# Who Decides on the Sell Side: The State of Sell-Side Curation in 2026

*By Jounce Media's global count for March 2025, nearly three in four auctions offered to buyers' platforms were curated, yet the public record does not show what a curated deal contains, what it costs or whether its quality claims hold. In the sources screened, no product documents a deal's sellers, selection rule and data provenance completely, none publishes its charge rates, and we found no independent matched evaluation of a named curated deal.*

Evgeny Popov · No Fluff Advisory · Research paper · November 2026 · Research cutoff 3 November 2026

## Abstract


*Who decides which inventory a buyer sees in programmatic advertising when the selection runs on the supply side, what a curated deal contains and costs, and whether its quality claims hold. Research current to the 3 November 2026 cutoff, with the United States as the main market.*

> **In short.**
> The public record shows how much of the auction curation occupies, not what a buyer is buying. By Jounce Media's count, 72.7% of the real-time-bidding auctions offered to buyers' platforms in March 2025 were curated, a global count of auctions, not spend [101, 64]. In the sources screened, no product among the 21 admitted to the matrix documents all three of a deal's sellers, selection rule and data provenance completely [110, 248, 25, 11, 258]. None publishes its charge rates or amounts [212, 231, 281, 115, 24]. We found no independent matched evaluation of a named curated deal [310, 101, 311]. For fourteen of fifteen exchange products the documentation opened does not settle whether a curated deal gets priority or a floor advantage by default [111, 250, 24, 266]. Five of eleven curation data suppliers passed into operating groups after 2023 [49, 41, 38, 36, 44, 34, 50]. Of the sell-side agent accounts read, one says what the agents selected, and every account comes from the parties that ran the campaign [313] [93, 120, 172, 168, 171]. Most of the paper's hypotheses are unresolved, because the evidence that would settle them is not public.

> **What a buyer can do with this.**
> Treat a curation fee as a routing charge until four things are in writing for the named deal: its seller and domain list, every charge with its amount, base, payer and collector, the deal's auction settings (priority, floor treatment and request sampling), and a measurement of its quality claim by someone without a stake, against comparable inventory before curation. [Chapter 12](#next) turns this into a checklist. This paper audits what vendors document in public by the cutoff; it does not test whether curated deals work.

### The question

Curation, as the market uses the word, moves the selection of inventory and the application of audience, context or quality data to the supply side, before the bid request reaches the buyer's platform, and exposes the result as a deal ID. The author's essays argue that curation is one move, that a curated deal carries no default priority, and that the margin moved to the sell side [1, 2]. This paper tests those positions on the public record without taking either side in advance. It asks how much trading runs through curated deals, what a deal contains and whether a buyer can check it, what priority it gets, what it costs, whether its quality claims are measured, who owns the curators, what the Google ad-tech cases change, and whether sell-side agents have reached production use.

### How it was done

Eight hypotheses and five coded questions were registered with their decision rules on 2 November 2026, alongside a product register and scoring rules with fixed anchors. The brief had one outside review by another AI model, whose factual findings were each re-opened at the source before any change. AI research agents then ran 10 workstreams, 1 vendor-universe sweep and 7 company profiles, and a separate AI verifier, told to refute, checked each file. 389 sources were opened and logged, and the ledger holds 454 claims, each traced to a source. Every scored cell was re-checked at its source by another verifier. The author waived the planned outside reviews of the evidence and the draft, and the evidence freeze fell on the day the research ran. This corrected edition of 4 November 2026 follows an outside AI review of the published paper, whose findings were checked at their sources. The corrections ledger holds 392 rows ([chapter 13](#method)).

### What it found

1. **The share series measure different things, and none measures curation.** The ANA's benchmark puts the median advertiser's private-marketplace share of tracked spend at 92.2% in Q4 2025, while the spend-weighted share fell after Q2 2025 and the advertiser panel changed [100, 309] [98, 100, 66]. Its private marketplace is any deal-ID trade [310, 287]. Jounce's 72.7% is a global share of auctions offered in one month [101, 64]. H1 is unresolved. [Chapter 2](#share).
2. **A buyer cannot establish what a named deal contains.** Six admitted products disclose part of a deal's seller composition, and the fullest is Equativ's publisher-name table for deals of 50 publishers or fewer [3] [110, 248, 25, 11, 258]. No product shows an identified deal's full seller list or checks delivery against it in public [110, 248, 25, 11, 258]. The supply-chain files define no curator role, and the one specification that names a curator, IAB Tech Lab's Deal Sync API 1.0, has no confirmed production use in the sources screened [143, 142, 141] [96] [96]. H6 is unresolved. [Chapter 3](#contents).
3. **Access is mostly undocumented.** Microsoft Curate packages every curated deal as an open auction, never overrides seller floors and samples deals that buyers stop bidding on [110, 111] [113]. Index Exchange documents net-of-fee competition and lets a package's creator choose a fixed price auction, which it defines as having priority over all other auction bid types [198, 109] [14]. Google lets curators bundle Preferred Deals, which it defines as fixed-price and pre-auction [24, 250]. The essay's no-default-advantage claim holds for Microsoft Curate and cannot be tested as a default for the other fourteen; two of them document a priority the curator can choose [111, 250, 24, 266]. [Chapter 4](#access).
4. **The curators have been absorbed.** Audigent went to Experian, Lotame and LiveRamp to Publicis, Semasio to Samba TV, and Adloox to Scope3 and then Peer39; Oracle retired its advertising products [49, 41, 38, 36, 44, 34, 50] [34]. Documented independent suppliers fell from four of eleven to two of ten, but control is undocumented for six of the eleven 2023 suppliers, so H5 is unresolved [49, 41, 38, 36, 44, 34, 50]. No exchange in the registers lost its independence [212] [231] [30]. [Chapter 5](#market).
5. **Fees exist in type, not in amount.** In the ANA's 2023 study, $710 of every $1,000 entering a demand-side platform reached the seller, and curation had no line of its own [26, 63]. Six exchange products document fee types and bases; none publishes a rate, and every "no fee" claim covers one party's charge [212, 231, 281, 115, 24] [188, 187, 200] [117]. No matched comparison of curated and open-auction cost is public, so H2a is unresolved, and no buyer-facing fee screen was observed, so H2b is unresolved too [115, 216, 290, 64] [212, 231, 281, 115, 24] [116, 12, 209, 128]. [Chapter 6](#fees).
6. **Quality claims come from the sellers.** Every quantified quality claim found for a curated product comes from a vendor, its agency or a partner, with no independent measurer and published method [119, 120, 121, 125]. The ANA's 2023 study found that made-for-advertising (MFA) sites took 1.3% of impressions in private-marketplace deals spanning 11 to 500 domains, 19.8% in deals spanning more than 500 and 26.8% in the open marketplace [26]. H3 is unresolved. [Chapter 7](#quality).
7. **Buyer-built marketplaces document the least.** Stagwell Curate launched on 16 July 2026 with its supply partners and agent connection described in the future tense, and no buyer-built marketplace discloses a fee [5] [72] [5, 72]. [Chapter 8](#marketplaces).
8. **The court found the advantages inside Google, and nothing is in force.** A Virginia court found First Look, Last Look and related practices anticompetitive, decided behavioural remedies on 2 September 2026 and had entered no judgment by 29 October 2026; none of the three court documents read mentions curated deals [78, 156] [156, 80] [80] [79, 156, 78]. [Chapter 9](#competition).
9. **One account says an agent selected inventory.** PubMatic's release of 9 June 2026 says agents on its platform "autonomously selected CTV inventory" for Havas's Movistar campaign in Spain [313]. Accounts of PubMatic's Geloso campaign and Magnite's LG buy describe agent-led activation, not what the agents selected [312] [172]. Every account comes from the transacting parties and none is independently measured, so H7 is supported with qualification [93, 120, 172, 168, 171]. [Chapter 10](#agents).
10. **On the Deal Quotient, the public record is thin.** 4 exchange curation platforms sit in the Strong band and 3 in Mid, none in Front rank; curated deal products and managed client marketplaces have too few well-documented rows for bands, and every band moves when missing cells are filled. Three in four counted cells rest on vendor marketing, and on stronger evidence alone no cohort would carry bands [314]. [Chapter 11](#dq).

### What it means

On this paper's reading, curation is sold as a quality purchase while it is documented as a routing arrangement. The selection rule, the sellers, the fees and the measurement that would make it a quality purchase are, in the sources screened, either not public or held by the parties that sell it. A buyer who pays a curation fee should ask for the named deal's contents, charges, auction settings and an independent measurement in writing ([chapter 12](#next)).

> **Interests.**
> The author has worked at Samba (TV measurement and advertising data) since May 2025. Samba owns Semasio, whose curated-deal product is in the register; it fell below the admission rule. He was EVP at Verve Group from September 2022 to August 2024; Verve's Brand+ Marketplace was scored and also fell below the admission rule. He worked earlier at Lotame, whose admitted row carries the interest mark, and at Hearts & Science. He co-leads the Signals & Measurement working group of AgenticAdvertising.org, which maintains AdCP, maintains an open-source AdCP signals adaptor, and No Fluff Advisory operates two agents in the AdCP registry; none earns any evidence weight. No vendor funded, reviewed or saw the paper. Every vendor is scored under the same published rules.



## 1. What curation is, and how this paper counts it


*What does the market mean by curation, and what do the standards record about who decides? On the supply side, the word names a selection or enrichment rule exposed as a deal ID, sold by exchanges, data companies, agencies and single publishers. Of the standards read, only IAB Tech Lab's new Deal Sync API has a field for the curator. Only exchange documentation says who collects the fee or can change a live deal.*

> **In short.**
> This paper codes four fields for every curated deal: mechanism, publisher multiplicity, transaction type and buyer eligibility. It sorts the purchased service into three cohorts by a frozen decision tree. The vendor sweep moved YMax to curated deal products and Seedtag NeuroX Curation to exchange platforms [11] [128]. Roku Curate packages Roku's own inventory with partner data, so it is coded single-publisher and listed [8]. OpenRTB 2.6 has no curator or curation-fee attribute, and ads.txt, sellers.json and SupplyChain 1.0 define no curator role [139] [143, 142, 141]. IAB Tech Lab's Deal Sync API 1.0, final on 6 February 2026, lets a deal name its curator and declare a fee type without an amount [261, 249, 96] [96] [96]. The searches found no curator or supply-side platform (SSP) confirming production support by 3 November 2026 [96]. On who decides (Q1), the rule owner is stated for all but four products in the three cohorts [59, 24, 248] [244] [5, 71]. On the pages opened, only five exchange platforms name a fee recipient and four a party able to change a live deal [115, 111] [109] [115, 24, 25, 109]. None of them documents notice to the buyer [115, 24, 25, 109].

### One word, many purchased services

"Curation" labels several different purchases. IAB Tech Lab's Deal Sync API defines it as "the selection and organization of inventory using technology and data" [96]. Microsoft Curate counts retailers, data companies, trading desks and other media companies among its curators, and lets buyers curate [56]. Audigent sells SmartPMP as publisher inventory and Audigent's first-party data under one deal ID [193, 116]. Stagwell runs Stagwell Curate for its own network's clients [5], and Roku sells Roku Curate on Roku's inventory [8]. Buying platforms use the word too. Google's Display & Video 360 Marketplace offers "curated bundles" [273]. Yahoo calls Backstage "curated MFA-free inventory" (MFA: made-for-advertising) [274, 275]. Walmart Connect calls Connect Select "a curated marketplace" inside Walmart's demand-side platform (DSP), the software buyers bid through [271, 272]. Viant's Direct Access page promises "Independence from SSP-curated deal ID limitations" [276].

This paper therefore uses a narrower definition. A curated deal is a deal ID under which a documented selection or enrichment rule is applied to an inventory package before the bid request reaches the buying platform. A single-publisher deal without a documented rule is a private marketplace (PMP) deal or a programmatic guaranteed transaction, and the open auction has no deal ID. A package built inside a buying platform is the comparator, described and never scored. How much trading meets the definition is the question of [chapter 2](#share).

### Four fields, coded apart

**Mechanism** takes five values: in-auction selection, in-flight enrichment, a seller or domain list, request sampling or deal-health optimisation, and DSP-side packaging as the comparator. One product can use several. Microsoft Curate's curators build deals from inventory categories, named sellers, allowlists and blocklists, with viewability or completion thresholds as hard limits [110, 111]. Microsoft adds its own cross-device data by default to any curated deal that targets a segment, an in-flight enrichment [110]. Its Curation Optimization samples deals a buyer has stopped bidding on [113]. Comscore's rule is a list: its Top 100, 500 or 2,000 site rankings [244]. Whether a rule selects inventory in or filters it out is recorded too. Integral Ad Science (IAS) announced IAS Curation in 2024 as applying avoidance, suitability and MFA filters [58, 262]. Adelaide sells high, medium and low AU (attention score) tiers as separate deal IDs [60].

**Publisher multiplicity** counts publisher entities, separately from exchange selling accounts and corporate owners. The Deal Sync API's pubcount field lets an inventory aggregator count as one publisher [96], so a declared value can differ from this paper's code. Multi-publisher curation is the primary subset; single-publisher curation is coded and reported with its exclusion.

**Transaction type** is recorded as the platform labels it. Microsoft sets every curated deal's auction type to Open Auction [110, 111]. Index Exchange defines a Marketplace deal ID as a private auction [109]. Google Ad Manager's curated packages bundle Private Auctions, which Ad Manager defines as auction-based, and Preferred Deals, which it defines as fixed-price and pre-auction [24, 250]. [Chapter 4](#access) codes what each label means for access.

**Buyer eligibility** is any seat, allow-listed seats or one buyer. OpenRTB's Deal object carries buyer-seat and advertiser-domain allow-lists [139]. A Microsoft curated deal serves one buyer, fixed once saved, while the same inventory stays biddable in the open exchange [110, 111, 56]. Commerce Grid allows one DSP seat per demand deal [266].

### Three cohorts by purchased service

The paper classifies the purchased service, not the company. A frozen decision tree asks in order:

1. Does a buyer purchase a tool to build and trade deals hosted on one exchange's inventory? Cohort A, exchange curation platforms.
2. Else, does a buyer purchase standardised deal IDs built by the operator? Cohort B, curated deal products.
3. Else, is it a curated supply pool run for a buyer's own clients? Cohort C, managed client marketplaces.
4. Else it goes to the comparator table, the data and standards table, or the listed products.

A separately purchased exchange service belongs in A whatever else the operator sells, and one-exchange operation does not exclude a B product. Features, aliases and activation variants never create rows. The provisional register was frozen on 2 November 2026, before the vendor sweep (workstream U). The eligible register followed on 3 November 2026, with every change recorded as a dated amendment. Of the 52 registered products, 22 are listed and 2 historical; the rest sit in the three cohorts.

Five cases test the tree.

- **YMax moved from A to B.** Yieldmo describes operator-built YMax packages that buyers activate through their own DSP as standard deal IDs, plus a managed tier [11]. The pages read describe no tool for third parties to build deals on its exchange, so step 2 applies [11].
- **Seedtag NeuroX Curation moved from B to A.** It is a self-serve interface giving activation teams pre-bid control of Seedtag's Neuro-Contextual Exchange to build web and CTV (connected TV) deals [128]. Step 1 applies.
- **Adloox is listed.** A buyer selects Adloox attention segments, at $0.15 per thousand impressions (CPM), while building its own auction package in PubMatic's Media Console [12]. The buyer builds the deal, so Adloox is a data supplier [12].
- **PubMatic Activate is listed** in the comparator table as a buying platform. PubMatic's buyer page, in workstream U's read, says curated deals "can be activated through PubMatic's Activate solution or your DSP of choice" [223, 20]. The verifier was refused access, so the read is unverified [223, 20]; Activate's own pages refused access too [4].
- **Roku Curate is single-publisher.** Partners such as Best Buy Ads, Criteo and Kroger supply data only, and on Roku's account "Roku packages the data and media supply" [8]. It is coded, listed and excluded from the multi-publisher subset.

Samba PMPs are also listed: Samba's site mentions PMPs in one homepage sentence and documents no deal product on the pages opened [268]. Permutive's and Onetag's curation products surfaced late and are listed for a later edition [267] [259].

> **Note on interests.**
> The author has worked at Samba, which owns Semasio, since May 2025. He was executive vice president at Verve Group from September 2022 to August 2024 and earlier worked at Lotame. Samba PMPs, Semasio's curated deals, Verve's Brand+ Marketplace and Lotame's Curated Marketplaces carry an interest mark under the same rules as every other row. None of these roles earns any evidence weight. [Chapter 13](#method) lists every interest.

### What the plumbing records

The OpenRTB bid request carries the deal, its floor and its allow-lists, and no standard field names a curator [139] [139].

| Instrument | Names a curator? | Fees | What it records | Status at the cutoff |
|---|---|---|---|---|
| OpenRTB 2.6 Pmp and Deal objects | No curator, packager or curation-fee attribute [139] | None | Deal ID, floor, auction-type override, seat and domain allow-lists, guaranteed flag [139] | 2.6-202606, 11 June 2026 [264, 139] |
| ads.txt 1.1, sellers.json 1.0 | No; a curator appears only as a paid seller account [143, 142, 141] | None | Authorised sellers; seller accounts typed PUBLISHER, INTERMEDIARY or BOTH [142, 263] | 2022 and July 2019 [141, 251] [142, 263] |
| SupplyChain 1.0 | No; every node must be in the payment flow [143] | None | The chain of paid intermediaries | 2019; non-payment participants deferred [142, 263] [143] |
| Deal Sync API 1.0 | Yes: curator domain and curator deal ID [96] | Type only: undisclosed, percentage, flat, CPM or none [96] | Publisher count, paid non-publisher data, in-flight inventory changes [96] | Final 6 February 2026; no production support confirmed [261, 249, 96] [96] |
| Media Rating Council (MRC) Auction Transparency Standards | No data format; covers any "Ad Auctioneer" [151] | Deal discounts and tech fees, gross or net, made known [151] | Floors, priority rules and superseding deals disclosed [151] [151] | January 2026; voluntary [151] |

OpenRTB 2.6 restricts bidding to a request's listed deals only when private_auction is set to 1 [139]. Its implementation notes (section 7.3) say an exchange "may be configured by a publisher to prioritize all Deal ID bids above open auction bids" [95]. They call such set-ups common practices and never use the word default [95]. Prebid recommends that deal line items in the ad server outrank open-market bids, and none of the 1,472 page paths in docs.prebid.org's sitemap contains "curat" [145, 265]. Both are configuration guidance; [chapter 4](#access) codes deployed defaults. Because SupplyChain 1.0 lists only entities in the direct flow of payment, a data or selection provider outside that flow cannot appear in it [143]. That step is an inference from the specification text [143, 142, 141].

The Deal Sync API separates three roles [96]. The origin is the system where the deal is first entered, typically an SSP [96]. The curator packages inventory, technology or data and is "most often, although not always" the party that sold the deal [96] [96]. The seller is the entity that contracted with the buyer [96]. Its curfeetype attribute carries no amount, and the guidance text calls it curationfee [96]. Version 1.0 supports no revisions, though attributes may change in flight [96], and it warns buyers that deal criteria are not guaranteed to appear in bid requests [96]. No cohort B page opened declares a fee type [96]. A specification field shows what a deal could declare; whether deployed deals declare it is not established in the sources screened [96].

IAB Tech Lab announced a Curation Workstream on 16 December 2024 and renamed Seller Defined Audiences as Curated Audiences [144, 252, 253]. By 3 November 2026 its site search returned no new curation specification, standard or guideline [144, 252, 253]. Its Curated Audiences page expected a release from the Curation Sub Working Group in Q4 2025, starting with deals standards [144, 252, 253]. The Deal Sync API, out for public comment on 4 December 2025, is the only output found [261, 249, 96] [144, 252, 253]. That absence is bounded to the Tech Lab's site search and the standards pages opened.

The MRC published its auction transparency standards in January 2026 [151]. They apply to any company that evaluates bids and sets prices "even if it is not the final auction in a multi-stage auction process" [151]. Reading that scope to cover exchanges running curated deals is an inference [151]. They say reserve prices should not be set per buyer and, in OpenRTB, should be managed through a Deal ID structure [151]. They require disclosure of "demand source priority" and of deals that "supersede an auction" [151], and advance notice of material auction changes [151]. Whether any exchange in the register has been audited against them is not established in the sources this chapter screened [151].

### Coding Q1

Curators own the selection rules and exchanges run them [59, 24, 248] [196]. On the pages opened, who collects the fee and who can change a live deal is written down only in five exchange platforms' own documentation [115, 111] [115, 24, 25, 109].

Q1 is a coding question. Its registered rule codes each field from vendor documentation or an official record, codes a marketing adjective as unknown, and leaves a field unknown where no opened source states it. "Not stated" below means not stated in the sources opened by 3 November 2026.

| Product (cohort) | Rule owner | Execution host | Data supplier | Fee recipient | Can change a live deal |
|---|---|---|---|---|---|
| Microsoft Curate (A) | Curator: retailer, data company, trading desk or buyer [56] | Microsoft's exchange, in the seller's final auction [110, 111] | Curator's segments; Microsoft cross-device by default [110] | Curator margin (none by default); Microsoft tech fee charged to the curator [115, 111] | Curator: margin, any time; Microsoft: sampling [115, 113, 110] [113] |
| Google Ad Manager Curation, Beta (A) | Curation partner [59, 24, 248] | Ad Manager [59, 24, 248] | Partner's own segments; Google's contextual signals and identifiers [59, 24, 248] | Partner, paid by Google out of the bid [24, 59, 248] | Partner: fee amount only [24, 25] |
| Index Marketplaces (A) | Marketplace partner: media owner, buyer or solution provider [258] | Index Exchange [109] | Data vendors, seventeen named in October 2025 [188] | Partner fee; Index's negotiated share [109] | Partner: fee disclosure [109] |
| Commerce Grid (A) | Curator [266] | Criteo's SSP [266] | Criteo's Commerce Audiences [266] | Curator margin; Criteo tech fee charged to the curator [266] | Curator: fee after sync [115, 24, 25, 109] |
| Maestro by Equativ (A) | Curator [114, 15] | Equativ [114, 15] | Not stated | Curator margin, basis not stated [114, 15] | Not stated |
| ClearLine (announced), OpenXSelect, Nexxen, TL Direct, FreeWheel Curation Hub, Seedtag NeuroX (A) | The buyer, curator, data partner or publisher using the tool [107, 108] [17, 10] [254] [130, 117] [16] [128] | Operator's exchange; FreeWheel names no SSP [16, 260] | Stated for all but FreeWheel: partners, the user's own data or the operator's signals [107, 108] [17, 10] [254] [130, 117] [128] | Not stated; TripleLift asserts no curation fee [130, 117] | Not stated; TripleLift says TL Spark adjusts floors [130, 117] |
| Buyer Hub, Sovrn Curated Deals (A) | Traders using an AI agent; not stated (Sovrn) [174] [277] | Not stated; Sovrn [174] [129, 256, 187, 257] | Not stated; Sovrn's Commerce Audiences [277] | Not stated | Not stated |
| PubMatic Connect, Brand+ Marketplace (A) | Not stated, on unverified reads [223, 20] [21] | Operator's exchange (unverified) [223, 20] [21] | Connect's data signals; Verve's software development kit (SDK) signals (unverified) [223, 20] [21] | Not stated | Not stated |
| Comscore-Certified Deal IDs (B) | Comscore: site rankings [244] | Magnite, then PubMatic [244] [227] | Comscore, Proximic, Adelaide [227] | Not stated | Not stated |
| SmartPMP, Experian, Adelaide, IAS, Lotame, Multilocal, Dstillery, YMax (B) | Operator; the buyer in Experian's Deal Engine [255] [11] | Mostly exchange records: Ad Manager, Index, Onetag, Sovrn, five SSPs (Adelaide), Microsoft Curate (Dstillery, 2022), Yieldmo (YMax); none for Experian [196] [187] [13, 196, 207] [129, 256, 187, 257] [60] [61, 70] [255] | Operator's own data, some with partners; not stated for YMax [193, 116] [193, 244, 58, 128, 13] | Not on any operator page [116, 12, 209, 128] [11] | Not stated |
| Semasio curated deals (B) | Unresolved: service page 404 [269, 270] | Not stated | Not stated | Not stated | Not stated |
| Stagwell Curate (C) | Stagwell [5, 71] | FreeWheel Curation Hub and Buyer Cloud, which "will power" it [72] | Stagwell signals; Sincera models [5, 71] | Not disclosed [5, 72] | Not stated [5] |
| WPP Media Premium Marketplace, dentsu Curate (C) | WPP Media; dentsu [7] [6] | Magnite and PubMatic (2022; North America, Europe, Middle East and Africa); Xandr (2020, Asia-Pacific) [7] [6] | Not stated; unnamed partners (dentsu) [6] | Payment models only (WPP) [7] | Not stated [7, 6] |
| Roku Curate (listed) | Roku [8] | Roku, or the advertiser's DSP [8] | Retail and other partners [8] | Not named; "lower fees" asserted [8] | Not stated [8] |
| Adloox segments (listed) | The buyer [12] | PubMatic, via Qwarry [12] | Adloox [12] | $0.15 CPM data fee; collector not stated [12] | Not stated |

Across the three cohorts, the rule owner is stated for all but four products: PubMatic Connect, Brand+ Marketplace, Sovrn Curated Deals and Semasio [223, 20] [21] [277] [269, 270]. The execution host is named for all but three, Buyer Hub, Experian and Semasio [174] [255] [269, 270]. Two of the named hosts, PubMatic's and Verve's, rest on unverified reads [223, 20] [21]. The data supplier is stated for all but six: Maestro, FreeWheel Curation Hub, Buyer Hub, YMax, Semasio and WPP Media's marketplace [114, 15] [16, 260] [174] [11] [269, 270]. A fee recipient is named for five products: Microsoft Curate, Google Ad Manager, Index Marketplaces, Commerce Grid and Maestro [115, 111] [24, 59, 248] [109] [266] [114, 15]. The same group without Maestro documents a party able to change a live deal [115, 24, 25, 109]. Each documents it in help or knowledge-base pages [115, 111] [24, 59, 248] [109] [114, 15] [266]. No cohort B or C operator names a fee recipient on the pages opened [116, 12, 209, 128] [5, 72] [7].

Where an exchange documents its platform in detail, it keeps decisions of its own. Microsoft samples curated deals without curator action, adds its cross-device data by default and always applies the AppNexus Blocklist [113] [110] [110, 111]. Commerce Grid deprioritises low-value curated deals in auction processing, coded here as request sampling [266]. Index opts media owners in to all Standard Marketplaces by default [109]. On Microsoft (by default), Google, Index and Commerce Grid the curator's fee comes out of the bid before it competes [110, 111] [24, 59, 248] [109] [266]. In Index's illustrative case, a $10 bid pays the media owner $8.08 after a 5% partner fee and a 15% Index share [109]. [Chapter 6](#fees) takes up the charges.

The buyer sees little of who decided. Google's subscribing agency sees the deal's name, pricing and a "created by" curator label, and no page opened says it sees the publishers [59, 24, 248]. Microsoft shows the buyer the deal name, while sellers can block named curators [110, 111]. Cohort B routes come mostly from exchange records, and those move. Google's launch post of 4 November 2024 named Lotame and Scope3 among eight partners [196]. Its help page, read on 3 November 2026, lists twelve partner entries and names neither [59].

Every documented power to change a deal after purchase sits with the curator, the exchange or the publisher [115, 24, 25, 109] [113]. Google's publishers can change or archive Marketplace Packages, cutting off subscribed buyers with no notification documented [25]. No exchange page opened documents notice to the buyer, an audit right, data export or a remedy [115, 24, 25, 109]. No source opened for the agency marketplaces or Roku states who can change a deal [5, 7, 6, 8]. The contracting seller is rarely stated: Microsoft's curators bring "specially negotiated rates with sellers" [56, 110], and Roku sells its own package [8]. No source screened says whether a curator acts as the buyer's agent or as a principal, so the registered rule leaves that field unknown.

In plain words, Q1 codes as follows. The selection rule belongs to whoever builds the deal. In cohort A that is the exchange's customer, who can be the buyer; in cohort B, the operator, usually a data or measurement company; in cohort C, the agency [56] [244] [5, 71]. In the three cohorts, every named host is an exchange or an exchange's curation tool, with FreeWheel's buying platform added for Stagwell [72] [16]. The best-documented exchanges keep sampling, default enrichment and blocklists for themselves [113] [110] [110, 111] [266]. Fee recipients and change rights are documented only by exchanges, for their own platforms [115, 24, 25, 109] [116, 12, 209, 128]. Unless the buyer curates its own deal, no page opened documents a change right for the buyer [115, 24, 25, 109] [116, 12, 209, 128]. What a buyer can see inside the deal is the subject of [chapter 3](#contents).



## 2. How much trading runs through curated deals


*Has the share of US open-web programmatic trading that runs through multi-publisher curated deals risen since 2023? No source screened by the cutoff measures that share, so H1 is unresolved [310, 101, 311]. The advertisers' benchmark counts deal-ID spend by a changing panel, Jounce Media counts auctions offered worldwide in one month, and the filing sections read show no deal share [98, 100, 66] [101, 64] [212, 231, 284, 30, 285, 282].*

> **In short.**
> The Association of National Advertisers (ANA) benchmark records private-marketplace (PMP) trading as a share of its contributing advertisers' tracked spend. The median across advertisers was 54.2% for the 2023 study and 92.2% in Q4 2025; the spend-weighted aggregate was 40.9% and 79.2% [26, 63] [100, 309]. From Q2 to Q4 2025 the median rose 4.4 points while the aggregate fell 3.4 points, and the contributing panel grew from 21 to 35 advertisers [98, 100, 66]. The ANA's PMP is any deal-ID trade: its glossary names two marketplace types and fixes neither publisher multiplicity nor the treatment of programmatic guaranteed [310, 287]. Jounce Media put curation at 72.7% of available real-time-bidding (RTB) auctions in March 2025, a global count of auctions offered to buyers' platforms [101, 64]. Jounce itself says buyer-controlled demand takes a far larger share of spend than its 18.6% of auctions [101, 64]. Its free 2023 edition has no comparable chart in its extractable text, so the auction series has one public point [104, 26, 280, 281, 101]. The FY2025 filings of Magnite, PubMatic and Nexxen disclose no deal, PMP or curated share in the sections read [212, 231, 284, 30, 285, 282]. H1 is unresolved, and the ANA's rise is a descriptive finding about deal-ID spend by its panel [310, 101, 311].

### Four sources, four units

Each source screened on deal-based trading counts something different, and none counts the share H1 asks about [310, 101, 311].

| Source | What it counts | Whose activity | Geography | Period | Deal buckets | Evidence |
|---|---|---|---|---|---|---|
| ANA Programmatic Transparency Benchmark | Share of tracked spend, as a median across advertisers and a spend-weighted aggregate | 21 to 35 contributing ANA member advertisers per period | US | September 2022 to December 2025, with gaps | Open marketplace, or private marketplace (any deal ID) | [26, 63] [100, 309] [310, 287] |
| Jounce Media, State of the Open Internet 2025 | Share of available RTB auctions presented to demand-side platforms (DSPs) | 1.5 billion bid requests from 44 exchanges | Global | March 2025 | Open auction; publisher-controlled PMP; curation (exchange- and curator-controlled) | [101, 64] |
| ISBA (the UK advertisers' trade body) and PwC, second supply chain study | Share of matchable impressions | 11 advertisers and 10 publishers, premium participants | UK | 1 September to 31 October 2022 | PMP deals | [106, 283] |
| Supply-side platform (SSP) annual filings | Each company's own activity | Magnite, PubMatic, Nexxen | Global registrants | FY2025 | None disclosed in the sections read | [212, 231, 284, 30, 285, 282] |

### The ANA series: deal-ID spend on a changing panel

The ANA's PMP share rose on both of its measures between the 2023 study and late 2025, while the panel, the period length and the channel mix changed under it. The 2023 study covered 21 marketers, $123 million of spend and 35.5 billion impressions from September 2022 to January 2023, reconciled by TAG TrustNet from impression log-level data [26]. It reported 41% of matched spend in PMP deals, spend-weighted and without connected TV (CTV) [26, 63]. The Q1 2025 report restates the same data as a 54.2% median and a 40.9% aggregate [26, 63]. The median takes one value per marketer [287, 102].

| Period as the ANA labels it | Months covered | Contributing of participating advertisers | Tracked spend | Median PMP share | Aggregate PMP share | Evidence |
|---|---|---|---|---|---|---|
| 2023 study | September 2022 to January 2023 | 21 | $123 million | 54.2% | 40.9% | [26] [26, 63] |
| 2024 update (later labelled Q4 2024) | April to September 2024 | 21 of 37 | $235 million | 56.2% | 65.9% | [102, 286, 63] [63] |
| Q1 2025 | November 2024 to March 2025 | 23 of 39 | $242 million | 64.5% | 67.2% | [63] |
| Q2 2025 | April to June 2025 | 21 of 39 | $146 million | 87.8% | 82.6% | [66, 98, 99] |
| Q3 2025 | July to September 2025 | 21 of 39 | $142 million | 88.8% | 81.6% | [98, 100] |
| Q4 2025 | October to December 2025 | 35 of 54 | $142 million | 92.2% | 79.2% | [100, 309] |
| Q1 2026 | January to March 2026 | 66 of 86 | $160 million | Not restated | Not restated | [103, 99] |

CTV entered the series with the 2024 update, which gave a PMP share of 59% without CTV and 66% with it, labelled "averages across marketers" [102, 286, 63]. Among the ANA's contributors, CTV trades almost wholly through deal IDs: 97% of CTV spend in that period and 100% in Q4 2025 [102, 286, 63] [100, 309]. Its share of tracked spend rose from 30.4% to 44.2% between Q1 and Q2 2025 [63] [66, 98, 99]. The period length also changed, from six months to five and then to quarters [102, 286, 63] [63]. The median's jump from 64.5% to 87.8% coincides with that shift, with contributors falling from 23 to 21 and with CTV rising in the mix [66, 98, 99]. The ANA publishes no matched-panel change in the editions read [66, 98, 99]. CTV does not explain the whole rise. Without it, the spend-weighted share was 41% in the 2023 study and 64.9% for web and mobile in Q4 2025, on different panels [26, 63] [100, 309].

The one near-matched step is Q2 to Q3 2025, when the contributor count held at 21; the ANA does not say the panel was identical [98, 100]. Over that step the median went from 87.8% to 88.8% and the aggregate fell one point to 81.6% [98, 100]. In Q4 2025 the panel grew to 35 contributors, the median reached 92.2% and the aggregate fell to 79.2% [100, 309]. A rising median with a falling aggregate fits new, smaller contributors buying almost entirely through deals while larger spenders used the open marketplace more. That reading is an inference, and no ANA edition read publishes anything that tests it [98, 100, 66]. Its own headlines, such as "Curation Becomes the Dominant Operating Model", read off the median [98, 100, 66].

The share series stops at Q4 2025, because the Q1 2026 report does not restate it [103, 99]. That report's cohort table gives PMP ad spend of 62.2% and 61.1% for the panel's lower- and higher-performing halves, on a basis it does not label [103, 99] [99]. Those figures extend neither line. The Q2 2026 report of 26 August 2026 is readable without login only as a title and a short summary [278, 279]. No Q3 2026 edition was listed on 3 November 2026 [278, 279].

![The median advertiser’s private-marketplace share rose to 92%; the spend-weighted share fell after mid-2025.](figures/fig-ana-two-series.svg)

*Figure fig-ana-two-series. **The median advertiser’s private-marketplace share rose to 92%; the spend-weighted share fell after mid-2025.** Share of contributing advertisers’ tracked programmatic spend that the ANA’s benchmark records as private-marketplace (any deal-ID) trading, by reporting period, as the median across advertisers and as the spend-weighted aggregate. From Q2 to Q4 2025 the median rose 4.4 points while the aggregate moved -3.4 points, and the contributing panel grew from 21 to 35 advertisers. Units: % of tracked spend. Scope: ANA benchmark contributors, predominantly US; periods as labelled (the 2023 study covers September 2022 to January 2023). Sources [63], [66], [98], [99], [100]. Caveat: The ANA’s private marketplace is any deal-ID trade: it includes single-publisher deals and does not isolate multi-publisher curation, so this is not a curation share (chapter 2). The panel changed between periods; no matched-panel series is published. Data: data/figure-data.json#ana_series*

### What the ANA's private marketplace contains

The ANA's private marketplace is the whole deal-ID bucket, and the benchmark does not publish how it sorts a trade into it. Its glossary defines a PMP as "A curated programmatic deal environment where buyers transact with selected publishers, often offering premium inventory" [310, 287]. It names only the open marketplace (OMP) and the PMP as marketplace types, and has no entry for curation, deal ID, preferred deal, programmatic guaranteed, auction package or private auction [310, 287]. The methodology page says the share of spend allocated to PMP deals "reflects investment in premium, curated inventory", and does not say how a transaction is classified [310, 287]. Single-publisher deals therefore sit in the same bucket as multi-publisher packages [310, 287]; [chapter 1](#defining) keeps the two apart.

The 2023 study once drew that line. It sorted PMP deals by domain count into publisher-specific deals (1 to 10 domains), seller-specific "auction packages" (11 to 500) and "SSP direct" deals (over 500) [26]. It said intermediary deal IDs "bundle together sites and apps from many publishers and rarely carry auction priority", their value proposition being "inventory curation" [26]. The benchmark's glossary carries no such split [310, 287].

The fee figures that travel with the series can be partly imputed. Where SSP log-level data or identifiers are missing, the methodology applies "estimated SSP fee rates" from "periodic SSP disclosures", and it publishes no match-rate or coverage figure [287, 102]. Its SSP cost line includes "curation services", so a curation fee collected through an SSP has no line of its own [287, 99, 26]. [Chapter 6](#fees) reads those figures.

### Jounce counts auctions offered

Jounce Media's split comes closest to the paper's multi-publisher subset, and it measures something else. In its State of the Open Internet 2025, a chart titled "Mix Of Available RTB Auctions" puts curation at 72.7%, the open auction at 18.6% and publisher-controlled private marketplaces at 8.7% [101, 64]. The caption reads "4 of every 5 auctions is an attempt by the sell-side to control demand" [101, 64]. The chart carries no date [101, 64]. AdExchanger, which calls Jounce a programmatic consultancy, dates the analysis to March 2025: 1.5 billion bid requests from 44 exchanges [101, 64]. Jounce's report covers web, mobile app and CTV worldwide [101, 64]. AdExchanger describes the 8.7% as single-seller deals with inventory from one publisher [101, 64].

The curation bucket joins two mechanisms. Exchanges can "package and sell supply from multiple publishers through a curated deal" [101]. Third-party curators sit on top of an exchange, "instructing the exchange to decorate each outbound bid request with one or more curated deal IDs" [101]. Jounce describes the exchange packages as multi-publisher, and AdExchanger calls the whole bucket multi-seller [101] [101, 64]. The chart reports exchange-controlled and curator-controlled demand as one bucket [101, 64].

The unit is a share of auctions offered, not of spend. A bid request carrying several deal IDs counts as several auctions [64]. AdExchanger reports that the average request carries an open auction and 3.5 deal IDs [64]. Large DSPs see more than 1,000 eligible deal IDs for a single impression and can bid on only one [64]. Jounce draws the consequence itself: "DSP-controlled demand is far more than 18.6% of programmatic ad spend" [101, 64]. Playwire, a publisher monetisation vendor, says in its summary of Jounce's work that curated deals "likely capture less than 73% of actual DSP spend" [9]. No source screened says how much less [310, 101, 311].

Jounce's series has one public point. The free 2023 edition's extractable text has no auction-mix chart with deal-ID shares [104, 26, 280, 281, 101]. The 2024 edition is for subscribers only, and no 2026 edition was listed on 3 November 2026 [104, 26, 280, 281, 101]. Jounce's Supply Path Benchmarking reports go to clients only [9, 64] [104, 26, 280, 281, 101]. A March 2023 auction mix on the 2025 definition is not established in the sources screened [104, 26, 280, 281, 101].

![By Jounce’s count, nearly three in four RTB auctions offered to DSPs in March 2025 were curated.](figures/fig-jounce-mix.svg)

*Figure fig-jounce-mix. **By Jounce’s count, nearly three in four RTB auctions offered to DSPs in March 2025 were curated.** Jounce Media’s split of available real-time-bidding auctions presented to DSPs, by who controls the demand: curation (exchange- and curator-controlled deal IDs), the open auction and publisher-controlled private marketplaces (single-seller deals). From its State of the Open Internet 2025; the March 2025 sample of 1.5 billion bid requests from 44 exchanges is as reported by AdExchanger. Units: % of available auctions. Scope: Jounce’s global bid-stream sample (web, app, CTV), March 2025. Sources [101]. Caveat: A share of auctions offered, not of spend or impressions bought; never compare it with the ANA’s spend share. No 2026 edition was public at the cutoff. Data: data/figure-data.json#jounce_mix*

### The UK studies and the filings

ISBA's studies and the parts read of the exchanges' annual reports add no share series [106, 283] [212, 231, 284, 30, 285, 282]. ISBA's 2020 study, with the Association of Online Publishers (AOP) and carried out by PwC, measured where money went [105, 106]. Across 15 advertisers and about £0.1 billion of UK spend, publishers received 51% of advertiser spend on average, with 15% an "unknown delta" [105, 106]. The second study ran from 1 September to 31 October 2022 and was published on 18 January 2023 [106, 283]. PMP deals made up about one-fifth of its 104 million matchable impressions and matched above 70%, "in part due to Deal IDs facilitating impression matching from DSP to SSP" [106, 283]. ISBA says the results "should not be taken as representative of the broader programmatic ecosystem" [106, 283]. PwC's summary recommends "well-curated PMPs" without defining the term [106].

The FY2025 filings of Magnite, PubMatic and Nexxen disclose no deal, PMP or curated share of their own activity in the sections read [212, 231, 284, 30, 285, 282]. Those sections are the business and risk-factor text, which end before the management discussion and the financial statements; each filing describes its fee as generally a percentage of spend [212, 231, 284, 30, 285, 282]. The research stream read Criteo's 10-K the same way, and the verifier did not recheck it [282]. Earnings calls and investor presentations were not screened [212, 231, 284, 30, 285, 282].

Magnite's 10-K describes ClearLine's "integrated curation capabilities" and a "Magnite Curator Marketplace" without a volume figure [212]. PubMatic's names "data partners and curators" as a customer group [231]. The nearest number sits in PubMatic's management discussion, outside those sections: supply-path optimisation (SPO) "represented approximately 55% of total activity" in 2025 [231]. The passages read do not define "total activity" [231], and SPO is not curation on the paper's definition ([chapter 1](#defining)).

### Why the two series cannot be joined

The ANA and Jounce figures share the word curation and little else a trend needs.

| Attribute | ANA benchmark | Jounce, State of the Open Internet 2025 | Evidence |
|---|---|---|---|
| Unit | Share of contributing advertisers' tracked spend | Share of available RTB auctions presented to DSPs | [26, 63] [101, 64] |
| Statistic | Median across advertisers, and a spend-weighted aggregate | Share of one pooled bid-stream sample | [287, 102] [101, 64] |
| Base | 21 to 35 contributing advertisers per period | 1.5 billion bid requests from 44 exchanges | [98, 100, 66] [101, 64] |
| Geography | US | Global | [26] [101, 64] |
| Time | 2023 study, then reporting periods from April 2024 to December 2025 | March 2025 only | [102, 286, 63] [100, 309] [104, 26, 280, 281, 101] |
| Deal bucket | Every deal-ID trade, single-publisher deals included | Exchange- and curator-controlled deals, with publisher-controlled PMPs counted apart | [310, 287] [101, 64] |

Setting Jounce's share beside either ANA figure compares auctions offered worldwide with dollars spent by a few dozen US advertisers [101, 64]. A spend share weights what was bought. An auction share counts what was offered, and each deal ID added to a request adds an auction without adding an impression [64]. Splicing the ANA's 2023 point onto Jounce's 2025 point would change unit, deal bucket and geography at once, and the slope would mean nothing. Inside the ANA series, the median and the aggregate stay apart for the same reason [98, 100, 66].

The register below gives each published figure its unit, base, period and exclusions, and says whether it may be combined.

**Market-size reconciliation**

| Estimate | Published | Metric | Value | Year | Geography | Includes | Excludes | Combinable with | Src |
|---|---|---|---|---|---|---|---|---|---|
| ANA 2023 study (Complete Report) | 2023-12 | PMP share of matched programmatic spend, spend-weighted | 41% | 2023 | US (21 ANA advertisers) | any deal-ID transaction (binary OMP/PMP); web display and video on sites and apps | CTV; agency fees | no: ANA series only | [26] |
| ANA 2023 study as restated in Q1 2025 report | 2025-05 | PMP share of tracked spend, median across advertisers | 54.2% | 2023 | US (21 contributors) | any deal-ID transaction | CTV | no: ANA median series only | [63] |
| ANA 2023 study as restated in Q1 2025 report | 2025-05 | PMP share of tracked spend, aggregate | 40.9% | 2023 | US (21 contributors) | any deal-ID transaction | CTV | no: ANA aggregate series only | [63] |
| ANA December 2024 update | 2024-12 | PMP share of spend, 'averages across marketers', excluding CTV | 59.0% | 2024 | US (21 contributors of 37) | any deal-ID transaction | CTV | no | [102] |
| ANA December 2024 update | 2024-12 | PMP share of spend, 'averages across marketers', including CTV | 66% | 2024 | US (21 contributors of 37) | any deal-ID transaction; CTV |  | no | [102] |
| ANA Q1 2025 report (restating Apr-Sep 2024) | 2025-05 | PMP share of tracked spend, median | 56.2% | 2024 | US (21 contributors) | any deal-ID transaction; all environments |  | no: ANA median series only | [63] |
| ANA Q1 2025 report (restating Apr-Sep 2024) | 2025-05 | PMP share of tracked spend, aggregate | 65.9% | 2024 | US (21 contributors) | any deal-ID transaction; all environments |  | no: ANA aggregate series only | [63] |
| ANA Q1 2025 report | 2025-05 | PMP share of tracked spend, median | 64.5% | 2025 | US (23 contributors of 39) | any deal-ID transaction; all environments |  | no: ANA median series only | [63] |
| ANA Q1 2025 report | 2025-05 | PMP share of tracked spend, aggregate | 67.2% | 2025 | US (23 contributors of 39) | any deal-ID transaction; all environments |  | no: ANA aggregate series only | [63] |
| ANA Q1 2025 report | 2025-05 | PMP share of spend excluding CTV | 53% | 2025 | US (23 contributors) | web and mobile app | CTV | no | [63] |
| ANA Q1 2025 report | 2025-05 | CTV spend executed via deal IDs in PMPs | 99.5% | 2025 | US (contributors with CTV) | CTV only |  | no | [63] |
| ANA Q2 2025 release / Q3 report / Q1 2026 table | 2025-08 | PMP share of tracked spend, median | 87.8% | 2025 | US (21 contributors of 39) | any deal-ID transaction; all environments |  | no: ANA median series only | [66] |
| ANA Q3 2025 report (Q2 column) | 2025-11 | PMP share of tracked spend, aggregate | 82.6% | 2025 | US (21 contributors of 39) | any deal-ID transaction; all environments |  | no: ANA aggregate series only | [98] |
| ANA Q3 2025 report | 2025-11 | PMP share of tracked spend, aggregate | 81.6% | 2025 | US (21 contributors of 39) | any deal-ID transaction; all environments |  | no: ANA aggregate series only | [98] |
| ANA Q4 2025 report (text) / Q3 report median row | 2026-02 | PMP share of tracked spend, median | 88.8% | 2025 | US (21 contributors of 39) | any deal-ID transaction; all environments |  | no: ANA median series only | [100] |
| ANA Q3 2025 report | 2025-11 | PMP share of CTV spend, aggregate | 99.2% | 2025 | US | CTV only |  | no | [98] |
| ANA Q3 2025 report | 2025-11 | PMP share of web plus mobile spend, aggregate | 65.2% | 2025 | US | web and mobile app | CTV | no | [98] |
| ANA Q4 2025 report | 2026-02 | PMP share of tracked spend, median | 92.2% | 2025 | US (35 contributors of 54) | any deal-ID transaction; all environments |  | no: ANA median series only | [100] |
| ANA Q4 2025 report | 2026-02 | PMP share of tracked spend, aggregate | 79.2% | 2025 | US (35 contributors of 54) | any deal-ID transaction; all environments |  | no: ANA aggregate series only | [100] |
| ANA Q4 2025 report | 2026-02 | PMP share of CTV spend, median and aggregate | 100% | 2025 | US | CTV only |  | no | [100] |
| ANA Q4 2025 report | 2026-02 | PMP share of web plus mobile spend, aggregate / median | 64.9% / 87.4% | 2025 | US | web and mobile app | CTV | no | [100] |
| ANA Q1 2026 report | 2026-06 | PMP ad spend share, cohort averages (lower / higher TrueAdSpend cohort) | 62.2% / 61.1% | 2026 | US (66 contributors of 86) | cohort means |  | no | [99] |
| ANA December 2024 update | 2024-12 | PMP CPM vs OMP CPM, 'average' | $7.46 vs $3.26 | 2024 | US | all environments |  | no | [102] |
| ANA December 2024 update | 2024-12 | CTV PMP CPM vs CTV OMP CPM | $15.00 vs $5.54 | 2024 | US | CTV |  | no | [102] |
| ANA Q1 2025 report | 2025-05 | CPM total / PMP / OMP, spend-weighted | $6.09 / $10.43 / $3.03 | 2025 | US | all environments |  | no | [63] |
| ANA Q1 2025 report | 2025-05 | CPM total / PMP / OMP, median | $5.62 / $9.37 / $3.28 | 2025 | US | all environments |  | no | [63] |
| ANA Q2 2025 release | 2025-08 | PMP CPM vs OMP CPM, median | $7.15 vs $4.41 | 2025 | US (21 contributors) | all environments |  | no | [66] |
| ANA Q3 2025 report (Q2 column) | 2025-11 | PMP CPM vs OMP CPM, spend-weighted | $7.43 vs $3.64 | 2025 | US (21 contributors) | all environments |  | no | [98] |
| ANA Q3 2025 report | 2025-11 | PMP CPM vs OMP CPM, spend-weighted | $9.88 vs $3.55 | 2025 | US (21 contributors) | all environments |  | no | [98] |
| ANA Q4 2025 report | 2026-02 | PMP CPM vs OMP CPM, 'overall' | $10.72 vs $4.25 | 2025 | US (35 contributors) | all environments |  | no | [100] |
| ANA Q1 2026 report | 2026-06 | CPM total / PMP / OMP (report text calls the total a median of $4.41) | $4.42 / $7.77 / $3.58 | 2026 | US (66 contributors) | all environments |  | no | [99] |
| ANA benchmark | 2024-12 to 2026-06 | Participating / contributing marketers | 21/21 (2023); 37/21 (Apr-Sep 2024); 39/23 (Q1 2025); 39/21 (Q2 2025); 39/21 (Q3 2025); 54/35 (Q4 2025); 86/66 (Q1 2026) | 2023-2026 | US | ANA members who requested LLD access / whose suppliers granted it |  | no | [99] |
| ANA benchmark | 2024-12 to 2026-06 | Tracked ad spend | $123M (2023); $235M (Apr-Sep 2024); $242M (Q1 2025); $146M (Q2 2025); $142M (Q3 2025); $142M (Q4 2025); $160M (Q1 2026) | 2023-2026 | US | contributing advertisers' reconciled spend |  | no | [99] |
| ANA benchmark | 2024-12 to 2026-06 | Tracked impressions | 35.5bn (2023); 38.5bn (Apr-Sep 2024); 41.9bn (Q1 2025); 18.6bn (Q2 2025); 17.5bn (Q3 2025); 16.7bn (Q4 2025); 20.9bn (Q1 2026) | 2023-2026 | US | contributing advertisers' reconciled impressions |  | no | [99] |
| ANA 2023 study | 2023-12 | TrueAdSpend (cents of each DSP dollar effectively reaching the consumer) | 36 | 2023 | US | transaction costs 29%; media productivity loss 35% | agency fees; ad serving; brand safety; CTV | no | [26] |
| ANA December 2024 update | 2024-12 | TrueAdSpend | 43.9% | 2024 | US | transaction costs 29.8%; media productivity loss 26.3% | CTV | no | [102] |
| ANA Q1 2025 report | 2025-05 | TrueAdSpend | 41.0% | 2025 | US | CTV now included |  | no | [63] |
| ANA Q1 2026 release | 2026-05 | TrueAdSpend index (market level) | 43.3% | 2026 | US | all environments |  | no | [103] |
| ANA 2023 study | 2023-12 | MFA share of impressions / spend (web-domain subset) | 21% / 15% | 2023 | US | DeepSee.io MFA list, web domains | CTV; apps | no | [26] |
| ANA 2023 study | 2023-12 | MFA spend, annual extrapolation | $10.1bn (15% of $67.3bn) | 2023 | global | $88bn open web less $20.3bn CTV |  | no | [26] |
| ANA 2023 study | 2023-12 | Efficiency opportunity | $22bn of $88bn | 2023 | global | $10bn MFA plus $12bn other |  | no | [26] |
| ANA Q1 2025 report | 2025-05 | Efficiency opportunity, re-based | $21.6bn (20.8% of $104bn) | 2024 | global (Madison and Wall base) |  |  | no | [63] |
| ANA Q2 2025 release | 2025-08 | Efficiency opportunity, re-based | $26.8bn | 2025 | global ($123bn base per Q3 report) |  |  | no | [66] |
| ANA Q4 2025 report | 2026-02 | MFA rate by marketplace, OMP / PMP | 0.98% / 0.23% | 2025 | US | spend share |  | no | [100] |
| ANA Q4 2025 report | 2026-02 | Median MFA exposure; top-quartile range | 0.45%; 5.5% to 31.4% | 2025 | US |  |  | no | [100] |
| ANA Q1 2026 release | 2026-05 | MFA exposure | 1.1% (0.4% to 0.6% through 2025) | 2026 | US |  |  | no | [103] |
| ANA December 2024 update | 2024-12 | Buy-side transaction costs: total / DSP platform / data and additional | 16.8% / 8.6% / 8.2% | 2024 | US | DSP charges as share of ad spending | SSP fees | no | [102] |
| ANA December 2024 update | 2024-12 | PMP DSP platform cost minus OMP DSP platform cost | -1.1 points | 2024 | US |  |  | no | [102] |
| ANA December 2024 update | 2024-12 | CTV DSP fees | 8.78% | 2024 | US | 'little difference between marketplace type' |  | no | [102] |
| ANA Q4 2025 report | 2026-02 | SSP platform costs, median | 12.7% | 2025 | US | may include estimated SSP rates per methodology |  | no | [100] |
| Jounce SOI 2025 | 2025 | Curation share of available RTB auctions (exchange- plus curator-controlled) | 72.7% | 2025 | global RTB (web, app, CTV) | every curated deal ID on a request counts as an auction | spend | no: auctions, not spend | [101] |
| Jounce SOI 2025 | 2025 | Open auction share of available RTB auctions | 18.6% | 2025 | global RTB |  | spend ('far more than 18.6% of programmatic ad spend') | no | [101] |
| Jounce SOI 2025 | 2025 | Publisher-controlled private marketplace share of available RTB auctions | 8.7% | 2025 | global RTB | single-publisher PMP and PG |  | no | [101] |
| Jounce via AdExchanger | 2025-05 | Sample | 1.5bn bid requests from 44 exchanges | 2025 | global |  |  | no | [64] |
| Jounce via AdExchanger | 2025-05 | Deal IDs per average bid request (plus one open auction) | 3.5 | 2025 | global |  |  | no | [64] |
| Jounce April 2025 SPB report via Playwire / AdExchanger | 2025-04 | Curated deal IDs with no fee beyond exchange take | 50% | 2025 | global | share of curated deal IDs (Playwire also says 'of bid requests') | spend | no | [9] |
| Jounce April 2025 SPB report via Playwire / AdExchanger | 2025-04 | Curated deal IDs with margin-based fee; median fee within subset | 34%; 14% | 2025 | global |  |  | no | [9] |
| Jounce April 2025 SPB report via Playwire | 2025-04 | Curated deal IDs with fixed CPM fee; median fee within subset | 10%; $0.48 | 2025 | global |  |  | no | [9] |
| Jounce April 2025 SPB report via Playwire / AdExchanger | 2025-04 | Curated deal IDs with static price | 7% | 2025 | global |  |  | no | [9] |
| Jounce April 2025 SPB report via Playwire | 2025-04 | Median effective margin; median effective CPM of curated deals | 30%; $0.51 | 2025 | global |  |  | no | [9] |
| Jounce SOI 2025 citing its December 2024 SPO report | 2025 | Largest MFA publishers contracting by more than 95% within 18 months of the ANA report | 18 of 20 | 2024 | global RTB | bid-request volume | spend | no | [101] |
| Jounce SOI 2023 | 2023 | Open internet ad spend projection | $85bn | 2023 | global | web, mobile app, CTV |  | no | [104] |
| Jounce SOI 2023 | 2023 | Open programmatic ad spend (chart label) | $81.1bn | 2023 | global |  | reservations $3.7bn | no | [104] |
| ISBA/PwC 2020 | 2020-05 | Publisher share of advertiser spend; unknown delta | 51%; 15% | 2020 | UK | 15 advertisers, ~£0.1bn |  | no | [105] |
| ISBA/PwC 2020 (restated in 2023 summary) | 2023-01 | Match rate; matched impressions | 12%; 31m | 2020 | UK |  |  | no | [106] |
| ISBA/PwC 2022 | 2023-01 | Impressions analysed / matchable / matched; match rate | 1.3bn / 104m / 61m; 58% | 2022 | UK | 11 advertisers, 10 publishers |  | no | [106] |
| ISBA/PwC 2022 | 2023-01 | Unknown delta overall / for PMPs | 3% / under 1% | 2022 | UK |  |  | no | [106] |
| ISBA/PwC 2022 | 2023-01 | Publisher revenue share (2020 restated) | 65% (57%) | 2022 | UK | excludes agency fees, verification, ad serving |  | no | [106] |
| ISBA/PwC 2022 | 2023-01 | PMP share of matchable impressions; PMP match rate | about one-fifth; above 70% | 2022 | UK |  |  | no | [106] |
| Adalytics MFA report | 2024 | Disclosed ANA 2023 participants observed on MFA sample | 16 of 17 | 2024 | US | 22-site sample |  | no | [68] |

### The essay's three-quarters

The author's earlier paper on contextual advertising states that "roughly three-quarters of the bid stream is now curated deals, and the margin moved to the sell side" [2]. On Jounce's unit the first clause holds: 72.7% of available RTB auctions in March 2025 fell in Jounce's curation bucket [101, 64]. "Bid stream" is the nearest noun for that unit. As a share of spend, of impressions bought or of US trading, the clause has no support in the sources screened, and Jounce says as much for spend [101, 64]. Jounce's curated deals also include packages built by the exchanges themselves [101], and "now" rests on one global month [104, 26, 280, 281, 101]. The margin clause is a fee claim, tested in [chapter 6](#fees). Here the sentence is a rival position with no evidence weight, and on the site it needs its unit, base and month.

This chapter weighs the author's own claim, so his interests apply. He has worked at Samba (TV measurement and ad data; owner of Semasio) since May 2025. He was EVP at Verve Group from September 2022 to August 2024, and earlier worked at Lotame and Hearts & Science. He co-leads the Signals & Measurement working group of AgenticAdvertising.org (AdCP, the Ad Context Protocol) and maintains an open-source AdCP signals adaptor. No Fluff Advisory operates two agents listed in the AdCP registry. None of these interests earns any evidence weight.

### What a series would need

A series that could settle H1 would combine four properties that no source screened combines [310, 101, 311].

- **A constant definition coded from the transaction.** A curated deal is a deal ID with a documented selection or enrichment rule ([chapter 1](#defining)), applied the same way every period. The ANA's label is binary, and its glossary is undated and unversioned [310, 287].
- **Publisher multiplicity per deal.** Publisher entities are counted for each deal, apart from domains, selling accounts and owners. The ANA's 2023 domain-count profiles show that log-level data can carry a proxy [26], though domains are not publishers.
- **A matched panel.** The same contributors are followed across periods, or each change is split into behaviour within the panel and change in its make-up. No ANA edition read publishes either [98, 100, 66].
- **A US base of spend or impressions bought, with channels split.** Auctions offered rise with every deal ID added to a request [64], and an all-channel share moves with the CTV mix [66, 98, 99].

It would also need a 2023 point on the same basis. A Jounce mix for March 2023 would still count auctions offered. IAB Tech Lab's Deal Sync API 1.0 has a Curation object meant to add "transparency into Curated deals that is not currently available in the bid stream" [96]. A specification field shows what could be recorded. It does not show that any exchange records it.

### Coding H1

H1, registered on 2 November 2026, states that the share of US open-web programmatic trading running through multi-publisher curated deals rose between 2023 and the cutoff. It is supported if a series rises that identifies multi-publisher curated transactions on a constant definition and an appropriately representative US population. It is contradicted if such a series is flat or falls. It is unresolved if no such series exists in the sources screened.

No series meets the test. The ANA series is US and measures spend, but its PMP bucket holds every deal ID and its panel changed without matched-panel figures [310, 287] [98, 100, 66]. Jounce's bucket approximates multi-publisher deals, but it counts auctions offered in a global sample and has one public point [101, 64] [104, 26, 280, 281, 101]. The ISBA and PwC figure is a UK share of matchable impressions over two months of 2022 [106, 283]. The filing sections read disclose no deal, PMP or curated share [212, 231, 284, 30, 285, 282]. In the sources screened by 3 November 2026, nothing identifies multi-publisher curated transactions on a constant definition [310, 101, 311].

H1 is coded unresolved, the registered default. The absence is a limit of the public record and says nothing about the direction of the share. The gated Q2 2026 ANA report and Jounce's client reports may hold more, and neither was readable [278, 279] [9, 64].

The proxies stand as descriptive findings. Deal-ID trading was a larger share of the ANA contributors' tracked spend in Q4 2025 than in the 2023 study on both measures. The median was 92.2% against 54.2%, and the aggregate 79.2% against 40.9% [26, 63] [100, 309]. The aggregate peaked at 82.6% in Q2 2025 and then fell while the median kept rising [66, 98, 99] [98, 100, 66]. In Jounce's March 2025 global sample, 72.7% of the auctions offered to DSPs fell in the curation bucket [101, 64].



## 3. What a curated deal contains, and whether a buyer can check it


*Can a buyer establish from documentation which sellers a curated deal carries, the rule that selected them and where the applied data came from? For none of the 21 admitted products can a buyer establish all three: no product documents the three elements completely, and 15 show no seller composition in the sources screened [110, 248, 25, 11, 258] [193, 244, 58, 128, 13] [193, 116, 190]. The supply-chain files cannot fill the gap, because they define no curator role [143, 142, 141].*

> **In short.**
> Six of the 21 admitted products disclose part of a deal's seller composition, 19 name a selection logic and 16 name a source type and provider for their data [110, 248, 25, 11, 258] [193, 244, 58, 128, 13]. None documents all three completely, and dentsu Curate documents none [6]. The fullest seller disclosures are partial: Equativ documents a publisher-name table for Maestro deals of 50 publishers or fewer, and PubMatic markets a console that shows buyers where inventory comes from [3] [222, 233]. No admitted product shows in public an identified deal with its full seller list, or a check that delivered impressions came from the sellers disclosed [110, 248, 25, 11, 258] [222, 233] [193, 116, 190]. Provenance stops at the source type, and Microsoft switches on its cross-device feature for segment-targeted deals by default without naming the graph's source [193, 244, 58, 128, 13] [110]. IAB Tech Lab's Deal Sync API 1.0 can declare a curator, a fee type and whether a deal is single- or multi-publisher, but no curator or SSP (supply-side platform) was found confirming production support [96] [96] [96]. Adalytics observed mis-declared inventory and ads served to declared bots in trading through what appear to be private marketplace (PMP) deal IDs, and tied none of them to an admitted product [67] [69] [67]. On the registered rule H6 is unresolved: the products split, with no majority documenting all three elements or none of them [110, 248, 25, 11, 258] [6].

### No admitted product documents all three elements in full

H6 asks whether a buyer can establish from documentation what a named curated deal contains. Three elements are coded for each of the 21 products in the matrix: seller composition, selection rule and data provenance. Each code is read off the product's matrix cell. A score of 4 or 5 codes as complete, 2 or 3 as partial, and 0, 1 or no disclosure found as none. Access unknown is kept for a page that could not be opened, and no element met that case. On composition, a general assertion of transparency scores 1 and so codes as none ([chapter 13](#method)).

| Element | Complete | Partial | None | Access unknown | Where the record stops |
|---|---|---|---|---|---|
| Seller composition | 0 | 6 | 15 | 0 | No buyer-facing seller or domain list for an identified deal [110, 248, 25, 11, 258] [193, 116, 190] [5, 71, 16] |
| Selection rule | 0 | 19 | 2 | 0 | Logic named; thresholds or update cadence missing [193, 194, 192, 190] [228, 224, 12] |
| Data provenance | 0 | 16 | 5 | 0 | Source type and provider named; no per-segment fields [193, 244, 58, 128, 13] [246, 230, 228, 222, 227] |

Three products code partial on all three elements: PubMatic Connect, Comscore-Certified Deal IDs and Yieldmo's YMax [222, 233] [244] [11]. One, dentsu Curate, codes none on all three [6]. It was launched on 15 September 2020 as an Asia-Pacific supply solution built on Xandr's technology, and no dentsu record of it from 2025 or 2026 was found [6]. Its release says the solution "is built upon principles of quality and transparency" and offers "added addressability through data partnerships", and it names no publisher, rule or data provider [6].

Microsoft Curate, Google Ad Manager and Audigent's SmartPMP code none on composition because their documentation shows the buyer no seller list or seller count for a deal [110, 111] [59, 24, 248] [193, 116, 190]. The DSP (demand-side platform) and exchange screens a logged-in buyer uses were not accessed, so the code describes documentation only, and the profiles record those screens as access unknown [110, 111] [193, 116, 190].

### Six products disclose part of a deal's composition

Six products disclose some of what a deal holds, each by a different route and none in full [110, 248, 25, 11, 258].

| Product | What is disclosed | Mode | Source |
|---|---|---|---|
| Maestro by Equativ | A publisher-name table for deals of 50 publishers or fewer; no domain list; larger deals not covered | Authenticated (help centre) | [3] |
| PubMatic Connect | Console marketing: an "Inventory Discovery Tool" for the past 30 days, "Log-Level Transaction Verification" and publisher lists | Authenticated (described in marketing) | [222, 233] |
| OpenXSelect | Curated Packages name example publishers per package, "and more"; three packages are Tubi-only | Public | [17, 10] [10] |
| YMax | "Every URL is visible before you commit"; packages "reviewable before you activate"; no seller list | Public (marketing) | [11] [11] |
| Comscore-Certified Deal IDs | Sellers by ranking band (Comscore's Top 100, 500 or 2,000 sites) and, on PubMatic, "many of the top CTV publishers" (connected TV) | Public (press releases) | [244] [227] |
| WPP Media Premium Marketplace | A 2023 page: integrations with "80% of top 20 connected TV publishers" in North America, six of them named | Public (dated 2023) | [18] |

Equativ's help centre page for creating Maestro deals, updated 15 October 2026, carries the table; it lists publishers, not domains, and no table is documented for larger deals [3].

PubMatic's console page says its fee transparency lets buyers and sellers "know exactly where the inventory is coming from", a marketing claim [222, 233]. Adloox's steps for the same console offer "Maximize Reach" beside "Upload or Select publishers", so a deal need not carry a publisher list at all [12].

OpenX's holiday package names "Wurl, Plex, Scripps, Frndly TV, Hallmark Media, Xumo, Dish, AMC, and more", with no domain or app list and no seller count [10].

Comscore launched the deal IDs on Magnite on 5 May 2025 and extended them to PubMatic on 10 June 2025; no product page for them was found at the cutoff, so current availability is unconfirmed [244] [227]. The latest WPP Media page found on Premium Marketplace is dated 9 February 2023, and no page opened says it is still sold [18].

The other 15 offer less. Microsoft documents only the deal name, plus an optional code for deal mapping, as exposed to the buyer [110, 111]. Google's subscribing agency sees the deal's name, pricing and a "created by" curator label [59, 24, 248]. Index asserts "Total transparency" and "Real-time dashboards and audit logs", and no page opened documents a buyer-facing list of the sellers in a package [198, 200, 187]. Lotame's capability page offers brands reach "with control and transparency" and names no SSP or publisher [13]. Stagwell's Matt Adams told Digiday that clients "can see our buy data", with no interface or example deal documented [5, 71, 16].

### Capability, a named deal and delivery

Three things are recorded separately:

- a documented capability: what a tool or page says a buyer can see;
- an observed disclosure: an identified deal ID with its seller or domain list;
- delivery verification: evidence that impressions delivered under the deal came from the sellers disclosed.

Only the first appears in the record, for Maestro, PubMatic and YMax [3] [222, 233] [11]. An observed disclosure was not found for any admitted product in the sources screened [110, 248, 25, 11, 258] [193, 116, 190]. OpenX's packages come nearest, but their lists are open-ended and no package is tied to an identified deal ID in the record [10]. Delivery verification was not found either [110, 248, 25, 11, 258]. PubMatic's "Log-Level Transaction Verification" and Index's "audit logs" are features asserted in marketing, with no public example of their output [222, 233] [258].

Composition can also move after purchase: on Index, a Marketplace package left with a blank publisher list targets all eligible publishers, and new publishers are added automatically [14] [14]. No page opened documents notice to the buyer when that happens [109, 14, 240, 206].

### Selection rules are named and their thresholds are mostly missing

Nineteen products name a selection logic, two name none, and no product documents both its thresholds and an update cadence [193, 194, 192, 190] [228, 224, 12] [244].

Microsoft lets curators select by its universal categories, applied from its review of inventory content, by seller categories, by seller and publisher include or exclude lists and by allowlists and blocklists, with the AppNexus Blocklist always applied [110]. Curator-set viewability or completion-rate thresholds act as hard limits on Microsoft's predicted rates, and the prediction models and category-review cadence are not described [110].

Index's package builder takes domain and app-bundle include and exclude entries totalling up to 10,000, with exclusions taking priority, up to four include segment rule sets and one exclude, and a minimum viewability probability [14]. Thresholds are set deal by deal and are not observable for any identified deal [14]. In Google's curation feature, still in Beta, partners build packages from targeting dimensions and up to ten of their own segments joined by AND logic [59, 24, 248].

Comscore names its thresholds: buyers target the Comscore Top 100, 500 or 2,000 site rankings, built from its MMX and Video Metrix data with Proximic contextual signals [244]. The ranking cadence is not stated [244]. The only update cadence found belongs to a partner: Adloox's "Top 20%" and "Top 40%" attention segments on PubMatic, "updated daily" [228, 224, 12]. Adelaide groups placements into high-, medium- and low-AU sets sold as separate deal IDs, AU being its attention-based quality metric, and states no cut-off for any tier [60]. IAS's launch release of 4 November 2024 says IAS Curation applies contextual avoidance, brand safety and suitability, made-for-advertising (MFA) filtration and viewability benchmarks "before it reaches a bidder", with no threshold stated [58, 262].

dentsu Curate and WPP Media's Premium Marketplace publish no selection rule: their pages describe a marketplace and name no audience match, contextual class, quality threshold or allow-list [6] [7] [18].

### Provenance stops at the source type

Sixteen products name a source type and a provider for the data they apply, and five fall short of that [193, 244, 58, 128, 13]. None documents collection method, intended use, permissions and legal basis for each segment in the sources screened [193, 244, 58, 128, 13] [246, 230, 228, 222, 227] [94, 240, 14, 188].

Microsoft's universal categories come from its own review of inventory content, coded here as a non-personal contextual input [110]. "Cross-device enhanced by Microsoft" is switched on automatically, at no cost, for any curated deal that targets a segment, and the documented opt-out route is the curator's account representative [110] [110]. The page does not describe the graph's data source or consent route [110].

Index lists its 49 active data vendors by name, with data types and markets, as of 21 October 2026, and a partner needs a vendor's permission to target its segments [94, 240, 14, 188]. Google names the contextual signals, user identifiers and secure signals it shares, and leaves the segments to the curation partner that owns them [59, 24, 248]. Audigent describes "100% 1st party audience data" and names partners such as Condé Nast and TransUnion, without saying that they supply its deal segments [193, 116, 190, 239, 194]. Its privacy policy, like Lotame's privacy notice, documents collection methods and legal bases for the company as a whole and for no curated deal [193, 116, 190, 239, 194] [208].

Among the records screened, only Lotame's case study with PHD EMEA documents provenance for a named segment [125]. For a campaign in the UK, Spain and Singapore, it gives the segment's collection route ("loyalty / credit cards and online actions across publisher sites") and states "No modeling was done" [125, 210] [125]. It does not name the party behind the card data or the permissions for the segment [125].

The five that fall short are Maestro, FreeWheel Curation Hub, OpenXSelect, dentsu Curate and WPP Media's Premium Marketplace [15, 16, 17, 6, 18]. Equativ's pages name no data provider or segment [15]. FreeWheel describes "standardized content metadata and audience signals" with no provider [16, 260] [16]. OpenX cites more than 80 directly integrated data partners and shows 28 as logos, none tied to a segment [17]. dentsu mentions "data partnerships" only, and WPP's pages name no data input [6, 18].

> **Note on interests.**
> The author has worked at Samba (TV measurement and ad data; owner of Semasio) since May 2025. He was executive vice president at Verve Group from September 2022 to August 2024 and earlier worked at Lotame and Hearts & Science. He co-leads the Signals & Measurement working group of AgenticAdvertising.org (AdCP, the Ad Context Protocol), maintains an open-source AdCP signals adaptor, and No Fluff Advisory operates two agents listed in the AdCP registry. In this chapter Lotame's Curated Marketplaces is an admitted, interest-marked row. Samba and Lotame appear among the data partners of Index Marketplaces and ClearLine, and Samba among OpenXSelect's partner logos [188] [108] [17]. None of these interests earns any evidence weight ([chapter 13](#method)).

### What the files let a buyer check

The industry's supply-chain files record who is paid for inventory and do not record who chose it [143] [143, 142, 141]. A search of ads.txt 1.1, sellers.json 1.0 and SupplyChain 1.0 on 3 November 2026 found no curator role in any of them [143, 142, 141].

sellers.json classifies each seller account as PUBLISHER, INTERMEDIARY or BOTH, by inventory ownership and by whether the advertising system pays the entity directly [142, 263]. A curator that is paid for inventory therefore appears as INTERMEDIARY or BOTH, and the file does not say what it does [142, 263]. SupplyChain 1.0 lists only entities in the direct flow of payment: each node's payment flag "should always be 1", and other participants are deferred to a future version [143]. A data or selection provider outside the payment flow therefore cannot be represented, an inference from that rule [143, 142, 141]. PubMatic's own blog says curators "don't sit in the media payment chain"; on this paper's reading, curator and data-partner payments on PubMatic run through its billing and leave no node in the chain [222, 225]. OpenRTB 2.6-202606, the bid-request protocol, carries no curator, packager or curation-fee attribute on its Pmp, Deal or Source objects; exchange-specific extensions, which could carry one, were not searched [139].

IAB Tech Lab finalised the Deal Sync API 1.0 on 6 February 2026 [261, 249, 96]. Its Curation object has three optional fields besides an extension placeholder [96]. They are `curator`, the canonical domain of the party that packaged the inventory, technology or data; `cdealid`, the deal ID in the curator's namespace; and `curfeetype`, a fee type from 0 (undisclosed) to 4 (no curation fee) [96]. The Deal object adds `pubcount` (undisclosed, single or multi-publisher), under which an inventory aggregator "may also be considered a single publisher", and `auxdata`, which flags paid non-publisher data without naming its provider [96]. No field names the data supplier or the fee recipient [96].

The specification itself warns buyers that deal criteria sent through the API are not guaranteed to be reflected in bid requests carrying the deal ID, and tells them to apply matching targeting themselves [96]. The searches found no curator or SSP confirming production support of the API by 3 November 2026, and no curator's product page opened declares a fee type [96].

The Media Rating Council (MRC) auction transparency standard of January 2026 requires OpenRTB auctioneers seeking compliance to carry complete SupplyChain objects, and requires DSPs to validate ads.txt seller IDs against sellers.json and report mis-declared chains [151]. Compliance and audit are voluntary [151]. Deal IDs do help trace a paid path after the fact [106, 283]. In the second study by ISBA, the UK advertisers' trade body, with PwC, PMP deals matched buy side to sell side above 70%, "in part due to Deal IDs facilitating impression matching" [106, 283]. Its participants were premium UK companies, observed in September and October 2022, and ISBA says the results are not representative of programmatic trading as a whole [106, 283].

### What observers found in deal-ID trading

The independent observations in the sources screened show failures in trading through apparent PMP deal IDs and tie none to an admitted product [67] [69]. Both come from Adalytics, which sells analytics to advertisers [67] [69].

Its 2024 Forbes report found that ads on the www3.forbes.com subdomain, which DeepSee.io classed as made-for-advertising, appeared to be relayed by Microsoft Xandr, TripleLift, PubMatic and Magnite as the "www" domain [67]. Some of that mis-declared inventory transacted through what appear to be PMP deal IDs, and the report gives no count of those impressions [67]. It identifies no Microsoft Curate deal [67].

Its pre-bid bot report documents ads for US brands served to declared data-centre bots, in examples dated March 2022 to November 2024 [69]. It names PMP deal IDs on Index Exchange, Magnite, TripleLift, Yieldmo, TrustX, GumGum, Kargo, PubMatic, Microsoft Xandr and Media.net [69]. The report says many of these platforms have stated publicly that they filter bots before the bid through HUMAN Security [69]. Declared crawler traffic is not necessarily fraud, and the examples give no invalid traffic (IVT) rate for any deal [69]. The Magnite examples predate ClearLine curation and none is identified as a ClearLine deal [69].

Supply-path integrity is the matrix dimension that checks a deal's inventory against ads.txt, sellers.json and SupplyChain over an observed sample. It is n/e (no evidence to score) for all 21 admitted products: for none was an observed deal sample, a duplication measure or IVT filtering by a named accredited measurer found [116, 195] [222, 225]. Index's statement that it screens and audits "every app and site on the exchange" is a self-description without a sample [19].

### Coding H6

H6, registered on 2 November 2026, states that a buyer can establish from documentation what a named curated deal contains: the sellers, the selection rule and the provenance of the data applied. It is supported if a majority of eligible products document all three elements, contradicted if a majority document none, and unresolved if access is unknown for most products or the coded elements split without a majority either way.

The products split. Of the 21 admitted products, none documents all three elements completely, and one, dentsu Curate, documents none of them [110, 248, 25, 11, 258] [6]. No element is coded access unknown, so the access clause does not apply. Coding the unopened buyer screens of Microsoft Curate and SmartPMP as access unknown would not trigger it either, since that covers two of the 21 products. Neither majority forms, and H6 is coded unresolved under the split clause.

In plain words, no admitted product lets a buyer establish from public documentation what its named deal contains; at best a buyer learns part of the seller list, a named selection logic and a source type [110, 248, 25, 11, 258] [193, 244, 58, 128, 13]. That 15 products show no seller composition is a descriptive finding outside the registered rule [110, 248, 25, 11, 258] [5, 71, 16]. Under the rule, the coding moves only if a majority of products document all three elements in full, or none of them. For composition, the strongest record would be an identified deal ID with a seller and domain list that a buyer can check against sellers.json, published in documentation or observed by a third party. A check of delivered impressions against that list would add the delivery verification that no product shows in public.



## 4. Priority, floors and sampling: what a curated deal gets


*Does a curated deal get priority, a better floor or exclusive supply by default? For most exchange products the documentation opened does not say [110, 25, 113]. Of fifteen screened, only Microsoft Curate and Google Ad Manager state auction-type and priority defaults, and nine document none of the five access attributes [110, 25, 113]. Microsoft's defaults give a curated deal no advantage [111, 250, 24, 266]. Google and Index document a route to priority that the curator chooses: a Preferred Deal, which Google defines as fixed-price and pre-auction, and a fixed price auction, which Index says has priority over all other auction bid types [24, 250] [14]. A setting the curator chooses is not a default, so the essay's claim holds for Microsoft Curate and is unresolved for the other fourteen [111, 250, 24, 266].*

> **In short.**
> The register holds 85 entries for fifteen exchange products, read on 3 November 2026 [110, 25, 113]. As first coded 62 were unknown; after the verifier's three corrections and one from the final live sweep, 60 are, and none is a recommendation or an observation [110, 25, 113] [14]. Microsoft Curate packages every curated deal as Open Auction and never overrides seller floors [110, 111]. By default its bids compete net of fees in the seller's final auction, and the supply stays open to other buyers [110, 111] [110, 111, 56]. It also samples curated deals that buyers stop bidding on, which can shrink a deal's bid requests [113]. Google Ad Manager's curated packages, in Beta, bundle Private Auctions and Preferred Deals, and Google defines a Preferred Deal as fixed-price and pre-auction [24, 250] [59, 24, 248]. A curator's floor applies only above the publisher's [59, 24]. In Index Exchange's worked example a $10 bid pays the media owner $8.08 after a 5% partner fee and a 15% Index share [109]. The OpenRTB protocol's notes leave deal priority and preferential floors to publishers, and Prebid recommends deal priority in ad-server set-up [95] [145, 265]. The Association of National Advertisers (ANA) reported in 2023 that intermediary-sold deals "rarely carry auction priority" [26]. The court's First Look, Last Look and Unified Pricing Rules findings concern Google's own exchange, and none of the three court documents read mentions curated deals [79, 156, 78]. Under the registered rule the essay's claim holds for Microsoft Curate and is unresolved for the other fourteen products; on Google Ad Manager and Index a curator can choose a transaction type that carries priority [111, 250, 24, 266] [14].

### Most products document nothing about access

Nine of the fifteen exchange products document none of the five access attributes on the pages opened on 3 November 2026: Magnite's ClearLine, PubMatic Connect, OpenXSelect, Nexxen Curated Marketplace, FreeWheel Curation Hub, InMobi's Buyer Hub, Sovrn Curated Deals, Yieldmo's YMax and Verve's Brand+ Marketplace [110, 25, 113]. For eight of them the stream found marketing pages only [110, 25, 113]. PubMatic's pages returned HTTP 403 to the stream and to its verifier, so PubMatic's unknowns record a failed fetch [110, 25, 113] [20]. The verifier did not reopen the InMobi, Sovrn and Verve pages, so those three codings are unverified [174] [277] [21]. Verve Group operates Brand+ Marketplace, and the author was executive vice-president (EVP) there from September 2022 to August 2024 [21]. The row carries the interest mark and earns no evidence weight.

The register, coded from the products' global documentation, has 17 rows because Google Ad Manager takes three, one per transaction type [110, 25, 113]. Each attribute is coded default (the platform's stated behaviour unless someone changes it), configurable (the curator or publisher chooses), recommended, observed, or unknown (the pages opened do not say). As first coded, auction type is documented for 5 rows (4 default, 1 configurable), priority for 4 (all default), floor treatment for 6 (all configurable), exclusivity for 7 (3 default, 4 configurable) and request sampling for 1 (default) [110, 25, 113].

![For most exchange products, the public record does not say how a curated deal is prioritised, floored or sampled.](figures/fig-access-register.svg)

*Figure fig-access-register. **For most exchange products, the public record does not say how a curated deal is prioritised, floored or sampled.** The access register of workstream E, with the verifier’s three corrections to Commerce Grid and the final live sweep’s reading of Index’s package page applied: for each exchange product and transaction type, whether auction type, priority, floor treatment, exclusivity and request sampling are documented as a default, a configurable setting, a recommendation or an observation, or not documented at all. 71% of entries are unknown. Units: register entries. Scope: exchange products in the register, documentation read on 3 November 2026. Sources [107], [108], [20], [109], [14], [17], [10], [110], [111], [112], [113], [114]. Caveat: Unknown means not documented in the pages opened, not absent. A documented default is the platform’s statement, not an observation of auctions. Data: data/figure-data.json#access*

The figure applies four corrections to that first coding. The verifier found Commerce Grid's knowledge base live where the stream had hit a 404 page, and re-coded three of Criteo's entries [110, 25, 113] [22, 23]. Priority moved from an asserted default to unknown, and request sampling from unknown to default, because the knowledge base describes deprioritising low-value deals [266]. Exclusivity moved to default because the glossary allows one DSP (demand-side platform) seat per demand deal [110, 25, 113] [23]. The final live sweep read Index's package page, which has the package's creator choose a First or Fixed auction with no default and defines a fixed price auction as "an auction that has priority over all other auction bid types", so Index's priority entry moves from unknown to configurable [14]. Priority is then documented for 4 rows, sampling for 2 and exclusivity for 8; unknown entries fall from 62 to 60 of 85 (71%), and nine products stay wholly unknown [110, 25, 113] [14].

Two deep profiles, verified after the register was coded, add records the figure does not show. Index Exchange's knowledge base says that, unless otherwise specified, Marketplace bids compete in the publisher's auction net of all fees and win only on a higher net bid [198, 109]. That rule nets fees and does not settle priority, since a fixed-price package carries the priority Index's package page describes; no page opened states an exclusivity default or a sampling rule [14] [198, 109]. A package floor must also exceed the publisher's floor grossed up for fees before the deal can transact [14, 199]. PubMatic's chief executive said in March 2025 that PubMatic applies no different floor rules to curated deals and that a curator wins only if the net price to the publisher is higher; no PubMatic source opened documents priority, exclusivity or sampling [226, 233].

Exclusivity in the register means buyer eligibility [110, 25, 113]. On the pages opened, only Microsoft documents whether the supply itself stays open to other buyers, and it does [110, 111, 56]. Yieldmo's "exclusive supply access" belongs to its managed YMax+ tier, with no mechanism stated [11].

### Two products state packaging and priority defaults

Microsoft Curate and Google Ad Manager are the only products that state defaults for auction packaging and priority, and they point in different directions [110, 25, 113].

Microsoft sets every curated deal's Auction Type to Open Auction [110, 111]. Curated transactions are not guaranteed deals and do not override seller floor or ad-quality settings [110, 111]. They compete in the seller's final auction against all other deal and open-exchange bids, net of fees unless a seller agrees otherwise with the curator [110, 111]. A dynamic-price deal enters the seller's first-price or second-price auction at the buyer's bid [112, 110]. Each deal serves one buyer, fixed once saved, and the same inventory stays biddable in the open exchange [110, 111, 56]. Microsoft leaves one route to an advantage: curators may bring preferred inventory access and negotiated seller rates, and a curated deal can draw supply from deals negotiated directly with sellers, on terms the pages do not describe [56, 110].

Google's curation partners can bundle Private Auctions and Preferred Deals into packages; the curation page does not mention Programmatic Guaranteed [24, 250]. Ad Manager defines a Private Auction as an auction among buyers allowlisted by company, and a Preferred Deal as a fixed-price, pre-auction transaction the publisher negotiates with specific buyers [24, 250]. Partners negotiate both directly with publishers [24, 250] [24]. A package built on Preferred Deals therefore carries pre-auction access by definition of the type, which the curator chooses to bundle [24, 250]. Google's publisher-curated Marketplace Packages, a closed beta limited to CTV (connected TV) and streaming inventory, run at Open Auction priority by default; they are single-publisher and outside the multi-publisher subset [25] [25]. Every Google curation page opened carries a Beta notice, so US availability and production use cannot be coded [59, 24, 248].

| Product and transaction type | Auction type | Priority | Floor treatment | Exclusivity (buyer eligibility) | Request sampling | Evidence |
|---|---|---|---|---|---|---|
| Microsoft Curate, curated deal | Default: Open Auction, first or second price as the seller runs it | Default: none | Configurable; seller floors never overridden | Default: one buyer; supply stays open | Default: Microsoft-controlled | [110, 111] [112, 110] [113] [110, 111, 56] |
| Google Ad Manager, Private Auction in a package | Default: auction | Unknown | Configurable, only above the publisher's floor | Default: allowlisted buyers | Unknown | [24, 250] [59, 24] |
| Google Ad Manager, Preferred Deal in a package | Default: fixed price | Default: pre-auction | Configurable, only above the publisher's floor | Default: specific buyers | Unknown | [24, 250] [59, 24] |
| Google Ad Manager, Marketplace Packages (single publisher) | Default: auction | Default: Open Auction priority | Configurable by the publisher | Configurable: public or exclusive | Unknown | [25] |
| Index Marketplaces, deal ID | Configurable: first or fixed price, no default stated | Configurable: a fixed price auction "has priority over all other auction bid types" | Unknown beyond a fixed-price fee rule | Configurable: selected buyers | Unknown | [109] [14] |
| Maestro by Equativ, deal | Unknown | Unknown | Configurable: market, custom or fixed | Configurable: buyer seats or DSPs | Unknown | [114, 15] |
| TL Direct (TripleLift), private marketplace (PMP) and curated deals | Unknown | Unknown | Configurable: fixed, floor or hybrid | Configurable | Unknown | [130, 117] |
| Commerce Grid (Criteo), deal IDs | Unknown | Stream: asserted default; verifier: unknown | Unknown | Verifier: default, one DSP seat per deal | Verifier: default deprioritisation | [266] [110, 25, 113] [22, 23] |
| Nine other products | Unknown | Unknown | Unknown | Unknown | Unknown | [110, 25, 113] |

### Floors bind upward, and the fee comes out of the bid

No product in the register documents a default floor advantage for curated deals, and the two that state a rule make the seller's or publisher's floor bind [59, 24] [110, 111]. Google applies a curator's package floor only where it exceeds the publisher's floor for the impression [59, 24]. Microsoft lets a curator set a floor above, at or below the seller's but never overrides seller floors [112, 110] [110, 111]. On this paper's reading, a lower curator floor cannot clear a bid below the seller's floor [112, 110]. Equativ and TripleLift document floor options without defaults [114, 15] [130, 117]. TripleLift says its TL Spark layer adjusts floors, without saying whether it does so on curated deals by default [130, 117].

Four products document that the curator's charge comes out of the buyer's bid before the auction ranks it [110, 111] [24, 59, 248] [109] [266]. Microsoft deducts any curator margin and its own tech fee before the bid enters the seller's auction, unless seller and curator agree otherwise [115, 111] [110, 111]. Google takes the partner's fee out of the bid price before the auction [24, 59, 248]. Commerce Grid's knowledge base does the same with a Criteo tech fee and the curator's margin, charged only on a win [266] [22]. Index gives the arithmetic: a $10 gross bid less a 5% partner fee and a 15% Index revenue share pays the media owner $8.08, the partner $0.50 and Index $1.42 [109]. The 15% is a worked example, since Index negotiates its share with each media owner [109]. The US Media Rating Council (MRC) auction standard encourages auctioneers to apply fees before the auction ("net bidding"), though it does not name curator fees [151]. [Chapter 6](#fees) covers the amounts and who sees them.

### Sampling: the documented defaults reduce a deal's traffic

Both request-sampling defaults in the register, Microsoft's as first coded and Commerce Grid's after verification, work against the curated deal [110, 25, 113]. Microsoft's Curation Optimization samples curated deals that a buyer has stopped bidding on, without curator action [113]. On a deal with little bidding or spend, curator and buyer may see fewer matched impressions and bid requests over time, which Microsoft attributes to the sampling applied to the deal [113]. No threshold, rate or opt-out is stated [113].

Commerce Grid's knowledge base describes deprioritising low-value curated deals in nearly the same words [266]. The verifier recorded the parallel without inferring a cause, and a two-week throttling threshold its fetch reported is unverified [266] [22]. On that reading, Commerce Grid's marketing claim to prioritise high-value deals in auctions describes throttling [111, 250, 24, 266]. On Microsoft, the bid requests a buyer sees under a curated deal ID depend on exchange sampling as well as on the curator's selection, an inference from one page that gives no rates [113]. [Chapter 3](#contents) takes up what a deal ID contains.

### What the standards say

OpenRTB leaves priority and floors to publishers, MRC's standard requires them to be disclosed, and none of the standards sets a default for curated deals [95] [151] [151]. OpenRTB 2.6-202606 (11 June 2026) is the current text of IAB Tech Lab's bid-request protocol [264, 139]. Its Pmp object defaults to an open auction accepting all bids [139]. Its Deal object carries a floor, an auction-type override, buyer-seat and advertiser-domain allow-lists and a guarantee flag [139]. Those wire-format defaults establish nothing about how an exchange configures a curated deal [139].

The non-normative implementation notes (section 7.3) say a publisher may configure an exchange to rank all Deal ID bids above open-auction bids, or to run a price auction between them with different floors by class [95]. They recommend a Deal ID for any non-price prioritisation or preferential seat floor [95]. Their use cases include a first-look agreement locked to one buyer, where priority is expected, and Deal IDs as package tokens with their own floors, where price decides; the notes never use the word default [95]. On this paper's reading the package-token case is the nearest to a multi-publisher curated deal [95]. IAB Tech Lab's Deal Sync API 1.0 adds that curated deals may vary in price request by request, so suppliers give bid guidance in place of a floor [96].

Prebid, the open-source header-bidding software, recommends that deal line items in the publisher's ad server rank above the line items for open-market bids [145, 265]. The page concerns header-bidding deals and mentions no curation, and none of the 1,472 URLs in the docs.prebid.org sitemap on 3 November 2026 has "curat" in its path [145, 265]. This is ad-server guidance, neither a deployed default nor advice on curated deals [145, 265].

MRC's Digital Advertising Auction Transparency Standards (January 2026) cover any company that evaluates bids, selects winners and sets prices, even in a non-final auction [151]. On this paper's reading an SSP (supply-side platform) running a curated deal's auction is in scope; compliance is voluntary [151]. They require disclosure of auction type and clearing method, of reserve prices before and after each auction, of the priority and business rules that decide winners beyond price, and of direct deals that supersede an auction [151] [151] [151]. These requirements sit in the section headed for closed-loop auctioneers, and how far they bind an OpenRTB exchange is left to the auditor [151]. They say floors should not be set per buyer and, as an encouraged practice in OpenRTB, should be managed through a Deal ID structure [151]. In the sources screened no auctioneer is recorded as accredited against them, and two of the fifteen register products publish auction-type and priority defaults [151] [110, 25, 113].

The ANA's 2023 Complete Report covered 21 US marketers, $123 million of spend and 35.5 billion impressions from September 2022 to January 2023 [26]. It sorted PMP deals by domain count and called those with 11 to 500 domains auction packages [26]. It stated that publisher-sold deal IDs "often include prioritized access to premium supply", while intermediary-sold ones bundle many publishers and "rarely carry auction priority" [26]. Both are unquantified observations of one advertiser-association sample, and the same report's PMP definition allows preferential access in some cases [26].

### What the court found, and what a judgment would change

The court documented access advantages for Google's exchange inside Google's ad server, and the remedies proposed for them stop at Google's open and private auctions [79, 156, 78]. On 17 April 2025 the US District Court for the Eastern District of Virginia found that Google had monopolised the worldwide open-web display publisher ad server and ad exchange markets (Dkt 1410) [78, 156]. It also found that Google had unlawfully tied DFP, its ad server, to AdX, its exchange [78, 156].

First Look required DFP publishers to give AdX a first right of refusal on every impression, even when they preferred other exchanges, so AdX advertisers could win when rivals offered more [78]. Last Look let AdX see the winning header-bidding bid, entered in DFP as a floor, and outbid it; by 2019 Google agreed to remove it and introduced Unified Pricing Rules at the same time [78]. Dynamic revenue share let AdX vary its take impression by impression, around 20% overall and lower on competitive impressions; that is an exchange take rate, a different series from curation fees [78]. Unified Pricing Rules barred publishers from setting higher floors for AdX than for other exchanges [78, 156]. These are observed, court-found advantages for one exchange, and Google is precluded from relitigating them in the multidistrict litigation in New York [79, 156, 78] [131, 132].

The remedies opinion of 2 September 2026 (Dkt 1858) rejected every structural remedy and accepted most behavioural ones as modified [156, 80]. It decided a bar on reimplementing First Look and Last Look and the deprecation of Unified Pricing Rules for all indirect transaction types, with publishers free to set different floors by buying tool and exchange [156]. It also decided that publishers may route programmatic guaranteed and preferred deals through Prebid [156]. It set a six-year term, with effect 60 days after a judgment is entered [156, 80].

The joint proposed final judgment of 2 October 2026 (Dkt 1861-1) would apply solely to Indirect Demand, meaning open or private auctions, for open-web display, and exclude Display & Video 360 (DV360), AdSense and AdMob [79]. It would bar First Look and Last Look for that demand, deprecate Unified Pricing Rules, and within 15 months give publishers an optional server-to-server route for programmatic guaranteed and preferred deals through Prebid [79]. Disputed brackets remain, among them 6 against 12 months for AdX-Prebid interoperability [79]. On CourtListener's docket copy of 29 October 2026 no judgment had been entered and no appeal filed [80]. The courts' own PACER records system was not checked, and nothing is in force [80].

For deals, the scope line matters. On this paper's reading, a curated Ad Manager package bundling Private Auctions would sit inside Indirect Demand if a judgment in this form were entered [79] [24, 250]. One built on Preferred Deals would be touched only by the optional Prebid route [79] [24, 250]. None of the three documents mentions curators, curated deals or third-party deal IDs [79, 156, 78]. The one use of "curate" is testimony that advertisers expect their DSPs to curate or prioritise exchanges, which is buyer-side selection [79, 156, 78]. [Chapter 9](#competition) sets out the full record.

### Coding H4 and the essay's claim

H4 is registered as a descriptive question with a coding rule, not a verdict. The essay's claim holds for a product whose documented defaults give no advantage in priority, floor or exclusivity, with no observed advantage recorded. It fails where a documented default gives curated deals one of those advantages. It is unresolved where defaults are undocumented in public, and protocol fields never establish a deployed default. The author's essay "The Great Repackaging" (16 July 2026) says a curated deal carries "no preferential access, no first look, no floor advantage by default" [1]. It is the position under test and carries no evidence weight.

- **Microsoft Curate: holds.** Open Auction packaging, no priority, binding seller floors and one buyer on supply that stays open, with no observed advantage recorded [111, 250, 24, 266]. Two qualifications travel with the code: deals can draw on seller-negotiated supply whose terms the pages do not describe, and the sampling default can cut a thinly bid deal's traffic [56, 110] [113].
- **Google Ad Manager Curation: unresolved, with a priority the curator can choose.** Google defines the Preferred Deal as fixed-price and pre-auction, and a curator who bundles one, negotiated with the publisher, gives the package that position [111, 250, 24, 266] [24, 250]. The pages opened set no default transaction type for a package, Private Auction packages have no documented priority, and curator floors apply only above the publisher's [24, 250] [59, 24]. In the register's own terms a setting the curator chooses is configurable, not a default, so the product's default stays undocumented.
- **Index Marketplaces: unresolved, with a priority the curator can choose.** Bids compete net of fees unless otherwise specified, and a package floor must exceed the publisher's floor grossed up for fees [198, 109] [14, 199]. The package's creator must choose a First or Fixed auction, with no default stated, and Index defines a fixed price auction as having "priority over all other auction bid types"; no exclusivity default is stated, and the source is Index's own knowledge base [14].
- **Commerce Grid: unresolved.** Its knowledge base describes the prioritisation claim as deprioritising low-value curated deals in auction processing, which is request sampling; priority and floors stay unknown [111, 250, 24, 266] [110, 25, 113].
- **The other eleven products: unresolved.** Equativ and TripleLift document configurable floors or buyer eligibility but no default priority [111, 250, 24, 266]. PubMatic's chief executive describes floors without an advantage, but priority is undocumented [226, 233]. The rest of the nine with no documented access attribute stay unknown [110, 25, 113].

In plain words, the essay's sentence is true of Microsoft Curate as documented and cannot be tested as a default on the public documentation opened for the other fourteen products [111, 250, 24, 266]. No product documents a default priority, floor advantage or exclusivity for curated deals. Two document a priority the curator can choose: Google through a publisher's Preferred Deal and Index through a fixed price auction [24, 250] [14]. Every coded entry is a platform's own statement, and none observes an auction [110, 25, 113]. Both routes to priority run through a deal type or a fixed price, and Microsoft's one open route to an advantage runs through a deal negotiated with the seller [24, 250] [14] [56, 110]. That fits the ANA's 2023 observation and OpenRTB's placement of priority with publishers [26] [95]. The evidence supports a narrower sentence: on Microsoft Curate a curated deal carries no priority or floor advantage by default; on Google Ad Manager and Index a curator can choose a transaction type that carries priority; elsewhere the pages opened publish no priority or floor default [111, 250, 24, 266] [110, 25, 113]. The first edition coded Index as holding and Google as failing. The final live sweep found that the first code read an undocumented exclusivity default and an undefaulted auction choice as no advantage, and the second read a curator's choice as a default ([chapter 13](#method)). Published defaults from the other products, or an independent observation of curated-deal auctions, would move the fourteen unresolved codes ([chapter 12](#next)).



## 5. Who owns the curators


*Who owns the companies that sell curated deals and curation data in the US, and are fewer of them independently controlled than in 2023? Since 2023 five of the eleven registered data suppliers have passed to a credit bureau, an advertising holding company, a TV data company or a contextual data company [49, 41, 38, 36, 44, 34, 50], while no exchange in the registers lost its independence. Documented independent suppliers fell from four of eleven to two of ten, but control is undocumented for six of the eleven 2023 suppliers, so H5 is unresolved on its registered rule [49, 41, 38, 36, 44, 34, 50].*

> **In short.**
> Five of the eleven curation data suppliers in the registers passed to operating groups after 2023: Audigent to Experian, a credit bureau; Lotame and LiveRamp to Publicis; Semasio to Samba TV; and Adloox to Scope3, then to Peer39 [38, 39] [40, 41] [49, 48] [36] [37, 50]. Oracle retired every Oracle Advertising product, Moat and Oracle Contextual Intelligence among them, on 30 September 2024 [34]. Funds managed by Novacap took Integral Ad Science (IAS) private on 23 December 2025, and DoubleVerify's sale to the parent of The Nielsen Company (US), at $13.60 a share, was pending at the cutoff [43, 44] [51, 52]. On the strict rule, documented independent suppliers fell from four of eleven in 2023 to two of ten active at the cutoff; on a looser reading that counts a supplier with undocumented control as independent, from ten to five [49, 41, 38, 36, 44, 34, 50]. Control is undocumented for six of the eleven 2023 suppliers, so H5 is unresolved on its registered rule, although every supplier whose code changed moved into an operating group or out of the market [49, 41, 38, 36, 44, 34, 50]. No exchange in the registers lost its independence between them: Magnite, PubMatic and Nexxen stayed listed, Equativ stayed with its financial sponsor, Microsoft kept Microsoft Curate, and Teads went from Altice to a listed buyer [212] [231] [30] [29] [56] [35, 42]. Scope3's rebrand to Apostra changes no count, and Multilocal's $4 million facility from OAREX is debt that left ownership unchanged on the company's own account [54] [55].

### The record since 2021

Acquisitions make up most of the record since 2021. One owner left the market, and the sources screened by the cutoff record no insolvency [34]. Each event is classified under shared rule 11 and Q4, which keep acquisition, merger, rebrand, restructuring, insolvency, product retirement and pivot apart. Each is dated from the filing or release that settles it, where one was opened.

| Date | Event | Type | Terms where disclosed | Source |
|---|---|---|---|---|
| 29 March 2021, announced | Vista Equity Partners agrees to buy a majority stake in TripleLift; no source opened confirms completion | Acquisition (majority stake), announced | Not disclosed | [27] [27] |
| 30 April 2021, closed | Magnite closes its purchase of SpotX from RTL Group | Acquisition | $640 million in cash and 12,374,315 Magnite shares, about $1.14 billion at the 29 April 2021 share price | [28] [28] |
| 2022 | Smart AdServer renamed Equativ | Rebrand | Not applicable | [29] [29] |
| 2022, day not captured | Tremor International acquires Amobee | Acquisition | Price not reached in the portion of the filing read | [30] [30] |
| 6 June 2022, completed | Microsoft acquires Xandr, Inc. (date from Microsoft's 2022 annual report; not re-checked by the verifier) | Acquisition | Not reached in the portion read | [31] [31] |
| 27 February 2023 | Bridgepoint confirmed as Equativ's "principal investor" (the headline says "new majority shareholder") | Change of control to a financial sponsor | Not disclosed | [29] [29] |
| January 2024 | Tremor International renamed Nexxen International | Rebrand | Not applicable | [30] [30] |
| 12 June 2024 | Equativ announces a "merger" with Sharethrough, which it later calls "Equativ's acquisition of Sharethrough in June 2024" | Merger, later described by Equativ as an acquisition | Not disclosed; the parties' combined net recurring revenue above $200 million | [32, 33] [32, 33] |
| 17 June 2024, notice; 30 September 2024, effective | Oracle ends all Oracle Advertising products and support, Moat Analytics and Oracle Contextual Intelligence among them | Product retirement (market exit) | Not applicable | [34] [34] |
| 1 August 2024, agreement; 3 February 2025, closed | Outbrain buys Teads from Altice Teads | Acquisition | $625 million in cash, subject to adjustments, and 43.75 million Outbrain shares; two Altice Teads board designees | [35] [35, 42] |
| 31 October 2024 | Samba TV announces its acquisition of Semasio | Acquisition, announced as made | Not disclosed | [36] [36] |
| 4 November 2024 | Scope3 announces its acquisition of Adloox (Adloox's own post is dated 5 November) | Acquisition, announced | Not disclosed | [37] [37, 50] |
| 4 December 2024 | Experian announces that it "has acquired" Audigent | Acquisition, announced as completed | Not disclosed; Experian's acquisition spend of $1,244 million in the year to 31 March 2025 does not break out Audigent | [38, 39] [38, 39] |
| 17 and 18 February 2025 | Nexxen cancels its depositary interests on London's Alternative Investment Market and trades on Nasdaq only | Listing change | Not applicable | [30] [30] |
| 6 March 2025, announced; April 2025, completed | Publicis acquires 100% of Lotame Solutions, Inc. | Acquisition | Cash consideration €127 million; provisional goodwill €96 million | [40, 41] [40, 41] |
| 6 June 2025 | Outbrain renamed Teads Holding Co. | Rebrand (corporate rename) | Not applicable | [42] [35, 42] |
| 9 June 2025 | The Sharethrough brand is folded into Equativ | Rebrand (brand retirement) | Not applicable | [33] [32, 33] |
| 24 September 2025, agreement; 23 December 2025, completed | Funds managed by Novacap take Integral Ad Science (IAS) private | Acquisition (take-private) | $10.30 a share in cash, valuing IAS at about $1.9 billion; Nasdaq listing ended | [43, 44] [43, 44] |
| 26 September 2025, agreement; 29 December 2025, closed | Comscore's recapitalisation retires all Series B preferred stock held by Charter Communications, Liberty Broadband and a Cerberus-advised vehicle | Restructuring | Each holder gets Series C preferred, common shares and $2.0 million due June 2028, and one seat on a seven-member board | [45, 46] [45] |
| 26 November 2025, completed | Omnicom acquires The Interpublic Group (IPG), parent of KINESSO's client marketplaces (the release does not mention KINESSO) | Acquisition (all-stock) | 0.344 Omnicom shares per IPG share; legacy holders own about 60.6% and 39.4% | [47] [47] |
| 28 February 2026, announced end | Media buying through Microsoft Invest, Microsoft's DSP (demand-side platform), is announced to end; no page opened confirms the end | Product retirement, announced | Not applicable | [56] |
| 17 May 2026, announced; 2 November 2026, completed | Publicis, through MMS USA Holdings, Inc., acquires LiveRamp, whose shares leave the New York Stock Exchange (NYSE) | Acquisition (take-private) | $38.50 a share in cash; enterprise value $2.167 billion; equity value $2.546 billion | [48, 49] [49, 48] |
| 3 June 2026 | Peer39 announces that it "acquired Adloox"; the post names no seller | Acquisition, announced | Not disclosed | [50] [37, 50] |
| 6 August 2026, announced; pending at the cutoff | Neptune BidCo US Inc., parent of The Nielsen Company (US), LLC, agrees to acquire DoubleVerify; stockholders approve on 29 October 2026 | Acquisition (take-private), pending | $13.60 a share in cash; enterprise value about $2.15 billion | [51, 52, 53] [51, 52] |
| 23 September 2026 | Scope3 announces that it "is becoming Apostra"; the footer names Apostra, Inc. | Rebrand, legal form not stated | Not applicable | [54] [54] |
| 1 October 2026 | OAREX provides Multilocal a receivables-collateralised "ADVANCE" facility | Financing (debt; ownership unchanged) | $4 million | [55] [55] |
| Date not stated on Microsoft's documentation | Xandr Curate renamed Microsoft Curate; sign-in still runs at curate.xandr.com | Rebrand | Not applicable | [56] |
| By 2 November 2026; date not in the sources screened | Samba TV trades as Samba | Rebrand | Not applicable | [57] |

Three rows needed care. Equativ called the Sharethrough deal a merger in June 2024 and an acquisition a year later; the sources opened give no legal closing day or ownership split [32, 33]. LiveRamp's closing 8-K names Publicis Groupe only for one section of the merger agreement, so Publicis as ultimate owner rests on its own release [49, 48] [48]. DoubleVerify moved its expected closing from the end of 2026 to the first quarter of 2027 [51, 52].

Some vendor pages lag the filings. IAS's 2024 launch page for IAS Curation still reads "Nasdaq: IAS" after the take-private [58, 59] [58]. Adloox's learning page of 17 July 2026 still says "Adloox is now part of Scope3", six weeks after Peer39 announced that it had bought Adloox [12] [37, 50]. Ownership in this chapter therefore comes from filings and the parties' releases, and vendor copyright lines are corroboration only [277, 130, 5, 13].

Several rows touch the author's interests. He has worked since May 2025 at Samba, which owns Semasio [36]. Lotame, now owned by Publicis, is a former employer, as are Hearts & Science, an Omnicom agency, and Verve Group, where he was executive vice president from September 2022 to August 2024 and which operates the Brand+ Marketplace [41, 21]. He co-leads the Signals & Measurement working group of AgenticAdvertising.org, which publishes AdCP (the Ad Context Protocol), and maintains an open-source AdCP signals adaptor. No Fluff Advisory operates two agents listed in the AdCP registry, and Apostra says it is built on AdCP [54]. None of these relationships earns evidence weight. The Semasio and Lotame rows rest on the acquirers' releases and Publicis's regulated half-year report [36] [40, 41], and Verve has no register row because its shareholder structure was not sourced [21].

### Two supplier registers

On the strict rule, documented independent curation data suppliers fell from four of eleven in 2023 to two of ten active at the cutoff [49, 41, 38, 36, 44, 34, 50]. The registers list companies selling curated deals or curation data in the US, each with its ultimate controlling group; whether each sold in the US in 2023 was not checked by the ownership stream. Control means majority ownership or documented control in filings and official releases, and unknown control stays unknown. Rebrands change no count. A financial sponsor such as a private-equity fund is coded as a non-operating owner [49, 41, 38, 36, 44, 34, 50].

| Supplier | 2023 control as documented | 2023 code | Controlling group at the cutoff (owner type) | Cutoff code | Source |
|---|---|---|---|---|---|
| Audigent | Private; cap table not opened | Unknown | Experian plc (credit bureau) | Not independent | [38] [38, 39] |
| Lotame | Private; "independent" in its acquirer's words | Independent | Publicis Groupe, 100% (holding company) | Not independent | [40, 41] [40, 41] |
| LiveRamp | Listed on the NYSE; no controlling holder in the sources opened | Independent | Publicis Groupe via MMS USA Holdings, Inc. (holding company) | Not independent | [49, 48] [49, 48] |
| Semasio | Private; prior owner not stated | Unknown | Samba TV, Inc., trading as Samba (TV data and measurement company) | Not independent | [36] [36] |
| Scope3, now Apostra | Private; no filing | Unknown | Apostra, Inc. (private) | Unknown | [54] [54] |
| Adloox | Private; prior owner not stated | Unknown | Peer39 (contextual data company) | Not independent | [37, 50] [37, 50] |
| Oracle Advertising (Oracle Contextual Intelligence, Moat) | Oracle Corporation (software company) | Not independent | Retired on 30 September 2024 | Exited | [34] [34] |
| Integral Ad Science | Listed on Nasdaq; controlling holder not established | Independent | Funds managed by Novacap (private equity) | Independent (financial sponsor) | [44, 43] [43, 44] |
| DoubleVerify | Listed on the NYSE | Independent | Listed; sale to Nielsen's parent pending | Independent until closing | [51, 52, 53] [51, 52] |
| Comscore (Proximic) | Listed; Series B holders held board designation rights | Unknown | Listed; three holders with one board designee each | Unknown | [45, 46] [45] |
| Multilocal | Private; no filing opened | Unknown | Private; the OAREX facility is debt | Unknown | [55] [55] |

![Documented independent curation data suppliers fell from four to two; most suppliers’ control is not documented in public.](figures/fig-independence.svg)

*Figure fig-independence. **Documented independent curation data suppliers fell from four to two; most suppliers’ control is not documented in public.** The two supplier registers of workstream D (H5): companies selling curated deals or curation data in the US, coded by control as documented in filings and official releases. On the strict rule the documented independent count fell from 4 of 11 to 2 of 10 active; on the looser reading that treats a private supplier with no documented parent as independent, from 10 to 5. Units: suppliers. Scope: curation data suppliers in the two registers, US. Sources [49], [41], [38], [36], [51], [44], [34], [50]. Caveat: Six of eleven 2023 suppliers have no documented control, so the strict count rests on a minority of the register; H5 is coded on both readings (chapter 5). Exchanges are tracked separately. Data: data/figure-data.json#suppliers*

The looser reading counts every supplier whose control is undocumented as independent, the listed Comscore included. On it the count fell from ten of eleven in 2023 to five of ten at the cutoff, with Audigent, Lotame, LiveRamp, Semasio and Adloox leaving independence and Oracle leaving the market [49, 41, 38, 36, 44, 34, 50]. The stream first reported four at the cutoff; the verifier recomputed five from the stream's own tables, and the corrected figure is used here [49, 41, 38, 36, 44, 34, 50].

Six of the eleven 2023 rows have no documented control (Audigent, Semasio, Scope3, Adloox, Comscore and Multilocal), and for the five private companies no cap table or registry filing was opened [49, 41, 38, 36, 44, 34, 50]. The four 2023 independents rest on thin records too. Lotame's code rests on Publicis describing it as independent, in its release and its half-year report, as it bought it [40, 41] [40]. Without that description the row would be unknown and the strict count would start at three, by this paper's arithmetic [49, 41, 38, 36, 44, 34, 50]. LiveRamp's, IAS's and DoubleVerify's codes rest on listing without a named controlling holder in the sources opened, and IAS's pre-deal holder was not established [43, 44] [49, 51].

Comscore is the closest call. Its closing 8-K of 29 December 2025 names the three holders, gives each one designee on a seven-member board and sets a voting threshold of 16.66% of the common stock, but states no combined percentage [46]. This chapter keeps Comscore unknown at both dates, as the figure does. Coding it independent at the cutoff would make the strict count four to three, and its 2023 row needs a source for the earlier designation rights before it can move [45] [49, 41, 38, 36, 44, 34, 50].

Two further changes would lower the cutoff count. DoubleVerify becomes not independent when its sale closes, because the buyer is an operating measurement company rather than a financial sponsor, and IAS is then the only documented independent supplier left [51, 52]. Counting private-equity ownership as control would remove IAS too [49, 41, 38, 36, 44, 34, 50] [43, 44].

Neither register is a census: membership follows the brief's vendor universe, and entrants since 2023 were not sourced by the ownership stream [49, 41, 38, 36, 44, 34, 50]. The vendor sweep found Experian's own Curated Deals, Deal Engine and Deal Marketplaces [255], PMPs (private marketplaces) in one sentence on Samba's homepage [268] and a Permutive curation product sold through SSP (supply-side platform) curation seats [267]. Google Ad Manager's curation page lists five partners absent from the supplier registers: Adelaide, Permutive, PrimeAudience, SWYM.ai and System1 [59] [59]. Adelaide and Dstillery, which sell or sold deal IDs, lack rows because no source opened documents their control [60, 61]. By this paper's inference from the register rules, the Experian line would enter as not independent, as Oracle's product line did, and the Samba line as unknown, because Samba TV's own control was not sourced [49, 41, 38, 36, 44, 34, 50]. On the same inference, entrants with undocumented control would narrow the looser fall and leave the strict one unchanged [49, 41, 38, 36, 44, 34, 50].

### What the registers show, and what they do not

They show absorption. At the cutoff the five suppliers that passed to operating groups sit with a credit bureau (Experian), an advertising holding company (Publicis, twice), a TV data and measurement company (Samba TV) and a contextual data company (Peer39) [49, 41, 38, 36, 44, 34, 50]. The next pending owner is a measurement company, Nielsen's parent, for DoubleVerify [51, 52]. The absorbed suppliers still appear under their own names: Google's beta curation page lists "Audigent/Experian" twice and "LiveRamp, Inc." among its twelve partners [59]. Experian's own curation page sells three services without mentioning Audigent [255], and on the sweep's read, which the verifier did not re-check, Audigent's SmartPMP page names no Experian ownership [193].

Exchange ownership moved differently. Ad-tech companies bought each other: Magnite bought SpotX in 2021, Tremor bought Amobee in 2022 and Equativ absorbed Sharethrough in 2024 [28] [30] [32, 33]. None of the registered exchange operators lost its independence: Magnite, PubMatic and Nexxen were listed at both dates [212] [231] [30]. PubMatic's insiders hold substantial control through dual-class shares, and Nexxen's four largest holders own about 52.3% together, with no single majority holder named [231] [30]. Equativ has had Bridgepoint as principal investor since February 2023 [29]. Vista's majority stake in TripleLift was announced in March 2021, with completion unconfirmed in the sources opened [27]. Microsoft has owned Xandr since 6 June 2022, on a reading of its annual report that the verifier did not re-check, and documents Microsoft Curate as its own product [31] [56]. Among the registered operators only Teads changed hands between the registers: Outbrain bought Teads from Altice and took its name [35, 42]. Altice Teads received 43.75 million Outbrain shares and two board designees, so Teads' control at the cutoff is coded unknown [35, 42]. KINESSO's client marketplaces passed from one holding company to another when Omnicom bought IPG ([chapter 8](#marketplaces)) [47].

The registers count companies and carry no trading volume, so they say nothing about concentration or share. The filings read do not fill that gap. In the Business and risk-factor sections of seven annual filings for fiscal 2025, from Magnite, PubMatic, Criteo, Comscore, The Trade Desk, Viant and Nexxen, none discloses a numeric take rate or a deal, PMP or curated share of its own activity [212, 231, 284, 30, 285, 282]. Each describes its fee as generally a percentage of spend, with flat fees or fixed prices per thousand impressions at Nexxen and Viant and negotiated SSP, technology and DSP fees at Criteo; the verifier re-opened five of the seven [212, 231, 284, 30, 285, 282]. The share series are in [chapter 2](#share) and the fees in [chapter 6](#fees).

No event here shows that a format, a metric or a product failed. Oracle's retirement is an exit by one owner, and this chapter reads no cause into it [34]. A change of owner does not show that a product changed either: Samba TV said Semasio would continue as an independent business under a general manager [36]. Whether any acquirer steers a supplier's deals toward its own clients or products is not established in the sources screened. Apostra's agentic platform is described in [chapter 10](#agents) [54].

### Coding H5 and Q4

**H5 is unresolved.** H5 is supported if the cutoff register holds fewer independently controlled suppliers than the 2023 register on the same definition, contradicted if it holds as many or more, and unresolved if control cannot be established from public records for enough suppliers to compare. Control is documented for five of the eleven 2023 suppliers and for seven of the ten active at the cutoff [49, 41, 38, 36, 44, 34, 50]. This paper reads a baseline whose majority is unknown as the registered unresolved case, and the verifier read the finding the same way [49, 41, 38, 36, 44, 34, 50]. The counts fall on both readings, from four to two and from ten to five [49, 41, 38, 36, 44, 34, 50]. The looser reading would meet the supported rule, but it counts unknown control as independent, which the registered outcome rules out ("unknown control stays unknown").

The lean is consistent. Suppose, as no recorded event contradicts, that Scope3 and Multilocal held the same control status at both dates. Then every assignment of the unknown rows leaves the cutoff count at least one below the 2023 count, by this paper's inference from the two registers [49, 41, 38, 36, 44, 34, 50]. That lean is not a verdict, because the assumption is untested and the cutoff register omits entrants. Three sets of records would decide H5: the 2023 ownership of Audigent, Semasio, Scope3, Adloox and Multilocal; Comscore's governance before its recapitalisation; and the control of the entrants the vendor sweep found [49, 41, 38, 36, 44, 34, 50] [45].

**Q4 is coded event by event.** Its rule lets a filing or a party's official release settle an event and uses reporting only to locate one. Every row except Microsoft Invest's rests on a filing or on a party's own release, post or page opened by the streams or their verifiers. The Invest row rests on Microsoft's announced end as recorded with its Curate documentation of 21 October 2025, which still names Invest as a route; the notice itself was not opened in these streams [56]. Several rows lack a key date. TripleLift's sale to Vista has no confirmed completion, Equativ's Sharethrough deal no closing day and the Amobee purchase no day [27] [32, 33] [30]. The renames of Xandr Curate to Microsoft Curate and of Samba TV to Samba carry no date in the pages cited [56, 57]. By type, the table holds thirteen acquisitions or mergers announced as made or completed, one pending acquisition, one announced majority sale without a confirmed close and one change of control to a financial sponsor. It also holds seven rebrands, two product retirements, one restructuring, one listing change and one financing, and no insolvency [34] [45] [55]. Under shared rule 11 none of these events, by itself, establishes that a format or a metric failed.



## 6. What curation costs, and who says so


*Does a curated deal cost a buyer more than the same impressions bought in the open auction, and who discloses the charges along the way? No source screened by the cutoff compares the two routes on the same opportunities, so H2a is unresolved [26] [99]. No admitted product publishes its charge rates or amounts, and 13 of the 21 state no charge type in the documentation read [212, 231, 281, 115, 24] [115, 24, 109, 117, 266] [116, 12, 209, 128]. No buyer-facing fee screen was observed and six of the 13 have no pricing evidence, so H2b is unresolved on its registered rule [212, 231, 281, 115, 24] [116, 12, 209, 128].*

> **In short.**
> In the Association of National Advertisers' (ANA) 2023 study, $1,000 entering a demand-side platform (DSP) paid $80 in DSP platform costs, $80 in DSP data and additional costs and $130 in supply-side platform (SSP) costs [26, 63]. Sellers received $710, and curation had no line of its own [26, 63] [287, 99, 26]. In the ANA benchmark for November 2024 to March 2025, private-marketplace (PMP) CPMs (cost per thousand impressions) ran above open-marketplace CPMs on both the median and the spend-weighted aggregate [63]. The two routes bought different impressions, so the gap does not measure what the same opportunities cost by either route [26] [66]. In Jounce Media's client-only sample, reported second-hand, half of curated deal IDs carried no curator fee beyond the exchange's take, 34% a margin (median 14%), 10% a fixed CPM (median $0.48) and 7% a static price [9, 64, 281]. Six exchange products document fee types and bases, and none publishes a rate [212, 231, 281, 115, 24] [115, 24, 109, 117, 266] [226, 222, 231]. Every "no fee" claim found covers one party's charge, and none speaks for the rest of the deal path [188, 187, 200] [117]. IAB Tech Lab's Deal Sync API can declare a curator's fee type but has no field for its amount, and no production declaration was found [96] [96]. In the sources screened, no named deal has all four charges disclosed with payer, collector, base and period [115, 216, 290, 64].

### Four series, kept apart

Curation costs reach the public record in four series, and no source screened reports all four for one deal [115, 216, 290, 64]. The curation fee goes to the curator, the data fee to the supplier of the segment, the SSP take rate to the exchange and the DSP take rate to the buying platform. Each figure below keeps its own base, period and geography, and none is added to a figure from another source.

| Series | Source and period | Base | Figure | Geography | Evidence |
|---|---|---|---|---|---|
| DSP take and DSP data | ANA study, September 2022 to January 2023, 21 advertisers | Dollar entering the DSP, sequential average | $80 platform; $80 data and additional, per $1,000 | Predominantly US | [26, 63] |
| DSP take and DSP data | ANA benchmark, November 2024 to March 2025, 23 advertisers | Each advertiser's spend, median | 7.9% platform; 1.0% data; 4.5% additional | Predominantly US | [63] |
| DSP take | UK Competition and Markets Authority (CMA), 2019 data | Value bought through the DSP, weighted average | 14% (range 5% to 42%), plus 2% third-party fees | UK | [288] |
| SSP take | ANA study, as above | Dollar entering the DSP, sequential average | $130 per $1,000 | Predominantly US | [26, 63] |
| SSP take | ANA benchmark | Each advertiser's spend, median, may be partly imputed | 13.7% (November 2024 to March 2025); 12.7% (Q4 2025) | Predominantly US | [63] [102, 100] |
| SSP take | CMA, 2019 data | Value sold through the SSP, weighted average | 16% (range 8% to 25%) | UK | [288] |
| SSP take | Liability opinion, United States v. Google, 17 April 2025 | Winning bid, open-web display, trial record | Google's exchange (AdX) about 20%; rivals often closer to 10% | US | [78] |
| Curation fee | Jounce, April 2025 report, via Playwire and AdExchanger | Curated deal IDs in a bid-stream sample | 50% none; 34% margin; 10% fixed CPM; 7% static price | Global | [9, 64, 281] |
| Data fee | Adloox page, 17 July 2026 | Impressions using the segment in a PubMatic auction package | $0.15 CPM | Not stated | [12] |

The rows cannot be stacked into one bill. The ANA's medians are taken across advertisers and its waterfall is a sequential average, and the report warns against comparing the two [63]. Each CMA rate takes the value passing through one intermediary as its base, in UK open display in 2019 [288].

### Where the advertiser's dollar went

In the ANA's 2023 study, $710 of every $1,000 entering a DSP reached the seller, and curation had no line of its own [26, 63] [287, 99, 26].

![In the ANA’s 2023 study, $710 of every $1,000 entering a DSP reached the seller; none of the fee lines is a curation fee.](figures/fig-ana-dollar.svg)

*Figure fig-ana-dollar. **In the ANA’s 2023 study, $710 of every $1,000 entering a DSP reached the seller; none of the fee lines is a curation fee.** The ANA’s Programmatic Media Supply Chain Transparency Study (complete report, 5 December 2023): of $1,000 entering the DSP, the sequential averages for DSP platform costs, DSP data and additional costs, SSP platform costs and seller revenue, across 21 advertisers and $123 million of spend from September 2022 to January 2023. Units: US dollars per $1,000. Scope: 21 ANA member advertisers, predominantly US, September 2022 to January 2023. Sources [26]. Caveat: A single study’s matched subset; later benchmark quarters report medians and cohorts on changing panels, so these components are not a trend (chapter 7). Data: data/figure-data.json#ana_dollar*

The ANA's Complete Report of 5 December 2023 covered 21 marketers, $123 million and 35.5 billion impressions from September 2022 to January 2023 [26, 63]. Transaction costs, "primarily DSP and SSP costs", took 29% of the dollar entering the DSP [26, 63]. The report's chart does not settle how the $80 splits between DSP data and additional costs, so this paper does not assign it [26, 63]. Agency fees and ad serving were outside the study, and SSP costs were computed only on impressions with SSP cost disclosure [26, 63].

The benchmark that followed reports medians across contributing advertisers [287, 102]. Its Q1 2025 report gives median SSP platform costs of 13.0%, 13.0% and 13.7% for the 2023 study, April to September 2024 and November 2024 to March 2025 [63]. Its 2023 column repeats the study's own averages [63]. On the sequential waterfall, transaction costs were 29.8% for April to September 2024, 26.1% for November 2024 to March 2025, 23.7% for Q4 2025 as restated in the Q1 2026 report, and 25.9% for Q1 2026 [63, 99]. Contributing panels ranged from 21 to 66 advertisers over these periods, so neither series is a trend [63, 99].

The SSP component can be partly imputed [287, 102]. Where SSP log-level data or identifiers are missing, the benchmark applies "estimated SSP fee rates" from "periodic SSP disclosures" and publishes no match rate [287, 102]. Its SSP cost line covers "investment in supply-side technology, inventory access, and curation services", so a curation fee collected through an SSP sits inside the SSP figure [287, 99, 26]. The 2023 study left out impressions without SSP cost disclosure, where the benchmark applies estimates, so the SSP line is not measured on one rule from 2023 to 2026 [26, 63] [287, 99, 26].

### The UK studies, the court and the filings

Outside the ANA and Jounce, the fee record is British, older or corporate [105, 288] [288] [212, 231, 284, 30, 285, 282]. ISBA (the UK advertisers' trade body) and PwC published their first study on 6 May 2020 [105, 288]. Across 15 advertisers and about £0.1 billion of UK spend, publishers received 51% of advertiser spend and 15% was an unknown delta [105, 288]. The second study, published 18 January 2023 on 1 September to 31 October 2022, used a basis without agency fees, verification and ad serving [283]. On that basis the 2020 results restate to a 57% publisher share and a 17% unknown delta [283]. In 2022 publishers received 65%, and the unknown delta fell to 3%, under 1% for PMP deals; the summary says the results do not represent the wider market [283]. The CMA put publishers at about 65% of advertiser spend in 2019 and found most SSPs keeping a higher share on open auctions than on private marketplaces, without figures [288].

The liability opinion of 17 April 2025 in United States v. Google found that AdX kept an overall take of around 20%, from 15% or lower to 25% or higher per impression, and that rival exchanges "often charged closer to 10%" [78]. The findings cover open-web display on the trial record, and the opinion is not a final judgment ([chapter 9](#competition)) [78].

Filings give bases without rates [212, 231, 284, 30, 285, 282]. In the Business and risk-factor sections read of seven FY2025 annual filings, none discloses a numeric take rate [212, 231, 284, 30, 285, 282]. Magnite defines its take rate as "our fee as a percentage of advertising spend" and says it tends to be lower for reserve auctions and on connected TV (CTV) [212]. PubMatic charges publishers "generally a percentage of the value" of impressions [231]. The Trade Desk's revenue was 21.6% of gross spend in 2025, this paper's ratio and an upper bound on its platform fee, since gross spend includes data and other services [284].

### What curators charge

The only curator-fee distribution in the sources screened is Jounce Media's, and it reaches the public second-hand [9, 64, 281]. Its April 2025 Supply Path Benchmarking report is client-only; Playwire (a publisher monetisation vendor, 10 April 2025) and AdExchanger (1 May 2025) report its fee types [9, 64, 281]. AdExchanger dates a Jounce analysis of 1.5 billion bid requests from 44 exchanges worldwide to March 2025 [9, 64, 281]. Neither account ties the fee distribution to that sample or states the fee sample's month [9, 64, 281] [9, 64]. The 14% median applies only within the 34% of deal IDs that carry a margin fee [9, 64, 281]. Playwire adds a median "effective margin" of 30% with no stated base, and it calls the unit both deal IDs and bid requests [9, 64, 281] [9, 64].

![In Jounce’s April 2025 report, half of curated deal IDs carried no curation fee and a third a percentage margin.](figures/fig-curation-fee-types.svg)

*Figure fig-curation-fee-types. **In Jounce’s April 2025 report, half of curated deal IDs carried no curation fee and a third a percentage margin.** Curated deal IDs by the fee the curator takes, from Jounce Media’s April 2025 Supply Path Benchmarking report as summarised by two secondary sources; neither account ties the distribution to the March 2025 request sample Jounce used elsewhere. A zero-fee deal still pays the exchange its standard take. Units: % of curated deal IDs. Scope: Jounce bid-stream data, global, as reported in April and May 2025. Sources [9]. Caveat: Second-hand: the two accounts describe the unit differently (deal IDs, bid requests), and only Playwire reports the median effective margin of 30%. A fee type is not an all-in cost (H2a). Data: data/figure-data.json#jounce_fees*

AdExchanger's account describes how the fees travel: curators and SSPs "factor their curation fees into the floor price passed to DSPs", and the SSP accrues and pays the curator's fee [64]. Jounce's public State of the Open Internet 2025 shows exchange fees embedded the same way, with illustrative take rates of 10%, 15% and 20% marking up a $0.60 floor [9, 64, 281] [115, 216, 290, 64]. For static-price deals, Jounce's Chris Kane said the split of the gap between declared and clearing price among exchange, data providers, curators and publisher is unknown [64]. AdExchanger framed curation fees as "the same old ad tech tax", and no SSP in the sources screened published a curated-deal fee schedule in reply [64, 288, 188, 117, 290].

Other margin figures have no comparable base. The Ozone Project, a UK publisher alliance, asserted "average curator margins at 44%" on 10 July 2025 with no sample, period or definition [289]. The author's earlier contextual paper gave median curation fees of 11 to 14%; no source for that figure was located, and this paper does not use it [2]. In a survey release of 17 June 2026, TripleLift, which sells curation, said 60% of buyers among 223 US and UK advertisers did not know what they pay for it [291]. The release names neither the fieldwork firm nor the dates [291].

### How the platforms document the stack

For six exchange products the type and base of the curation charges are on record, and none publishes a rate [212, 231, 281, 115, 24] [115, 24, 109, 117, 266] [226, 222, 231]. Four take the curator's charge out of the buyer's bid before the auction ranks it, as [chapter 4](#access) records [115, 111] [24, 59, 248] [109] [266].

| Product | Curator's charge: type and base | Platform's own charge | Buyer visibility | Amounts | Evidence |
|---|---|---|---|---|---|
| Microsoft Curate | No margin by default; a percentage of the buyer's bid net of buyer fees, or a fixed CPM, under an unstated maximum | Broker or Margin Auction Service Charge to the curator; seller charge on the buyer's media cost | Not stated | Worked examples only | [115, 216] [115, 216, 111] |
| Google Ad Manager Curation (Beta) | Percentage of media or CPM, set by the partner; "taken out of the bid price before it enters the auction" | Google's share unstated; Google pays the partner | Visible or hidden at the partner's choice on one page; "fully transparent to the buyer" on another | None | [24, 290] [24, 59, 248] |
| Index Marketplaces | Partner fee as a percentage of media spend or a fixed CPM; data-vendor fees as a CPM or percentage | Index revenue share, negotiated with each media owner | Partner chooses a disclosed or undisclosed model | Worked examples only | [109] [197] [109] |
| Maestro by Equativ | Margin added to the publisher's recommended bid, or deducted from the deal floor after fixed costs | Fixed costs and Equativ's own fee, amounts unstated | Not stated | Worked example only | [3] [114, 15] |
| Commerce Grid (Criteo) | Curator margin; bids submitted net of fees, charged only on a win | Criteo tech fee charged to the curator | Not stated | None | [266] [22] |
| PubMatic Connect | Curator sets its own fee | "Core SSP fee" plus a data fee "based on the curator's fee", negotiated partner by partner, per the CEO in trade press | Asserted for the authenticated console | None | [226, 222, 231] [222, 233] |

The worked examples show the arithmetic and stop there. Index splits a $10 gross bid, with a 5% partner fee and a 15% Index share, into $0.50 for the partner, $1.42 for Index and $8.08 for the media owner [109]. The 15% is illustrative, since Index negotiates its share with each media owner [109]. Equativ's help centre takes a EUR 0.59 deal floor and a 10% margin to a media price of EUR 0.531 [3]. Microsoft shows a 30% margin charge on a $0.18 margin as $0.06, where 30% of $0.18 is $0.054 [216].

Beyond the corporate fee basis in its 10-K, PubMatic's own charges on a curated deal are described, in the sources opened, only in its chief executive's interview with AdExchanger of 7 March 2025 [226, 222, 231]. The one priced input documented on PubMatic is a third-party data fee, for Adloox attention segments hosted by Qwarry at "$0.15/CPM" in PubMatic's auction-package flow, with payer and collector unstated [12] [12]. It is the only data-fee amount with a base found for a curated-deal input [12] [12]. Among the data curators, Audigent names its charge types without a base or amount: "PMP pricing is determined by bid floors and curation fees only" [192, 193, 190, 191].

### Every "no fee" covers one party

A "no fee" claim can mean a zero incremental charge, a charge inside another, a subsidy or an undisclosed arrangement [188, 187, 200] [117]. Each one found covers a single party's charge, and Commerce Grid's win-only rule covers only the curation charges [266].

| Claim | Whose charge | What stays in the path | Settled as | Evidence |
|---|---|---|---|---|
| Microsoft: curators take no margin by default | The curator's margin | Microsoft's charges to curator and seller; buyer fees | Zero curator margin by default; other charges contractual | [115, 216] |
| Index, 7 October 2025: "Index does not charge data fees" | Index's own data charge | Data-vendor fees, the partner fee and Index's revenue share | Zero Index data fee; the data vendor is still paid | [188] [197] |
| Index product page: "without any added platform fees" | An Index fee added on top | Index's revenue share, deducted from the bid | On this paper's reading, an included charge | [188, 187, 200] [109] |
| TripleLift TL Direct: "No TripleLift curation fees"; own segments "at no extra cost" | TripleLift's curation and segment charges | TripleLift's exchange take, unstated | Unsettled: zero, included or subsidised cannot be told from the page | [117] |
| Jounce: half of deal IDs with no curator fee | The curator's fee | The exchange's standard take, undisclosed per deal | Zero curator fee only | [9, 64] |
| Commerce Grid: "Fees only apply if curated deals win impressions" | All curation charges | Criteo's tech fee and the curator's margin, on a win | A win-only rule | [266] |

### The Deal Sync API carries a fee type and no amount

The one standard field for curation fees found declares a fee type and leaves the amount out [96]. IAB Tech Lab finalised the Deal Sync API 1.0 on 6 February 2026 [261, 249, 96]. Its Curation object carries `curfeetype`: 0 undisclosed, 1 percentage of spend, 2 flat fee, 3 CPM and 4 "no fee is paid for curation services" [96]. The guidance says the attribute gives "the type of fee being applied in the bidstream, but not what that fee is" [96]. A separate `auxdata` attribute flags fee-bearing data from outside the publisher without naming the provider or the charge [96]. A value of 4 can sit beside a data charge: the specification's own example is a data company that packages its data and takes no fee for curation itself [96]. No cohort B vendor page opened declares a fee type, and the searches found no curator or SSP confirming production support of the API by 3 November 2026 [96].

### Who discloses what across the admitted products

None of the 21 admitted products publishes an amount for the charges it controls, and 13 state no charge type in the documentation read [212, 231, 281, 115, 24] [115, 24, 109, 117, 266] [116, 12, 209, 128]. Each was coded from its pricing-transparency cell, on public documentation read by 3 November 2026.

![No admitted product publishes its charge amounts; 13 of 21 state no charge type in their public documentation.](figures/fig-fee-disclosure.svg)

*Figure fig-fee-disclosure. **No admitted product publishes its charge amounts; 13 of 21 state no charge type in their public documentation.** The 21 products admitted to the matrix, coded for H2b from their pricing-transparency cells: every charge with amount and base (none), charge types and their bases without amounts, a declared type only, a general assertion such as “no curation fees”, or nothing documented. Units: products. Scope: products admitted to the matrix, public documentation read by 3 November 2026. Sources [115], [24], [109], [3], [116], [117]. Caveat: Buyer-facing fee screens inside platforms were not observed, so H2b is coded unresolved on its registered rule (chapter 6). A “no fee” assertion can describe an included charge or a subsidy; it is coded as an assertion, not as a zero. Data: data/figure-data.json#h2b*

| Code | Products | Evidence |
|---|---|---|
| Every charge with amount and base (0) | None | [212, 231, 281, 115, 24] |
| Types and bases, no amounts (6) | Microsoft Curate, Google Ad Manager Curation, Index Marketplaces, Maestro by Equativ, Commerce Grid, PubMatic Connect | [115, 216, 111] [24, 290] [109] [3] [266] [226, 222, 231] |
| Declared type, no base (2) | SmartPMP (Audigent); WPP Media Premium Marketplace ("all-inclusive, commission-based or SaaS-based payment options") | [192, 193, 190, 191] [7] |
| General assertion only (7) | TL Direct, ClearLine, Seedtag NeuroX Curation, Adelaide High-Attention PMPs, Lotame Curated Marketplaces, dentsu Curate, Stagwell Curate | [117] [212, 62] [62] [116, 12, 209, 128] [209, 24, 188] [6] [5, 72, 71, 236] |
| Nothing documented (6) | FreeWheel Curation Hub, OpenXSelect, YMax, Comscore-Certified Deal IDs, IAS Curation, Multilocal curation | [16, 260] [17, 10] [11] [116, 12, 209, 128] |

Buyer-facing fee screens inside the platforms were not observed [212, 231, 281, 115, 24]. PubMatic asserts "Two-Way Fee Transparency" for its authenticated console, and Google and Index let the curator or partner choose whether the buyer sees the fee [222, 233] [24, 59, 248] [109]. PubMatic's code rests on its chief executive's account and its 10-K; its Connect for Publishers page returned an error to the fetch [226, 222, 231] [20].

> **Note on interests.**
> Lotame, a former employer of the author, is coded here as a general assertion only, and Index's data-vendor launch list includes Samba TV and Lotame [209, 24, 188] [188]. The author has worked at Samba (TV measurement and ad data; owner of Semasio) since May 2025 [36]. He was executive vice president at Verve Group from September 2022 to August 2024 and earlier worked at Lotame and Hearts & Science. He co-leads the Signals & Measurement working group of AgenticAdvertising.org (AdCP, the Ad Context Protocol) and maintains an open-source AdCP signals adaptor, and No Fluff Advisory operates two agents listed in the AdCP registry. None of these earns any evidence weight. Without the Lotame row, 12 of 20 products state no charge type, and the H2b code is unchanged [209, 24, 188].

### A price gap on different impressions

Deal-ID impressions cost the ANA's advertisers more per thousand than open-marketplace (OMP) impressions in every period in the table, on each measure reported [63]. None of these figures compares the same opportunities [66].

| Period | Contributing advertisers | Median CPM, PMP and OMP | Aggregate CPM, PMP and OMP | Evidence |
|---|---|---|---|---|
| 2023 study, as tabulated in the Q1 2025 report | 21 | $6.68 and $2.23 | $5.83 and $2.75 | [63] |
| November 2024 to March 2025 | 23 | $9.37 and $3.28 | $10.43 and $3.03 | [63] |
| April to June 2025 | 21 | $7.15 and $4.41 (the release says "averaging"; the median label is unconfirmed) | $7.43 and $3.64 | [66, 98, 99] |
| October to December 2025 | 35 | Not recorded | $10.72 and $4.25 | [100, 309] |

Three things separate this gap from the cost H2a asks about. First, the ANA's PMP is any deal ID, and it carries CTV, which ran 99.5% through deal IDs in November 2024 to March 2025 [63]. CTV prices sit higher: in April to September 2024 the average CTV CPM was $15.00 through PMP deals against $5.54 in the open marketplace [102, 286, 63]. Second, no figure holds inventory, period and buyer constant, so each route's CPM prices whatever impressions it bought [66]. Third, a CPM bundles the media price paid to the publisher with intermediary fees, and the series does not say which drives the gap [26]. For April to September 2024 the benchmark found PMP DSP platform costs "1.1 percent" below OMP costs, and the CMA found SSPs keeping a smaller share on private marketplaces [102, 100] [288].

The closest cut is still cross-sectional. In the 2023 study, deals spanning more than 500 domains cost $5.52 against $2.75 in the open marketplace and delivered 19.8% made-for-advertising (MFA) impressions against 26.8% [26]. The report calls this "quality similar to the OMP deals, but at double the cost" [26]. The domain-count profile is not this paper's multi-publisher definition, and the CPMs average 21 advertisers' impressions without matching them on inventory, period or buyer [26]. In Q1 2026 the ANA ranked 66 advertisers by how much of each DSP dollar reached consumers [99]. The better half paid a $5.45 CPM against $7.40, with PMP shares of 61.1% and 62.2%, and the halves are defined by that outcome [99]. Jounce's Kane told AdExchanger that curation "does not appear to be inflating supply chain costs", a consultant's assertion without a matched comparison [64].

### Coding H2a, H2b and Q3

**H2a** holds that, for comparable eligible opportunities, a curated deal's all-in net cost exceeds the open auction's. A documented matched comparison showing a higher cost supports it, one showing an equal or lower cost contradicts it, and the lack of any leaves it unresolved. No source screened by 3 November 2026 compares the two routes on the same inventory, period and buyer with every charge recorded [26] [99] [115, 216, 290, 64]. The ANA's CPM gaps price different impressions, its cohorts are defined by outcome, and the fee documentation carries types and worked examples [66] [212, 231, 281, 115, 24]. Under the rule, a zero curator surcharge or a corporate take rate would not settle all-in cost. H2a is coded unresolved, the registered default; the absence says nothing about the direction of the difference.

**H2b** holds that every charge in each scored product's deal path is disclosed with its basis. A majority disclosing every applicable charge type with its basis supports it, and a majority disclosing none contradicts it. It is unresolved when access to buyer-facing fee screens cannot be established for most products, so that most codes are access unknown. The test is on charge types and bases, and amounts belong to Q3. Six products document the type and basis of each charge they control, and none documents every charge in the deal path, the DSP's take among them, so the supported rule is not met [212, 231, 281, 115, 24] [115, 24, 109, 117, 266] [115, 216, 290, 64]. Thirteen products state no charge type in the documentation read: seven make only a general assertion and six document nothing on the pages opened [116, 12, 209, 128] [16, 260] [17, 10]. The matrix scores the six as no evidence, and no buyer-facing fee screen was observed for any product, the registered unresolved condition [212, 231, 281, 115, 24] [116, 12, 209, 128]. Absence of accessible evidence cannot contradict, so at most the seven assertions count toward the contradicted rule, short of a majority of 21. H2b is coded unresolved, and the code is the same without the interest-marked Lotame row [209, 24, 188]. The record shows charge types and bases documented for six products and amounts for none, not that any charge is hidden.

**Q3** counts a charge as disclosed when its amount or rate and its base are public, and codes type-only declarations as type only. On that rule no curated deal or product in the sources screened discloses its curation fee; Adloox's $0.15 CPM is the one disclosed data-fee amount, with payer and collector unstated [212, 231, 281, 115, 24] [12]. SSP take appears as corporate bases, court findings for open-web display and benchmark medians that fold curation into SSP costs [212, 231, 284, 30, 285, 282] [78] [287, 99, 26]. The DSP's take on a curated deal is outside every sell-side document read, and the fullest reconciliation within one transaction perimeter is Index's worked example on illustrative rates [115, 216, 290, 64] [109]. No public source screened gives all four charges, with payer, collector, base and period, for one named curated deal [115, 216, 290, 64].

In plain words: the public record screened shows which fees exist and in what order they leave the bid. It does not show what any buyer paid, all in, for a named curated deal, or whether that was more than the open auction would have cost.



## 7. Does curation deliver the quality it sells


*Curated deals are sold on less made-for-advertising inventory, higher viewability, cleaner traffic and more attention. In the sources screened by 3 November 2026, we found no independent matched evaluation of a named curated deal against its own promise. Every quantified product claim comes from a vendor, its agency or a partner, so H3 is unresolved.*

> **In short.**
> The Association of National Advertisers (ANA) measured US advertisers' made-for-advertising (MFA) exposure by deal type in 2023 [26]. MFA sites took 1.3% of impressions in private-marketplace (PMP) deals spanning 11 to 500 domains, 19.8% in deals spanning more than 500 and 26.8% in the open marketplace [26]. On an unstated base, the ANA benchmark put Q4 2025 MFA rates at 0.98% in the open marketplace and 0.23% in PMPs [100]. It warned that no buying path is immune without continuous auditing [100]. Adalytics documented MFA, mis-declared and declared-bot inventory sold through apparent PMP deal IDs, as examples without rates [67] [68] [69]. Every quantified claim found for a curated product comes from a vendor, its agency or a partner, and none names an independent measurer with a published method [119, 120, 121, 125] [60, 209, 129, 70]. No vendor source opened reports an MFA share, invalid-traffic rate or suitability rate for a named curated deal [119, 121, 124, 58, 129]. In the sources screened, no independent measurement tests a named deal's promise, so H3 is unresolved and no claim meets the registered margin on independent evidence [310, 101, 311].

### What H3 tests

H3 is tested one promise at a time, and each promise is one of two kinds. A threshold promise sets an absolute level, such as an MFA share or a viewability rate. An improvement promise beats a comparator, which must be comparable eligible inventory before curation over the same period. The margin was registered before extraction: for a rate, the smaller of 10 percentage points and the vendor's claimed uplift; 2 points for invalid traffic (IVT); 10% relative for attention and carbon. Only MFA, viewability, IVT, suitability, attention and carbon count, and no metric stands in for another. Only independent measurement with a published method can support or contradict a promise; vendor and commissioned evidence leave it unresolved.

### The ANA record: deal profiles and quarterly buckets

The strongest independent evidence is the ANA's, and it compares groups of deals with the open market of the same advertisers [100]. TAG TrustNet reconciled its 2023 study from impression log-level data, matching DSP (demand-side platform) records with verification data [26] [26]. It covered 21 marketers, $123 million of spend and 35.5 billion impressions from September 2022 to January 2023, with no CTV (connected TV) [26]. It sorted PMP deals by domain count into publisher-specific deals, seller-specific auction packages and SSP (supply-side platform) direct deals curated by the SSP or an agency [26]. Intermediary deal IDs, it said, bundle many publishers, rarely carry auction priority and sell inventory curation [26]. CPMs (costs per thousand impressions) in the table are averages across matched data [26].

| Deal profile (domains per PMP) | CPM | MFA share of impressions | IVT share of impressions |
|---|---|---|---|
| 1 to 10, publisher-specific | $6.03 | 0.02% | 0.2% [26] |
| 11 to 500, auction packages | $8.49 | 1.3% | 0.2% [26] |
| Over 500, SSP direct | $5.52 | 19.8% | 1.4% [26] |
| Open marketplace | $2.75 | 26.8% | 1.1% [26] |

The verification pass re-checked only the CPM and MFA columns [26].

The two larger bands split against the 10-point margin. On this paper's arithmetic, the 11-to-500 band ran 25.5 points below the open market on MFA and the over-500 band 7.0 points below [26]. The over-500 band's IVT share was 0.3 points above the open market's [26]. The report said the largest deals delivered "quality similar to the OMP deals, but at double the cost", OMP being its open marketplace [26]. For some participants, MFA passed 30% of activity inside PMP deals [26]. None of this tests a product: the bands count domains without a selection rule, the comparator is other inventory, and no uncertainty is published [26]. The study also found that MFA sites pass standard viewability, IVT and brand-safety screens, so neither a viewability nor an IVT figure speaks to MFA [26].

The benchmark that followed dropped the domain-band split [26]. It names only two marketplace types, open and private, and its glossary fixes neither publisher multiplicity nor the treatment of programmatic guaranteed [310, 287]. For October to December 2025 (35 contributing advertisers, $142 million) it reported MFA rates of 0.98% in the open marketplace and 0.23% in PMPs, without saying whether these are medians [100, 309] [100]. The report credited the PMPs' curated structure, said no buying path is immune without continuous auditing, and said actively managed open-market buying can match PMPs on TrueCPM efficiency [100]. Exposure was concentrated: the median was 0.45%, and the top quartile of advertisers ran from 5.5% to 31.4% [100]. For April to June 2025, Fiducia, the benchmark's technology partner, reported 29% of MFA impressions inside PMPs [65].

The MFA series shifts between editions. The Q2 2025 release said median MFA spend fell from 2.3% to 0.8% between the first and second quarters [66, 65]. The Q1 2026 report said MFA sat between 0.4% and 0.6% throughout 2025 before rising to 1.1% [103, 99]. Meanwhile the contributor panel went from 23 advertisers to 21, 35 and then 66, with no matched panel published [63] [98, 100, 66] [103, 99]. The 2023 figures, 21% of impressions and 15% of spend on a web-domain subset, sit on another base [26]. No trend can be read across these editions.

### UK traceability, MFA supply and the academic record

The second study by the UK advertiser body ISBA and PwC covered 11 advertisers, with fieldwork in September and October 2022 [106, 283]. PMP deals were about one-fifth of matchable impressions, matched buy side to sell side at above 70% and left an unknown delta under 1% [106, 283]. These figures trace money, and ISBA said they should not be taken as representative [106, 283]. PwC recommended investing more in well-curated PMPs, while placing their lower fraud, viewability, brand-safety and data-leakage risks outside the study [106].

Jounce Media is a consultancy with clients on both sides of the market [101, 64]. It reported that 18 of the 20 largest MFA publishers in the global bid stream shrank by more than 95% within 18 months of the ANA's 2023 report [101]. One MFA operator launches a new site every three to four weeks to stay ahead of block lists [101]. Inference: an MFA exclusion is only as current as its list, so a promise to exclude MFA must be tested against a dated list.

The academic record screened is one 2023 doctoral thesis from Erasmus University Rotterdam [118]. Its abstract argues that premium contracts, which it calls PMPs, lead to adverse selection in open real-time bidding (RTB) without increasing advertisers' utility; the full text returned an access error [118]. The search behind it was one query, so the absence of other academic studies is bounded to that query [310, 101, 311].

### Adalytics: deal IDs on MFA, mis-declared and bot inventory

Adalytics' reports are the only third-party observations of specific PMP deal IDs in the sources screened, and they show failures as examples without rates [68] [69]. Adalytics sells analytics to advertisers and states no funding for these reports, so its independence is unknown [68].

In 2024 it reported that the www3.forbes.com subdomain, which DeepSee.io classed as MFA, was relayed as the main www domain by Microsoft Xandr, TripleLift, PubMatic and Magnite [67]. Some of this inventory transacted through apparent PMP deal IDs [67]. The report concludes that a PMP deal ID can still leave a campaign exposed to MFA, and allows for false positives [67]. A Forbes response in trade coverage was not opened, so it is not assessed [67].

In January 2024 Adalytics observed ads from 16 of the 17 disclosed participants in the ANA's 2023 study on 22 MFA-classified sites [68]. It also observed apparent PMP deals on such sites through Yieldmo, PubMatic, TripleLift, OpenX and Sharethrough [68]. The report quotes the exchanges' own statements: Sharethrough's covers removing MFA sites from its off-the-shelf deals and custom PMPs created after 1 July 2023 [68]. OpenX's covers blocking known MFA traffic on its auction packages against Jounce's list, and PubMatic's covers eliminating MFA from the auction packages it sells [68]. Each is a threshold promise of the kind H3 tests. The report disclaims any quantitative audit and names no curated deal covered by those statements [68].

A third report documents ads for US brands served to declared data-centre bots through PMP deal IDs, March 2022 to November 2024 [69]. The exchanges named include Index Exchange, Magnite, TripleLift, Yieldmo, TrustX, GumGum and Kargo [69]. Verification found more on PubMatic, Microsoft Xandr and Media.net [69]. Most of them say they filter bots pre-bid with HUMAN Security, and the report cites HUMAN's 2019 claim that Kargo's curated marketplace ran below 0.3% IVT [69]. The authors note that declared crawlers are not necessarily fraudulent [69]. Examples cannot test a 0.3% ceiling, which needs a rate. In the same data the Integral Ad Science (IAS) pixel labelled declared data-centre bots as human about 17% of the time, so an IVT measurer needs checking too [69]. IAS, DoubleVerify and HUMAN hold MRC (Media Rating Council) accreditations for IVT detection, and no vendor rebuttal appears on the pages [69].

The 40 study records coded for this paper, independent, commissioned and vendor, are listed below with design, sample, sponsor and reach.

**Study evidence table**

| Study | Year | Design | Sample | Setting | Format / channel | Outcome measure | Headline result | Uncertainty | Sponsor | Generalisability | Src |
|---|---|---|---|---|---|---|---|---|---|---|---|
| ANA Programmatic Media Supply Chain Transparency Study | 2023 | Log-level data reconciliation (DSP, SSP, verification) across advertisers; Kroll interviews; DeepSee.io MFA overlay | 21 advertisers; $123M; 35.5bn impressions | US open-web programmatic (sites and apps; no CTV) | display and video | cost waterfall; OMP/PMP spend split; MFA share; CPMs by marketplace | 36 cents of each DSP dollar reaches the consumer; 59% OMP / 41% PMP; MFA 21% of impressions and 15% of spend with CPMs 25% lower | none published; averages across advertisers | ANA (advertiser trade body); TAG TrustNet/Fiducia technology partner | self-selected ANA members; MFA on a web-domain subset | [26] |
| ANA Programmatic Transparency Benchmark (quarterly series) | 2024-2026 | Ongoing log-level reconciliation of contributing advertisers; medians and quartiles with one value per marketer; spend-weighted aggregates in cost waterfall; estimated SSP fee rates where SSP logs are missing | 21 to 66 contributors per period; $142M to $242M per period | US open-web programmatic incl. CTV from Apr-Sep 2024 | web, mobile app, CTV | PMP share (median and aggregate); CPMs by marketplace; TrueAdSpend; MFA; fees | median PMP 56.2% (Apr-Sep 2024) to 92.2% (Q4 2025); aggregate 65.9% to 79.2%; aggregate fell Q2-Q4 2025 while median rose | no confidence intervals; panel composition changes each period | ANA and TAG TrustNet; Fiducia | contributing ANA members; binary OMP/PMP taxonomy cannot isolate multi-publisher curation | [99] |
| ISBA/AOP/PwC Programmatic Supply Chain Transparency Study | 2020 | End-to-end supply-chain mapping and DSP-to-SSP impression matching from log-level data | 15 advertisers, 8 agencies, 5 DSPs, 6 SSPs, 12 publishers; ~£0.1bn; 290 of 1,000+ chains matched (31m impressions) | UK premium programmatic | display | publisher share of spend; unknown delta; match rate | publishers 51%; unknown delta 15%; match rate 12% | none published | ISBA members (advertiser-funded) with AOP; PwC | premium UK participants only | [105] |
| ISBA/PwC Programmatic Supply Chain Transparency Study II | 2023 | Test of the Taskforce audit toolkit; DSP-to-SSP impression matching from log-level data | 11 advertisers, 7 agencies, 6 DSPs, 6 SSPs, 10 publishers; 1.3bn impressions analysed, 104m matchable, 61m matched | UK premium programmatic | display | match rate; unknown delta; publisher share; PMP match rate | match 58%; delta 3% (PMP under 1%); publishers 65%; PMP deals about one-fifth of matchable impressions, match above 70% | none published; 'should not be taken as representative' | ISBA; PwC under agreement of 25 April 2022 | premium UK participants only | [106] |
| Jounce Media, The State Of The Open Internet 2025 | 2025 | Bid-stream analysis of available RTB auctions by control type; publisher and exchange case studies | 1.5bn bid requests from 44 exchanges in March 2025 (per AdExchanger) | global RTB web, mobile app, CTV | all | mix of available RTB auctions; MFA publisher volumes; duplication | curation 72.7%, open auction 18.6%, PMP 8.7% of auctions | none; unit is auctions not spend | Jounce Media (consultancy; free publication) | Jounce's sample of exchanges; global | [101] |
| Jounce Media, April 2025 Supply Path Benchmarking Report (via Playwire and AdExchanger) | 2025 | Bid-stream analysis of curated deal floors versus open-auction floors to infer fee type | same March 2025 sample | global RTB | all | distribution of curation fee types and medians | 50% zero fee; 34% margin-based (median 14%); 10% fixed CPM (median $0.48); 7% static price; median effective margin 30% | none; two secondary accounts differ on unit wording | Jounce Media (client-only report) | sample of deal IDs; collector of fees unknown | [9] |
| Adalytics, Forbes www3 subdomain report | 2024 | Observational: Similarweb, screenshots, archives, Prebid debug logs; one 52-slide session | 201+ impressions in one session; hundreds of brands; four exchanges named | US open web (one publisher) | display | domain mis-declaration; transaction via PMP deal IDs; absence of verification tags | mis-declared inventory observed transacting via apparent PMP deal IDs on Xandr, TripleLift, PubMatic, Magnite | examples only; 'could be false positives'; intent not asserted | unknown (not stated); Adalytics is a commercial vendor | one publisher; not a rate | [67] |
| Adalytics, MFA sites January 2024 report | 2024 | Observational: panel browser extension, manual inspection, crawlers, archives on 22 MFA-classified sites | 22 sites; brand counts per vendor (e.g. Magnite 11,671; Amazon DSP 2,601) | US open web | display | brands and vendors observed on MFA sites; PMP deals carrying MFA | 16 of 17 disclosed ANA participants observed; PMP deals on MFA sites via Yieldmo, PubMatic, TripleLift, OpenX, ShareThrough | 'not a comprehensive or quantitative audit' | unknown (not stated); Adalytics is a commercial vendor | 22-site sample chosen for MFA criteria; not a rate | [68] |
| Adalytics, pre-bid bot detection and filtration report | 2024-2025 | Observational: declared bot traffic from HTTP Archive, URLScan.io and an anonymous crawler vendor matched to ad source code | over two million websites, more than seven years of crawl data; dozens of brand examples | US brands; data-centre crawlers worldwide | display and video | ads served to declared bots incl. via PMP deal IDs; vendor filtration claims | PMP deal IDs on Index Exchange, Magnite, TripleLift, Yieldmo, TrustX, GumGum, Kargo served declared bots; IAS pixel labelled declared data-centre bots human about 17% of the time | examples; 'not all bot traffic is fraudulent'; caveats section not read | unknown (not stated); Adalytics is a commercial vendor | declared crawler traffic; not an IVT rate for any deal | [69] |
| Balocco, Asymmetric information in programmatic advertising (doctoral thesis) | 2023 | Theoretical models with empirical support (chapter 3); mechanism-choice model (chapter 4); waterfall vs header bidding fee model (chapter 5) | not readable (abstract only) | programmatic display (exchanges, publishers, advertisers) | display | adverse selection from PMPs; exchange fee incentives | PMPs cause adverse selection in RTB without raising advertiser utility; header bidding removes exchanges' incentive to keep sell-side fees low | not readable | academic (Erasmus University Rotterdam); no sponsor stated | model-based; data unknown | [118] |
| PubMatic and Agence 79, France: CTR | 2026 | single campaign, vendor case study | not stated | France; Agence 79 (Havas Media Network) for an unnamed French mass-market retailer | PubMatic Connect (curation) | CTR | CTR five times higher | not stated | PubMatic (measured by Agence 79 internal data) | comparator: the same campaign on the open market; method published: no; evidence class: vendor_assertion | [119] |
| PubMatic and Agence 79, France: video completion rate | 2026 | single campaign, vendor case study | not stated | France; Agence 79 for an unnamed retailer | PubMatic Connect (curation) | video completion rate | +16.7% versus target | not stated | PubMatic (measured by Agence 79 internal data) | comparator: the campaign's set target; method published: no; evidence class: vendor_assertion | [119] |
| PubMatic and Agence 79, France: viewability | 2026 | single campaign, vendor case study | not stated | France; Agence 79 for an unnamed retailer | PubMatic Connect (curation) | viewability | +12.5% versus an 80% minimum viewability target (relative or points not stated) | not stated | PubMatic (measured by Agence 79 internal data) | comparator: the agency's 80% target; method published: no; evidence class: vendor_assertion | [119] |
| OpenX and Chalice AI for Hyundai: cost per high-value action | 2026 | single campaign, 'Early results' | not stated | not stated; Hyundai | OpenXBuild custom decisioning (listed; curated video activation) | cost per high-value action | 20% reduction | not stated | OpenX and Chalice AI (measured by not stated) | comparator: the next-best alternative / second-best-on-plan; method published: no; evidence class: vendor_assertion | [120] |
| OpenX and Chalice AI for Hyundai: CPM | 2026 | single campaign, 'Early results' | not stated | not stated; Hyundai | OpenXBuild custom decisioning (listed; curated video activation) | CPM | 67% decrease | not stated | OpenX and Chalice AI (measured by not stated) | comparator: the next-best alternative / second-best-on-plan; method published: no; evidence class: vendor_assertion | [120] |
| Microsoft Curate 'by the numbers': CPA | 2023 | marketing aggregate from first-party data | not stated | not stated; not stated | Microsoft Curate | CPA | 74% lower CPA | not stated | Microsoft Advertising (measured by Microsoft (first-party data)) | comparator: not stated; method published: no; evidence class: vendor_assertion | [121] |
| Microsoft Curate 'by the numbers': CPM | 2023 | marketing aggregate from first-party data | not stated | not stated; not stated | Microsoft Curate | CPM | 25% lower CPM | not stated | Microsoft Advertising (measured by Microsoft (first-party data)) | comparator: not stated; method published: no; evidence class: vendor_assertion | [121] |
| Microsoft Curate 'by the numbers': viewability | 2023 | marketing aggregate from first-party data | not stated | not stated; not stated | Microsoft Curate | viewability | +12% viewability rate (rate, relative or points not stated) | not stated | Microsoft Advertising (measured by Microsoft (first-party data)) | comparator: not stated; method published: no; evidence class: vendor_assertion | [121] |
| Audigent SmartPMPs outperform DMP data and open exchange: CPC | 2026 | 'rigorous testing', vendor case page | not stated | not stated; not stated | SmartPMP | CPC | 12% to 67% lower CPC | not stated | Audigent (a part of Experian) (measured by not stated) | comparator: traditional DSP-level data to open exchange inventory; method published: no; evidence class: vendor_assertion | [122] |
| Audigent SmartPMPs outperform DMP data and open exchange: CTR | 2026 | 'rigorous testing', vendor case page | not stated | not stated; not stated | SmartPMP | CTR | 1.6x to 2.5x higher CTR | not stated | Audigent (a part of Experian) (measured by not stated) | comparator: traditional DSP-level data to open exchange inventory; method published: no; evidence class: vendor_assertion | [122] |
| Audigent SmartPMPs outperform DMP data and open exchange: video completion rate | 2026 | 'rigorous testing', vendor case page | not stated | not stated; not stated | SmartPMP | video completion rate | 12.5% to 42% higher VCR | not stated | Audigent (a part of Experian) (measured by not stated) | comparator: traditional DSP-level data to open exchange inventory; method published: no; evidence class: vendor_assertion | [122] |
| Audigent for Alcon: target KPI (sign-ups / cost per online eye exam) | 2026 | single advertiser, before-after, vendor case page | not stated | not stated; Alcon | SmartPMP (curated PMPs) | target KPI (sign-ups / cost per online eye exam) | exceeded target KPI by 27%; cost per online eye exam down 87% from the start of optimisations | not stated | Audigent (a part of Experian) (measured by Alcon floodlight tag (conversions); reported by Audigent) | comparator: Alcon's target KPI; the campaign's starting point; method published: no; evidence class: vendor_assertion | [123] |
| Lotame Curation launch, early trials: campaign efficiencies | 2024 | 'early trials', vendor release | not stated | not stated (product in North America, EMEA, APAC); not stated | Lotame Curated Marketplaces (launched as Lotame Curation) | campaign efficiencies | 20% better campaign efficiencies | not stated | Lotame (interest mark) (measured by not stated) | comparator: not stated; method published: no; evidence class: vendor_assertion | [124] |
| Lotame Curation launch, early trials: video completion rate | 2024 | 'early trials', vendor release | not stated | not stated; not stated | Lotame Curated Marketplaces (launched as Lotame Curation) | video completion rate | improved video completion rates by 15% | not stated | Lotame (interest mark) (measured by not stated) | comparator: not stated; method published: no; evidence class: vendor_assertion | [124] |
| Lotame Curation launch, early trials: viewability | 2024 | 'early trials', vendor release | not stated | not stated; not stated | Lotame Curated Marketplaces (launched as Lotame Curation) | viewability | 21% higher viewability | not stated | Lotame (interest mark) (measured by not stated) | comparator: not stated; method published: no; evidence class: vendor_assertion | [124] |
| Lotame and PHD EMEA, global CPG drinks brand: view-through rate | 2025 | single brand, multi-market, agency allow-lists plus supply-led optimisation across unnamed SSPs | not stated | UK, Spain, Singapore; PHD EMEA (Omnicom) for an unnamed global CPG drinks brand | Lotame Curated Marketplaces | view-through rate | consistently above 70% | not stated | Lotame (interest mark) (measured by not stated (Lotame Analytics referenced)) | comparator: industry benchmarks of 60% (unsourced); method published: no; evidence class: vendor_assertion | [125] |
| Lotame and PHD EMEA, global CPG drinks brand: cost savings | 2025 | single brand, multi-market | not stated | UK, Spain, Singapore; PHD EMEA for an unnamed global CPG drinks brand | Lotame Curated Marketplaces | cost savings | projected cost savings of up to 33% | not stated | Lotame (interest mark) (measured by not stated) | comparator: not stated; method published: no; evidence class: vendor_assertion | [125] |
| Adelaide High Attention Markets launch: Colgate on PubMatic: 'media quality' (AU implied) | 2023 | first campaigns, vendor release | not stated | not stated; Colgate | Adelaide High-Attention PMPs (PubMatic auction packages) | 'media quality' (AU implied) | 40% increase in media quality | not stated | Adelaide (measured by not stated) | comparator: standard media; method published: no; evidence class: vendor_assertion | [126] |
| Adelaide High Attention Markets launch: aggregate: performance / funnel lift | 2023 | vendor aggregate and a meta-analysis of nearly 40 case studies | nearly 40 case studies; advertiser count not stated | global ('20+ countries'); 'dozens of advertisers' | Adelaide High-Attention PMPs | performance / funnel lift | 20-80% better performance than standard viewable media; 31% upper-funnel and 56% lower-funnel lift | not stated | Adelaide (measured by Adelaide) | comparator: standard viewable media; traditional metrics; method published: no; evidence class: vendor_assertion | [126] |
| Teads and Adelaide high-attention omnichannel PMPs: attention score (AU) | 2023 | PMP composition claim, vendor release | AU base: 9.3 billion Teads impressions, 65 campaigns, 22 formats, 20,000+ domains | not stated; not stated | Adelaide High-Attention PMPs (Teads) | attention score (AU) | average attention score 31% higher | not stated | Teads and Adelaide (measured by Adelaide) | comparator: Adelaide's Teads AU benchmarks; method published: no; evidence class: vendor_assertion | [127] |
| Teads and Adelaide high-attention omnichannel PMPs: case lifts: brand and conversion outcomes | 2023 | per-advertiser outcome lifts, vendor release | not stated per case | not stated; five unnamed advertisers (hospitality, dairy, clothing retail, streaming, fast food) | Adelaide High-Attention PMPs (Teads) | brand and conversion outcomes | 12.5% higher unaided awareness; 32.7% increase in familiarity; 41.6% higher store visitation; 455.6% higher site visits; 110.8% higher checkout actions; 8x higher in-store visits | not stated | Teads and Adelaide (measured by Adelaide and unnamed third-party brand measurement partners) | comparator: not stated (lower-AU placements implied); method published: no; evidence class: vendor_assertion | [127] |
| Yieldmo YMax vs IAB contextual: CTR and CPC | 2026 | 'head-to-head testing on identical inventory' | not stated | not stated; not stated | YMax | CTR and CPC | +81% CTR and -65% CPC | not stated | Yieldmo (measured by not stated) | comparator: IAB contextual targeting; method published: no; evidence class: vendor_assertion | [11] |
| Yieldmo Companion Audiences vs ID-based: CTR and incremental reach | 2026 | 'In testing' | not stated | not stated; not stated | YMax | CTR and incremental reach | +33% CTR and +91% incremental reach | not stated | Yieldmo (measured by not stated) | comparator: ID-based targeting; method published: no; evidence class: vendor_assertion | [11] |
| Seedtag NeuroX Curation: neural engagement (vendor metric) | 2026 | vendor page figure | not stated | not stated; not stated | Seedtag NeuroX Curation | neural engagement (vendor metric) | 3.5x higher neural engagement | not stated | Seedtag (measured by not stated) | comparator: non-contextual ads; method published: no; evidence class: vendor_assertion | [128] |
| Multilocal homepage ranges: performance (undefined) | 2026 | marketing tiles, no design | not stated | not stated; not stated | Multilocal curation | performance (undefined) | 20-40% increase in performance; 30-50% boost; 24-28% boost in yield | not stated | Multilocal (measured by not stated) | comparator: not stated; method published: no; evidence class: vendor_assertion | [129] |
| Multilocal homepage ranges: CTR | 2026 | marketing tiles, no design | not stated | not stated; not stated | Multilocal curation | CTR | 75% higher CTR than projected | not stated | Multilocal (measured by not stated) | comparator: projected CTR; method published: no; evidence class: vendor_assertion | [129] |
| Multilocal homepage ranges: video completion rate | 2026 | marketing tiles, no design | not stated | not stated; not stated | Multilocal curation | video completion rate | 38% increase in VCR | not stated | Multilocal (measured by not stated) | comparator: not stated; method published: no; evidence class: vendor_assertion | [129] |
| Multilocal homepage ranges: carbon footprint | 2026 | marketing tiles, no methodology named | not stated | not stated; not stated | Multilocal curation | carbon footprint | cut by 40-60% | not stated | Multilocal (measured by not stated) | comparator: traditional programmatic ads; method published: no; evidence class: vendor_assertion | [129] |
| Equativ Maestro operational claim: planning time | 2026 | vendor page | not stated | not stated; not stated | Maestro by Equativ | planning time | up to 40% reduction in planning time | not stated | Equativ (measured by not stated) | comparator: not stated; method published: no; evidence class: vendor_assertion | [15] |
| TripleLift TL Direct GA release: Forrester TEI: ROAS, labour cost, incremental profit, hours | 2026 | Forrester Total Economic Impact financial model; design not described in the release | not stated | US (release dateline); composite (TEI) | TL Direct | ROAS, labour cost, incremental profit, hours | $3.70 ROAS; $357,000 avoided labour costs over three years; $1.3 million incremental profit over three years; 40 hours saved per campaign | not stated | TripleLift (Forrester TEI commissioned by TripleLift) (measured by Forrester (commissioned)) | comparator: not stated; method published: not in the release (TEI document not opened); evidence class: commissioned_measurement | [130] |

### What vendors claim

Every quantified claim found for a curated product comes from a vendor, its agency or a partner, without an independent measurer or a published method [119, 120, 121, 125] [60, 209, 129, 70].

> **Note on interests.**
> The author has worked at Samba, which owns Semasio, since May 2025. He was EVP at Verve Group from September 2022 to August 2024 and earlier worked at Lotame and Hearts & Science. He co-leads the Signals & Measurement working group of AgenticAdvertising.org, which publishes AdCP (the Ad Context Protocol), and maintains an open-source AdCP signals adaptor. No Fluff Advisory operates two agents listed in the AdCP registry. Lotame's claims carry the interest mark [124]; Samba documents no deal product with a quality claim, and Semasio's deal page returned an error [57, 268] [269, 270]. None of these interests earns any evidence weight ([chapter 13](#method)).

| Product | Claim as published | Kind | Comparator | Source of the figure | Registered dimension |
|---|---|---|---|---|---|
| PubMatic curation, Agence 79 (France, 2026) | CTR (click-through rate) five times higher; VCR (video completion rate) +16.7% and viewability +12.5% against targets | Improvement; threshold | Same campaign on the open market; an 80% viewability target | Agence 79 internal data | Viewability [119, 292] |
| PubMatic curation (2025 release; 2022 Goodway Group case) | Effective CPMs at least 25% above non-curated deals; CPMs 37% better than buy-side data; viewability 24% over a benchmark | Improvement | Non-curated deals; buy-side data; an undefined benchmark | PubMatic; Goodway internal data | Viewability (Goodway) [119, 230, 246] |
| OpenXBuild with Chalice AI, Hyundai (2026) | Cost per high-value action -20%; CPMs -67% | Improvement | Next-best alternative on the plan | OpenX and Chalice AI | None [120] |
| Microsoft Curate | CPA (cost per acquisition) -74%; CPM -25%; viewability +12%; average rate 10% to 15% above the buyer's threshold | Improvement; threshold | None; the buyer's threshold | 2023 Microsoft first-party data; Microsoft documentation (threshold) | Viewability, base unstated [121, 56] [121, 110] |
| Audigent SmartPMP | Cost per click -12% to -67%; CTR 1.6 to 2.5 times; VCR +12.5% to +42%; Alcon's target beaten by 27% | Improvement; threshold | DSP-level data on open-exchange inventory; the advertiser's target | Audigent | None [122, 123] [122, 123, 187] |
| Lotame Curation and Curated Marketplaces (interest mark) | Trials: efficiencies +20%, VCR +15%, viewability +21% (2024); PHD EMEA: view-through rate above 70%; page: no MFA | Improvement; threshold | None; an unsourced 60% benchmark | Lotame, including a PHD EMEA case (UK, Spain, Singapore) | Viewability (no baseline); MFA (unmeasured) [124] [125] [209, 13] |
| Adelaide high-attention PMPs | Teads placements scoring 31% above Adelaide's Teads benchmarks; Colgate media quality +40% (2023); outcome lifts | Improvement | Adelaide's own benchmark; standard media | Adelaide; a Kantar study, funding unstated | Attention, scored by the metric's owner [126] [127] [60] |
| YMax (Yieldmo) | CTR +81%, cost per click -65% | Improvement | IAB (Interactive Advertising Bureau) contextual targeting on identical inventory | Yieldmo | None [11] |
| Seedtag NeuroX Curation | Neural engagement 3.5 times higher | Improvement | Non-contextual ads | Seedtag | None [128] |
| Multilocal | Carbon footprint -40% to -60%; performance ranges | Improvement | Traditional programmatic ads | Multilocal; no methodology | Carbon [129, 293] [129] |
| Index Marketplaces | Conversion rate +56%; cost per booking -41%; 0% of Dos Equis impressions outside college-football content | Improvement; threshold | None, or undefined benchmarks | Index case studies | None [201, 202, 203, 204] |
| Kargo curated marketplace (2019) | IVT below 0.3% | Threshold | None | HUMAN Security, via Adalytics | IVT [69] |
| PubMatic and OpenX auction packages; Sharethrough deals and PMPs (statements quoted in 2024) | MFA removed or blocked | Threshold | None | The exchanges, via Adalytics | MFA [68] |

PubMatic's Agence 79 release does not say whether its 12.5% viewability gain over an 80% target is relative or in points, so achieved viewability was 90% or 92.5% [119, 292]. A version claiming 14.3% higher video performance redirected to this case on 3 November 2026 [119, 292]. The Hyundai figures belong to OpenXBuild custom decisioning, a listed comparator product, and OpenXSelect was not screened [119, 120, 121, 125]; [chapter 10](#agents) codes the agent. Microsoft's thresholds are hard limits on predicted rates, so its threshold promise rests on a model [110, 111] [121, 110].

Figures from three further products code as neither kind. Equativ publishes a planning-time cut and a Forrester Wave score, which rates vendor capability [15]. The Forrester Total Economic Impact study commissioned by TripleLift models platform economics [130]. Dstillery's 2022 blog post quotes CPAs from KORTX, one of its customers [70]. The comparators closest to the H3 rule, PubMatic's same campaign on the open market and Yieldmo's identical inventory, come with no published design [119, 292] [11]. Adelaide's figures use its own attention unit (AU), whose validation the attention paper in this series covers, and Teads' release mixes measured lifts with projections [126] [127]. Multilocal's case studies could not be opened [129, 293].

### Products that make no quality claim

Several products publish no quantified quality claim in the sources opened, leaving H3 nothing to test. Stagwell Curate cites pre-vetted inventory and Sincera-powered models without a figure [5, 58, 107, 254, 189, 260, 255, 196]. IAS Curation describes MFA filtration and reports no MFA share [5, 58, 107, 254, 189, 260, 255, 196] [58, 262]. Magnite's ClearLine release, Nexxen Curated Marketplace, FreeWheel Curation Hub and Google Ad Manager Curation carry no quality figure [5, 58, 107, 254, 189, 260, 255, 196]. Experian promises more predictable performance than standard PMPs without a number [5, 58, 107, 254, 189, 260, 255, 196]. Index's January 2026 AU release and Comscore's June 2025 Certified Deal IDs release carry no deal-quality figure [204, 227]. Samba mentions PMPs in one homepage line [57, 268]. Verve's marketplaces, Criteo Commerce Grid, InMobi Buyer Hub and the WPP Media, dentsu and KINESSO marketplaces were not screened for quality claims [119, 120, 121, 125]. Each absence is bounded to the documents opened.

### What a test that could settle H3 would need

A settling test needs four things. First, a named deal's written promise: metric, threshold or uplift, period and definition, such as MRC's viewability and IVT guidelines [146] [148, 149]. The exchanges' MFA statements already name their scope [68]. Second, a comparator: the same publishers and placements, bought by the same buyer in the same period outside the deal. The cleanest design splits eligible bid requests at random between the deal and the open auction. The ANA's bands compare different inventory, so they describe without deciding [26].

Third, a measurer with no financial interest in the deal, stated funding and, where it exists, MRC accreditation for the metric, testing each metric separately [26]. Fourth, an estimate with an interval read against the registered margin, replicated across deals and periods, since MFA supply moves within weeks [101].

The data paths exist. The ANA reconciles log-level data and once split PMPs by domain band [26] [26]. ISBA and PwC matched PMP impressions at above 70%, helped by deal IDs [106, 283]. Jounce read curated deal IDs on a global sample of 1.5 billion bid requests from 44 exchanges in March 2025, AdExchanger reports [101, 64]. A benchmark with a deal-level field, or a buyer releasing deal-level logs, could run the test.

### Coding H3

On the registered rule H3 is unresolved, and no promise meets the registered margin on independent evidence. Every quantified claim on a curated product rests on vendor, agency or partner data, which the rule's third clause codes as unresolved [119, 120, 121, 125] [60, 209, 129, 70]. The table's claims on registered dimensions are viewability (PubMatic's two cases, Microsoft, Lotame), attention (Adelaide), carbon (Multilocal), MFA (Lotame and the three exchanges) and IVT (Kargo). PubMatic's viewability target, Microsoft's threshold, Lotame's no-MFA promise, the exchanges' exclusions and Kargo's ceiling are threshold claims. The other viewability figures, Adelaide's score and Multilocal's cut are improvement claims. In the sources screened, no independent party measured any of them [310, 101, 311].

"Supported" needs independent measurement with a published method that meets the margin for the stated promise, and none exists in the sources screened [310, 101, 311]. The ANA's 11-to-500 band beat the open market on MFA by more than the margin, but it is no product's promise and its comparator is other inventory [26].

"Contradicted" needs independent measurement finding the promise unmet, or the comparator no worse beyond the margin. Adalytics' examples carry no rates, name no deal covered by the exchanges' statements and come from a firm with unstated funding [68] [69]. The ANA's over-500 band was 0.3 points worse than the open market on IVT, inside the 2-point margin [26]. The benchmark's MFA rates compare two buying paths of the same advertisers, not the same inventory before and after curation [100].

The verdict is the registered default. One independent measurement of a named deal, with a published method, would be enough to code that deal's promise.



## 8. Buyer-built marketplaces and the comparators


*When an agency group builds its own curated marketplace, who sets the rule, whose technology runs it and who is paid? On the public record the agency sets the rule, and an exchange the operator names runs it or is announced to. No buyer-built marketplace discloses a fee, a fee recipient or the party that contracts the deal. The comparators packaged inside demand-side platforms (DSPs) use the same vocabulary and run their rules inside one buying platform, which keeps them out of the curation subset.*

> **In short.**
> Stagwell launched Stagwell Curate on 16 July 2026 and called it available immediately to clients of its agencies [5]. Stagwell owns the rules [5, 71]. FreeWheel was announced in April as the technology that will power it, and Magnite joined it as a named supply partner in July, both in the future tense [5, 71] [72]. The connection of Stagwell's Media Machine buying agents is also future tense, and Stagwell's Q3 2026 results release names neither product [5, 71]. The other agency marketplaces have older records: WPP Media's Premium Marketplace from 2022 and 2023, dentsu Curate from 2020 in Asia-Pacific, and KINESSO's unnamed client marketplaces from 2023 [7] [18] [6] [294]. Neither Omnicom's completion filing for Interpublic nor Publicis's Lotame release names a media marketplace [47] [40]. No buyer-built marketplace discloses a fee, take rate or margin in the sources screened by the cutoff [5, 72] [7] [6] [294]. Roku Curate packages Roku's own inventory with partner data, so it falls outside the multi-publisher subset [8]. The DSP-side comparators select inside one buying platform, and among their pages only Amazon's states a rate, a 1% technology fee on guaranteed deals with no base given [297] [274, 275] [276] [299].

> **Note on interests.**
> The author worked at Hearts & Science, an Omnicom agency, and at Lotame, which Publicis announced it would acquire in March 2025 [40]; both groups appear in this chapter. He co-leads the Signals & Measurement working group of AgenticAdvertising.org (AdCP, the Ad Context Protocol, on which Apostra says it is built) and maintains an open-source AdCP signals adaptor. No Fluff Advisory operates two agents listed in the AdCP registry. None of these earns any evidence weight.

### Stagwell Curate: the agency's rule on exchange technology

Of the agency marketplaces screened, only Stagwell Curate has a 2026 launch record [5] [18] [6] [294]. Stagwell's release of 16 July 2026 calls it "a new centralized deal marketplace and bespoke inventory curation platform" for streaming TV, video, display and audio [5]. The release says it is "available immediately to clients across the Stagwell network through their existing agency teams with no DSP migration required" [5]. Digiday's report the same day said Stagwell is "set to launch" it [71]. The release names no client, spend figure or deal count [5].

On its own record Stagwell is the rule owner [5, 71]. The platform lets it "apply its own intelligence, quality standards and performance signals upstream", and no third party is named as setting the rules [5, 71]. Its supply-quality models are "powered by The Trade Desk's Sincera" [5, 71]. Digiday adds that the system uses AI agents built in Claude and The Trade Desk's OpenSincera data for a "domain health check"; the release names neither [5, 71] [71].

The execution host appears in two documents less than three months apart, both in the future tense [5, 71] [72]. Stagwell's FreeWheel release of 27 April 2026, read in ExchangeWire's reproduced copy because the wire original refused access, says "Alongside other partner integrations, FreeWheel will power Stagwell Curate" [72] [72]. It adds that "FreeWheel's Curation Hub and Buyer Cloud will integrate into Stagwell's media acquisition layer", and does not mention Magnite [72]. The July release says partners "including Freewheel and Magnite will provide access to deep, direct publisher relationships" and names neither FreeWheel product [5, 71]. The April architecture may therefore have changed by launch [72].

The release says the Media Machine's agents will get "instant access to Stagwell's preferred publisher relationships, pre-negotiated PMP deals and exclusive pricing" (PMP, private marketplace) [5]. The operator's label is PMP deals, and no selection rule is published [5, 72]. Whether each deal meets this paper's definition of a curated deal therefore cannot be read from the record ([chapter 1](#defining)). No fee, take rate or margin is disclosed; the April release promises "material reduction in ad tech fees" without a figure, a directional claim that discloses nothing ([chapter 6](#fees)) [5, 72].

No live agent connection is documented in the sources screened [5, 71]. Stagwell launched The Media Machine on 19 June 2026 with "over 20 intelligent agents", humans "firmly in control of every critical decision" and no client named [183, 5]. The Curate release says "its buying agents will connect directly to Stagwell Curate" [5, 71]. Stagwell's Q3 2026 results release, an 8-K exhibit of 29 October 2026, names neither Stagwell Curate nor The Media Machine, FreeWheel, Magnite, curation or a marketplace [5, 71]. Its only agent wording is "agentic Stagwell Machines" [5, 71] [73]. The quarter's 10-Q names none of them in its financial notes, and its management discussion was not read [5, 71] [74]. [Chapter 10](#agents) codes the agent record under H7.

### The older agency marketplaces

GroupM introduced Premium Marketplace on 24 February 2022 as "a unified programmatic marketplace that will increase media buying transparency and efficiency" [7]. It covered connected TV (CTV), digital video and display in EMEA (Europe, the Middle East and Africa) and North America, "Supported by global partnership agreements with Magnite and PubMatic" [7]. Clients could choose "all-inclusive, commission-based or SaaS-based payment options" (SaaS, software as a service), types without rates [7].

On 9 February 2023 GroupM said the marketplace hosted "roughly 30 active advertisers" and had integrations with 80% of the top 20 North American CTV publishers [18]. It named Fox, AMC, Univision, LG, Philo and Xumo [18]. It reported an average win rate of roughly 35% inside against about 15% outside, and one client's CPMs (cost per thousand impressions) nearly one-third lower, all without method, period or sample [18]. A lower CPM inside a marketplace does not establish a lower all-in cost [18]. No later page or 2025 or 2026 statement was found in the searches logged, so the product's status is coded unknown [18].

Dentsu Aegis Network launched dentsu Curate on 15 September 2020 on Xandr's technology platform, to "curate supply direct from publishers" into "a central marketplace" [6]. It ran in Hong Kong, South East Asia and Singapore, to be scaled to other Asia-Pacific markets, as a "Supply Path Optimisation & Curated Marketplace offering" [6]. Its product text names no US availability [6]. Its viewability and performance gains are vendor claims without method [6]. No 2025 or 2026 dentsu record was found in the one search logged; a reported December 2022 India launch surfaced only as an unopened search lead [6].

KINESSO's page of 21 September 2023 describes "creating curated marketplaces specific to each client's function and needs" and names no product, exchange, SSP (supply-side platform), DSP or deal mechanism [294]. The page states Interpublic ownership, describing IPG Mediabrands as Interpublic's "media and marketing solutions division" [294]. Omnicom's 8-K exhibit of 26 November 2025 announced the completion of its acquisition of Interpublic at 0.344 Omnicom shares per Interpublic share, and mentions no marketplace, programmatic product or curation [47]. Current KINESSO and Omnicom Media pages were not opened, so whether the client marketplaces continue under Omnicom is not established in the sources screened [294].

Publicis announced on 6 March 2025 that Lotame "will be positioned as part of Epsilon, accessible to all Publicis clients and teams", price undisclosed [40]. The release describes Lotame's "end-to-end data marketplaces" and does not mention curation, deals, supply or media buying [40]. Lotame's Curated Marketplaces are a cohort B product, coded with that cohort ([chapter 11](#dq)) [40]. [Chapter 5](#market) dates the ownership changes.

### Retailer off-site supply, listed here

Retailers' off-site curated supply is listed here and not scored; scoring it is left to a separate retail media paper, if one is run. Walmart Connect introduced Connect Select on 27 April 2026 as "a curated marketplace" inside Walmart DSP; Magnite's headline calls it "Walmart DSP Marketplace Connect Select" [271, 272]. At launch it "includes inventory from VIZIO, premium CTV publishers, including Paramount, and Warner Bros. Discovery" [271, 272]. The article also lists FreeWheel, Index Exchange, Magnite and PubMatic as its SSPs, so on the operator's list the supply is multi-publisher and multi-exchange; Magnite's blog names only VIZIO [271, 272]. Deals "are available through Inventory Controls and Deal Desk", and the article credits the SSPs with "post-auction optimizations" [271, 272]. Neither page publishes a fee, minimum, deal ID or selection rule [271, 272].

Supermarket News reported Kroger Precision Marketing's Private Marketplace on 20 October 2021 [75]. As reported, it applied Kroger audience data in the buyer's "preferred ad-buying platform" through "a deal ID that allows for appropriate programmatic use" [75]. The inventory, publishers and exchanges are not described, so publisher multiplicity is unknown [75]. The same article records Publicis Groupe among the alpha testers [75].

Albertsons Media Collective's page describes off-site display through Google's Display & Video 360 (DV360) and Collective TV, and a partnership with The Trade Desk for its "verified-buyer audience and measurement solutions" [295, 296]. It shows no curated inventory, PMP, deal ID or publisher list, so the route is audience activation inside buying platforms, a comparator mechanism [295, 296]. Instacart's off-platform page redirected to a European corporate page, and PubMatic's April 2024 Instacart release refused access, so Instacart's curated supply is access unknown [76].

### Roku Curate, the single-publisher boundary case

Roku's page of 27 April 2026 calls Roku Curate "a premium advertising package" [8]. It "combines Roku's proprietary insights and inventory" with "in-market targeting data from partners like Best Buy Ads, Criteo, Fandango, Fetch, Instacart and Kroger" [8]. The partners supply data, and no sentence on the page adds another publisher's inventory [8]. Roku contrasts it with a standard deal ID or PMP, where "the publisher sends inventory, while the targeting and decisioning happen at the DSP level" [8]. With Curate, "Roku packages the data and media supply" [8]. Roku can activate the package itself or "send the package to the advertiser's preferred demand-side platform", none named, and promises "lower fees" without an amount or a named collector [8].

On this paper's definitions Roku Curate is single-publisher curation: partner data applied to one publisher's inventory before the package reaches the buyer ([chapter 1](#defining)) [8]. The primary analytical subset is multi-publisher curation, so Roku Curate is listed and coded, and excluded from that subset and from the scored cohorts [8].

### The comparators, as their vendors describe them

The comparators are described and never scored as curation. Each row is the vendor's own description, which records what the vendor says it offers and nothing about use or performance.

| Product (operator) | The vendor's description | Where selection runs | Fee stated | Date |
|---|---|---|---|---|
| OpenPath (The Trade Desk) | "direct connections between advertisers and a publisher's ad server"; "does not function as an intermediary"; "hundreds of publishers" | Inside The Trade Desk's DSP; no selection rule stated | None; a 3x win-rate claim on October 2024 platform data across "dozens of advertisers", no method | Page of 24 February 2026 [297] |
| OpenSincera (The Trade Desk) | A free application of Sincera's ad-experience metadata; no curation or deal role | A data product | Free | Announced 13 May 2025; available 6 June 2025 [237] |
| DV360 Marketplace (Google) | "curated bundles", "Lists of curated inventory that advertisers can assign directly across line items"; publishers, packages and deals in one list | Inside DV360; who assembles packages or bundles is not stated | Not stated | Undated help page [273] |
| Inventory Tier brand suitability control (Amazon DSP) | Three advertiser-selected tiers, Expanded, Standard and Restrictive, balancing reach and brand safety | A setting inside Amazon DSP | Not stated | 14 November 2025; page not reopened by the verifier [298] |
| Certified Supply Exchange and Amazon Publisher Direct (Amazon Ads) | An exchange programme, no exchange named; invitation-only links to "thousands" of publishers; quality measures "at no additional cost" | Amazon DSP | "a single 1% technology fee" on guaranteed deals with publishers outside Amazon; base not stated | Undated article, served in Italian; quotations translated [299] |
| Yahoo Backstage (Yahoo DSP) | "a direct path to curated, premium publisher inventory"; "over 100 publishers" at announcement; now "curated inventory of trusted, MFA-free publishers" (MFA, made-for-advertising) | Publishers integrate directly with Yahoo DSP; no curator named | None; "reduce tech tax" claimed | Announced 19 June 2023 [274, 275] |
| Viant Direct Access | "Viant's innovative supply path optimization program"; 19 CTV and audio publisher logos; "Independence from SSP-curated deal ID limitations" | Viant's DSP; no selection rule stated | None; "Reduce ad tech fees" promised | Undated page [276] |
| PubMatic Activate | Its pages refused access; PubMatic's buyer page, read in a browser and unverified because the verifier's fetch was refused, says curated deals "can be activated through PubMatic's Activate solution or your DSP of choice" | PubMatic's own buying route | Access unknown | Not established [223, 20] [4] |
| Apostra (formerly Scope3) | "one place to buy and sell every kind of advertising", "built on AdCP"; sellers "stay in control of your pricing, packaging, and relationships"; no curated-deal product | Apostra's platform | None stated | 23 September 2026 [54] [54] |
| OpenXBuild (OpenX) | Buyers "deploy your data and decisioning in a secure container within OpenX cloud infrastructure" to "enrich and shape bid requests before they reach the DSP" | Inside the exchange, on the buyer's or its model vendor's rule | None stated | Hyundai release of 16 June 2026 [300, 120] [120] |

Most fee language among the comparators is directional and carries no figure [274, 275] [276]. Yahoo promises to "reduce tech tax" by "eliminating a supply chain hop", and Viant to "Reduce ad tech fees and non-working media spend" [274, 275] [276]. Amazon's 1% is the only stated rate, a DSP charge on one transaction type and not a curation fee ([chapter 6](#fees)) [299]. Yahoo's "MFA-free" claim names no measurer, so it stays a vendor assertion with no evidence weight ([chapter 7](#quality)) [274, 275].

Other records add to three rows. The vendors' release of 15 December 2025 says the AdCP-enabled PubMatic and Butler/Till campaign ran "through PubMatic's Activate platform" [93, 120, 172, 168, 171] [77]. The OpenX release says Chalice AI's models "operate as always-on bidding agents within OpenX's infrastructure" for Hyundai [120]. It claims 20% lower cost per high-value action and 67% lower CPMs than an undefined "second-best-on-plan", with no period, sample or method [120]. Apostra says "campaigns are live" in six markets, the US among them, and names no customer or transaction; [chapter 10](#agents) codes these under H7 [54].

### What separates DSP-side packaging from sell-side curation

Location separates them. In this paper a curated deal applies a documented selection or enrichment rule before the bid request reaches the buying platform, and a package built inside that platform is DSP-side packaging ([chapter 1](#defining)). OpenPath, Backstage and Direct Access route publishers straight into one DSP, and none states a selection rule beyond membership [297] [274, 275] [276]. DV360's bundles are lists a buyer assigns to line items [273], and Amazon's tiers, on a page the verifier did not reopen, are a DSP setting [298]. Roku draws the same line: with a standard deal ID or PMP, "the targeting and decisioning happen at the DSP level" [8].

The vocabulary does not sort the products. "Curated" appears on Yahoo's direct path, DV360's bundles and Walmart's DSP marketplace, and Viant sells independence from "SSP-curated deal ID limitations" [274, 275] [273] [271, 272] [276]. The register therefore codes the execution host, which the adjective never states.

Reach follows from location. A DSP-side package lives in one platform; Stagwell sells Curate with "no DSP migration required", and Roku can send its package to the advertiser's own DSP [5] [8]. Digiday quotes Stagwell's Matt Adams that Curate should reduce reliance on "ubiquitous" marketplace solutions such as OpenPath, a quotation the verifier did not recheck [71]. On Stagwell's account, The Trade Desk, which operates OpenPath, also supplies the Sincera data behind Curate's quality models [5, 71] [297] [237].

Two products sit near the line. Connect Select lives inside Walmart DSP, with SSP partners running "post-auction optimizations", so on the operator's words the marketplace sits on the buying side and the SSPs act after the auction [271, 272]. OpenXBuild inverts exchange curation: the buyer's or its vendor's rule runs inside the exchange before the request reaches the DSP [300, 120]. Cohort C, on these records, pairs a buy-side rule owner with sell-side execution, as when Stagwell applies its standards "upstream" on supply that FreeWheel and Magnite are announced to provide [5, 71] [72].

### Coding Q1 for buyer-built marketplaces

Q1 is a coding question with no verdict. Its rule codes each field from vendor documentation or an official record, codes a marketing adjective as unknown, and leaves unknown what no opened source states. Each code here records what the operator says it does. Stagwell's execution host rests on future-tense statements, which establish an announced architecture and document no deployment.

| Marketplace | Rule owner | Execution host | Data supplier | Contracting seller, fee recipient, change party | Mechanism |
|---|---|---|---|---|---|
| Stagwell Curate (launched 16 July 2026) | Stagwell | FreeWheel (Curation Hub, Buyer Cloud), with Magnite; announced in the future tense; the client's existing DSP | Stagwell's signals; The Trade Desk's Sincera for quality models | Unknown | Pre-negotiated PMP deals under Stagwell's standards applied upstream; selection rule unpublished [5] [5, 71] [72] [5, 72] |
| WPP Media Premium Marketplace (2022; last page 2023; status unknown) | GroupM, now WPP Media | Magnite and PubMatic, roles not separated | Unknown | Unknown; payment-option types only | Seller list of direct publisher integrations; selection rule unpublished [7] [18] [18] |
| dentsu Curate (2020, Asia-Pacific; no later record) | dentsu (Dentsu Aegis Network) | Xandr's platform | Unnamed data partnerships | Unknown | Seller list of publishers with data enrichment [6] |
| KINESSO client marketplaces (2023; no product name) | KINESSO | Unknown | Unknown | Unknown | Unknown [294] |
| Omnicom | No named marketplace on the record opened | Unknown | Unknown | Unknown | Unknown [47] |
| Publicis | No named media marketplace; Lotame coded in cohort B | Unknown | Unknown | Unknown | Unknown [40] |

Of the four agency marketplaces, the rule owner is coded for all four, each the agency itself on its own pages [5, 71] [7] [6] [294]. The execution host is coded for three, one of them as announced intent [72] [7] [6]. A data supplier is named for one, The Trade Desk's Sincera for Stagwell [5, 71]. The contracting seller, the fee recipient and the party able to change a deal after purchase are unknown for all four, and no row publishes its selection rule [5, 72] [7] [6] [294].

The listed rows code the same way. Walmart Connect owns the rule for Connect Select, which runs in Walmart DSP with four named SSPs [271, 272]. Kroger's buyer designs the audience within Kroger's data and activates it "in your tech stack" [75]. Albertsons' advertiser selects audiences in DV360 or The Trade Desk [295, 296], and Instacart is access unknown [76]. Roku owns the rule, contracts the package and executes it itself or in the advertiser's DSP, on data from six named partners [8]. The fee recipient is unknown for every listed row; Roku's "lower fees" names no collector [271, 272] [75] [8].

In plain words: in a buyer-built marketplace the agency decides what goes in, and an exchange runs it, or is announced to, where the agency names one [5, 71] [72]. The public record does not say who is paid, who contracts the deal or who can change it after purchase [5, 72]. No buyer-built marketplace discloses a fee in the sources screened by the cutoff; a client fee schedule, a contract term or a current product page would move these codes [5, 72] [7] [6] [294].



## 9. The competition record


*What does the competition record establish for sell-side decisioning? A US court found that Google's exchange held first-look, last-look and floor advantages inside Google's own ad server [79, 156, 78]. The parties' proposed judgment reaches Google's open and private auctions for open-web display [79]. No judgment appears on the docket copy of 29 October 2026, so no US remedy is in force [80]. None of the three Virginia court documents mentions curated deals or curators [79, 156, 78].*

> **In short.**
> On 17 April 2025 a federal court in Virginia found that Google had monopolised the worldwide open-web display publisher ad server and ad exchange markets [78, 156]. It found four auction practices anticompetitive, First Look and Last Look among them [78]. Its remedies opinion of 2 September 2026 rejected divestiture and accepted most behavioural remedies as modified, with a six-year term [156, 80] [156, 80]. The joint proposed judgment of 2 October 2026 still carries disputed brackets and, apart from an optional route for direct deals, covers only Google's open and private auctions for open-web display [79] [79] [79]. No judgment or appeal appears on the docket copy of 29 October 2026 [80]. In New York, Google may not relitigate the Virginia market, tie and conduct findings against the publisher movants or the competitor plaintiffs [131, 132]. The European Commission fined Google EUR 2.95 billion in 2025 but had reached no final conclusion on a structural remedy by 12 March 2026, and the UK case is open [134, 133] [137] [138, 157, 158]. In the US the privacy rules that govern the personal data used in in-flight enrichment are state laws: nineteen in force at the cutoff, four enacted in 2026 for later years [85].

### The dated steps

The US final judgment is still a proposal and the New York cases are pending [80] [132]. By 12 March 2026 the Commission had not decided how Google must end its conflicts of interest, and the UK file is open [137] [138, 157, 158]. The table lists every step and the figure a selection of them. Each step comes from a court or regulator record, except Google's own account of its EU plan. Steps are coded found, decided, proposed, ordered or in force, with filed, opened, pending and open where none fits. In the table, E.D. Va. is the US District Court for the Eastern District of Virginia and S.D.N.Y. MDL the multidistrict litigation in the Southern District of New York. CMA is the UK's Competition and Markets Authority, and Dkt marks a docket entry.

![A court has found Google liable and decided remedies, but no judgment is in force; Europe’s remedy is undecided.](figures/fig-docket.svg)

*Figure fig-docket. **A court has found Google liable and decided remedies, but no judgment is in force; Europe’s remedy is undecided.** Selected dated steps in the Google ad-tech proceedings, each from the court’s docket or the regulator’s own document. The E.D. Va. remedies opinion of 2 September 2026 rejected an AdX divestiture and accepted behavioural remedies; the parties’ proposed final judgment of 2 October 2026 still carried disputed terms, and nothing had been entered by 29 October. The full list of 22 steps is in chapter 9. Units: dates. Scope: E.D. Va., S.D.N.Y. multidistrict litigation, European Commission and General Court, UK CMA. Sources [79], [78], [131], [132], [80], [133], [134], [135], [136], [137], [138]. Caveat: Status-sensitive: dockets read by 3 November 2026. A remedy proposed or decided is not a remedy in force; the opinion’s timetable runs from entry of a final judgment. Data: data/figure-data.json#docket*

| Date | Forum | Step | Status | Source |
|---|---|---|---|---|
| 22 June 2021 | European Commission | Case AT.40670 opened | Opened | [134, 133] |
| 25 May 2022 | UK CMA | Ad-tech investigation opened | Opened | [138, 157, 158] |
| 24 January 2023 | E.D. Va. | Complaint, No. 1:23-cv-00108 | Filed | [79, 80] |
| 14 June 2023 | European Commission | Statement of objections | Found (preliminary) | [134, 133] |
| 6 September 2024 | UK CMA | Statement of objections | Found (preliminary) | [138, 157, 158] |
| 17 April 2025 | E.D. Va. | Liability opinion, Dkt 1410 | Found | [78, 156] |
| 5 September 2025 | European Commission | Prohibition decision and EUR 2.95 billion fine | Decided | [134, 133] |
| 27 October 2025 | S.D.N.Y. MDL | Issue-preclusion order for publisher movants | Ordered (interlocutory) | [131, 132] |
| 14 November 2025 | Google | Compliance proposal, by Google's account | Proposed | [135, 136] |
| 20 November 2025 | General Court | Annulment action T-794/25 lodged | Pending | [135, 136] |
| 12 January 2026 | General Court | Official Journal notice of the action | Pending | [135, 136] |
| 13 January 2026 | S.D.N.Y. MDL | Pre-Trial Order No. 20 | Ordered | [132] |
| 15 January 2026 | European Commission | Provisional non-confidential decision published | Decided (published) | [134, 133] |
| 23 February 2026 | S.D.N.Y. MDL | Consolidated competitor complaint | Filed | [132] |
| 12 March 2026 | European Commission | Answer to Parliament: no final conclusion on remedy | Found | [137] |
| 29 April 2026 | S.D.N.Y. MDL | Google's motion to dismiss | Pending | [132] |
| 19 May 2026 | UK CMA | Timetable updated; case open | Open | [138, 157, 158] |
| 18 June 2026 | S.D.N.Y. MDL | Case management plan | Ordered | [132] |
| 25 June 2026 | S.D.N.Y. MDL | Preclusion extended to competitor actions | Ordered | [131, 132] |
| 2 September 2026 | E.D. Va. | Remedies order and sealed opinion, Dkt 1857 and 1858 | Decided | [156, 80] |
| 16 September 2026 | E.D. Va. | Remedies opinion unsealed, Dkt 1860 | Decided | [156, 80] |
| 2 October 2026 | E.D. Va. | Joint proposed final judgment, Dkt 1861-1 | Proposed | [79] |

The Virginia docket copy, last updated 29 October 2026, lists nothing after 2 October; PACER, the federal courts' own system, was not checked [80].

### What the Virginia court found

On 17 April 2025 the US District Court for the Eastern District of Virginia (E.D. Va.) found two monopolies, a tie and four anticompetitive practices, all concerning Google's own ad server and exchange [78] [79, 156, 78]. The monopolies were in the worldwide open-web display publisher ad server and ad exchange markets [78, 156]. It found DFP, Google's ad server, unlawfully tied to AdX, its exchange, dismissed an advertiser ad network count and entered no remedy [78, 156].

First Look gave AdX a first right of refusal on every DFP impression, even where the publisher wanted to rank other exchanges first [78]. AdX buyers could therefore win against higher rival bids [78]. Last Look let AdX see the winning header-bidding bid, entered in DFP as a floor, and outbid it [78]. By 2019 Google agreed to remove Last Look and introduced Unified Pricing Rules [78]. These barred publishers from setting higher floors for AdX than for rival exchanges, or for AdWords than for other buying tools [78, 156]. They still allowed higher floors on rival exchanges than on AdX [78, 156].

Dynamic revenue share let AdX vary its take impression by impression to "exploit the last look advantage" [78]. Its overall take was around 20%, with 15% or lower on competitive impressions and 25% or higher on less competitive ones [78]. Google staff recognised that Index Exchange, Magnite and Xandr often charged closer to 10% [78]. These court characterisations of exchange take rates on the trial record join no curation-fee series and say nothing of 2026 rates ([chapter 6](#fees)) [78] [78].

### The remedies opinion of 2 September 2026

The remedies opinion rejected every structural remedy and accepted most behavioural ones as modified [156, 80]. It cites closing arguments from the remedies-trial transcript of 21 November (2025, by inference from the opinion dates); when the trial opened is not established in the sources screened [156, 80]. The order (Dkt 1857) and sealed opinion (Dkt 1858) rejected divestiture of AdX, open-sourcing of DFP's final auction logic and a contingent divestiture of the rest of DFP [156, 80]. The court found structural remedies "neither realistic nor needed", and the opinion was unsealed on 16 September [156, 80].

The accepted remedies would bar the DFP-AdX tie and any return of First Look or Last Look, and would deprecate Unified Pricing Rules for all indirect transaction types [156]. Rival ad servers and Prebid, the open-source header-bidding software, would get AdX real-time bids for indirect transactions, "including private auction and open auction deals" [156]. Publishers keeping DFP could route programmatic guaranteed and preferred deals through Prebid [156]. Data sharing, non-discriminatory bidding and buy-side rules for AdWords and Display & Video 360 (DV360) complete the set [156].

The court chose Google's six-year term over the plaintiffs' fifteen years [156, 80]. The judgment should be implemented within 15 months and take effect 60 days after entry, with work on the Monitor and Technical Committee starting on entry [156, 80]. The parties were directed to file one joint proposed judgment within 30 days, labelling disputed versions by proponent [156, 80].

### The proposed judgment and its brackets

Except for one routing provision for direct deals, the joint proposal of 2 October 2026 (Dkt 1861-1) applies only to Indirect Demand, meaning Open Auctions or Private Auctions, for open-web display [79]. It excludes AdSense, AdMob, DV360 and every Google tool except Google Ad Manager and Google Ads [79]. Its Private Auction excludes Preferred Deals, Programmatic Guaranteed Deals and similar individually negotiated transactions [79].

Within that scope Google would be barred from reimplementing First Look or Last Look, and would offer Header Bidding Trafficking to all DFP publishers within 6 months [79]. Unified Pricing Rules would go, with no uniform-floor requirement or fee discount [79]. Within 15 months Google would build an optional server-to-server route through Prebid for programmatic guaranteed and preferred deals [79]. The Effective Date, from which these deadlines run, is 60 days after entry and tolled during any stay; the judgment expires on the Effective Date's sixth anniversary [79] [79].

The plaintiffs want AdX interoperability with Prebid and rival ad servers within 6 months; Google wants 12 [79]. The plaintiffs define AdX as all Ad Exchange functionality, while Google limits it to its open and private auctions [79]. Google would make the Effective Date "Subject to the outcome of any motion to stay this Final Judgment pending appeal" [79]. The sides also differ on who proposes the Monitor slate, on what data DFP passes to Prebid, and on 30 or 60 days to appoint the Technical Committee [79]. A notice of appeal in this case is due within 60 days of entry, so none can precede a judgment [80].

### The private suits and the New York litigation

The sell-side damages suits sit in, or are coordinated with, one multidistrict litigation (MDL) in the Southern District of New York (S.D.N.Y.) [132] [81]. The OpenX complaint and the PubMatic and Magnite releases cite the April 2025 opinion by date; Teads' release cites the Virginia ruling undated [301, 302, 303, 132, 304, 305]. No docket text or docket card read shows any of the nine actions in the table dismissed or decided [132] [81].

| Plaintiff | Filed | First court and number | S.D.N.Y. number |
|---|---|---|---|
| OpenX | 4 August 2025 | E.D. Va. 1:25-cv-01282 | 1:25-cv-10817 [301, 302, 303, 132, 304, 305] [132] |
| PubMatic | 8 September 2025 | E.D. Va. 1:25-cv-01482 | 1:25-cv-10819 [132] [82] |
| Magnite | 16 September 2025 | E.D. Va. 1:25-cv-01541 | 1:25-cv-10818 [132] [82] |
| Equativ and Sharethrough | 13 October 2025 | E.D. Va. 1:25-cv-01755 | 1:26-cv-00140 [132] |
| Index Exchange | 10 November 2025 | E.D. Va. 1:25-cv-02003, voluntarily dismissed | 1:25-cv-10477, filed 17 December 2025 [301, 302, 303, 132, 304, 305] [82] |
| Kargo | Not found | S.D.N.Y. | 1:25-cv-10630 [132] |
| Sovrn | 25 February 2026 | S.D.N.Y. | 1:26-cv-01587 [132] [81] |
| TRUSTX | 16 April 2026 | S.D.N.Y. | 1:26-cv-03135 [132] [81] |
| Teads | 3 August 2026 | S.D.N.Y. | 1:26-cv-06591, coordinated per entries of 14 August and 14 October 2026 [301, 302, 303, 132, 304, 305] [81] |

Equativ's case arrived on 8 January 2026 by conditional transfer order [132]. Index Exchange's New York action is a fresh filing [82]. Equativ, PubMatic, OpenX and Magnite filed a consolidated complaint on 23 February 2026, and Google's motion to dismiss was fully briefed by 12 June [132]. Whether it has been decided cannot be read from the docket copy, whose later entries show no text [132]. Fact discovery in the eight competitor actions is due to close on 10 November 2026 and expert discovery on 30 April 2027 [132].

Preclusion carries named Virginia findings into these cases [131, 132]. On 27 October 2025 Judge P. Kevin Castel barred Google from relitigating, against the publisher movants, the two markets, their worldwide scope and the tie [131, 132]. The bar also covers Unlawful Tying, First Look, Last Look, Dynamic Revenue Share and Unified Pricing Rules [131, 132]. A stipulated order of 25 June 2026 extended the same preclusion to the competitor cases; its own list of findings was not opened [131, 132]. Preclusion fixes liability facts for damages and orders no change in conduct [131, 132].

Publishers have sued too: docket cards show S.D.N.Y. suits filed between 13 January and 13 July 2026 by The Atlantic Monthly Group, Vox Media, McClatchy, Advance Publications, Ziff Davis, the Los Angeles Times, Mediavine, WP Company, MH Sub I (WebMD), Nexstar and TEGNA, and Valnet [81]. None was opened, so what they plead is not recorded here [81]. Reporting says Teads alleges Last Look was never fully retired; its release filed with the US Securities and Exchange Commission does not say so, and the allegation is untested [78].

### Europe and the UK

The European Commission has decided an infringement and a fine; its case register shows no remedy decision for Google's conflicts of interest [134, 133] [137]. On 5 September 2025 it fined Google EUR 2.95 billion in case AT.40670 [134, 133]. It found Google dominant in publisher ad servers and open-web buying tools across the European Economic Area, and abusive since at least 2014 in favouring AdX [134, 133]. One example was telling AdX in advance the best rival bid it had to beat [134, 133]. It required Google to end the self-preferencing and its conflicts of interest, and gave it 60 days to say how [134, 133]. The Commission's preliminary view was that only divesting part of Google's services would end the conflicts [134, 133].

Google's plan of 14 November 2025, by its own account, lets publishers set different minimum prices for different bidders and avoids a break-up [135, 136]. Google and Alphabet's annulment action at the General Court, T-794/25, lodged on 20 November 2025, is pending [135, 136]. Its pleas contest findings that DFP favoured AdX bids over header bidders before and after 2019 [83]. Virginia found that Google agreed by 2019 to remove Last Look; what the Commission found after 2019 only the unopened decision text would say [78] [83].

On 12 March 2026 Executive Vice-President Ribera told the European Parliament that the Commission "has not reached a final conclusion" on a structural remedy and was consulting on Google's plan [137]. The case register and one web search found nothing later [137]. Virginia found structural remedies unneeded; on 12 March 2026 the Commission's preliminary view still favoured a structural remedy [156, 80] [137].

The UK's Competition and Markets Authority (CMA) has an open ad-tech case, and its case page records no decision [138, 157, 158]. The case opened on 25 May 2022, and objections followed on 6 September 2024 [138, 157, 158]. On a page last updated on 19 May 2026, its latest phase, estimated for December 2025 to July 2026, is "Considering next steps, whilst monitoring international developments" [138, 157, 158]. The CMA's strategic market status designations of Google, in general search from 10 October 2025 and in mobile from 22 October 2025, are a separate regime [138, 157, 158]. Their update logs to 23 September 2026 mention no ad tech; the requirement documents themselves were not opened [138, 157, 158].

### The privacy rules on in-flight enrichment

> **Note on interests.**
> The privacy rules below apply to ad-data suppliers. The author has worked at Samba (TV measurement and ad data; owner of Semasio) since May 2025. He was executive vice president at Verve Group from September 2022 to August 2024 and earlier worked at Lotame and Hearts & Science. He co-leads the Signals & Measurement working group of AgenticAdvertising.org (AdCP, the Ad Context Protocol), maintains an open-source AdCP signals adaptor, and No Fluff Advisory operates two agents listed in the AdCP registry. None of these earns any evidence weight.

The privacy rules govern the personal data that in-flight enrichment joins to a bid request [85]. In the US they are state laws, and the statutes themselves were not opened [85]. By the cutoff twenty-three states had enacted comprehensive privacy laws [85] [84]. Nineteen, on the International Association of Privacy Professionals' (IAPP) chart of 24 November 2025, were in force by 1 January 2026 [85] [85]. Alabama, Louisiana, Oklahoma and Vermont enacted theirs in 2026 [84]. Vermont's takes effect on 1 January 2028; the other three take effect in 2027 according to search results the verifier did not open, so those dates are unverified [85] [84]. The laws reach enrichment through definitions of sale, sharing, targeted advertising and sensitive data, which this paper did not code [85].

The California Privacy Protection Agency (CPPA) completed rules on risk assessments and automated decisionmaking technology in September 2025 [159]. It states that its California Consumer Privacy Act (CCPA) regulations and data-broker registration and deletion rules took effect on 1 January 2026; phased compliance dates are not on its page [159]. The Federal Trade Commission (FTC) finalised a consent order against the data broker Mobilewalla on 14 January 2025 [160]. It bans collecting consumer data from real-time bidding (RTB) exchanges except to take part in the auctions [160]. It was the first time the FTC alleged that practice to be unfair; the vote was 4-1, and the order binds Mobilewalla alone [160].

The EU Court of Justice held on 7 March 2024 (C-604/22) that a TC String is personal data when it can reasonably be linked to an identifier such as an IP address [161]. The string is the consent signal of IAB Europe's Transparency and Consent Framework (TCF) [161]. A body that sets such a framework can be a joint controller for the string, a role that does not extend automatically to participants' advertising processing [161]. By IAB Europe's account, Belgium's Market Court on 14 May 2025 annulled the regulator's 2022 decision except that joint-controller finding, limited to TC Strings [162, 163]. A law firm reports the EUR 250,000 fine as upheld [86].

The register below holds these instruments beside the standards coded in earlier chapters. Two parts of it predate the verifier's corrections: its state list stops at the nineteen laws in force, and its Belgian row reads the 2022 decision as annulled in full [85] [162, 163].

**Rules and governance instruments**

| Jurisdiction | Instrument | What it requires | Effective | Scope | Law or practice | Src |
|---|---|---|---|---|---|---|
| Global (IAB Tech Lab, US-based) | OpenRTB 2.6-202606 Pmp and Deal objects | Pmp.private_auction (default 0) restricts bidding to listed deals when 1; Deal carries bidfloor/bidfloorcur, at (1/2/3), wseat and wadomain allow-lists, guar (default 0), mincpmpersec, durfloors; no curator or fee attribute | 2026-06-11 (release 2.6-202606; objects unchanged by the 2025 revisions) | bid-request protocol for open and private-marketplace trading | technical standard | [139] |
| Global (IAB Tech Lab, US-based) | OpenRTB 2.6 Implementation Notes, 7.3 PMP & Direct Deals | Recommends that any prioritisation other than by price, and any preferential floor for a seat, carry a Deal ID; describes publisher-configured priority of Deal ID bids over open-auction bids as one common practice | undated companion to 2.6 (read 2026-11-03) | non-normative bidding and policy guidance | implementation guidance | [95] |
| Global (IAB Tech Lab, US-based) | OpenRTB 3.0 Deal object (in Item) | Deal.id, flr, flrcur, at (1/2/3; 500+ exchange-specific), wseat, wadomain; Item.private (default 0); no guar, mincpmpersec or durfloors | 2022-03 | bid-request protocol (3.0 line) | technical standard | [140] |
| Global (IAB Tech Lab, US-based) | ads.txt 1.1 | Publisher-side list of authorised sellers; 1.1 adds managerdomain and ownerdomain; no curator role | 2022-07 (spec version table) / 2022-08 (Tech Lab page) | web and app publishers; read by buyers | technical standard | [141] |
| Global (IAB Tech Lab, US-based) | sellers.json 1.0 | Advertising systems publish each paid seller account as PUBLISHER, INTERMEDIARY or BOTH; version value 1.0 only | 2019-07-31 | SSPs, exchanges and intermediaries | technical standard | [142] |
| Global (IAB Tech Lab, US-based) | SupplyChain object 1.0 | Every node is a payment participant (hp 'should always be 1'); complete flag; chain ends at the inventory owner when complete; non-payment entities deferred to a future version | 2019-07-31 (per Tech Lab page; document undated) | OpenRTB bid requests (Source.schain in 2.6) | technical standard | [143] |
| Global (IAB Tech Lab, US-based) | Deal Sync API 1.0 (Deals API) | Out-of-band deal sync SSP to DSP; Deal.origin, seller, pubcount, auxdata and dinventory; Curation object curator, cdealid and curfeetype (0-4, a type not an amount); no revisions, proposals or negotiations in 1.0; buyers told deal criteria are not guaranteed in bid requests | public comment 2025-12-04 to 2026-01-31; final 2026-02-06 | deal metadata exchange between origin (typically SSP) and receiving systems (typically DSP) | technical standard | [96] |
| Global (IAB Tech Lab, US-based) | IAB Tech Lab Curation Framework / Curated Audiences workstream | Announced a 2025 Curation Workstream under the Addressability Working Group and the rebranding of Seller Defined Audiences as Curated Audiences; no standalone curation specification found on iabtechlab.com by 2026-11-03 | announced 2024-12-16; no instrument by the cutoff | announcement only | workstream announcement | [144] |
| Global (Prebid.org, US-based) | Prebid 'Deals in Prebid' ad-ops documentation | Recommends ad-server deal line items carry a priority higher than open-market header-bidding line items; hb_deal_BIDDERCODE targeting; no curation page on docs.prebid.org | undated (read 2026-11-03) | publisher ad-server configuration for header-bidding deals | configuration guidance | [145] |
| US (MRC; international adoption encouraged) | MRC Viewable Ad Impression Measurement Guidelines v2.0 (desktop) | Display viewable at >=50% of pixels on an in-focus tab for >=1 continuous second post-render; video at 50% of the ad's pixels for 2 continuous seconds | 2015-08-18 | desktop browser display and video measurement | measurement guideline | [146] |
| US (MRC; international adoption encouraged) | MRC Mobile Viewable Ad Impression Measurement Guidelines | Mobile web and in-app viewable at >=50% of density-independent pixels for >=1 continuous second; supersedes prior mobile guidance | 2016-06-28 | mobile web and in-app measurement | measurement guideline | [147] |
| US (MRC; international adoption encouraged) | MRC Invalid Traffic Detection and Filtration Standards Addendum (IVT 2.0) | Defines GIVT (list/rule-based) and SIVT (requiring advanced analytics, multi-point corroboration, human intervention) categories and filtration requirements for accredited measurers | 2020-06 (update of v1.0 of 2015-10-27) | digital measurement vendors under MRC audit | measurement standard | [148] |
| US (MRC; international adoption encouraged) | MRC 2024 IVT Interim Updates memo | Adds requirements on privacy implications for IVT (GIVT and SIVT), CTV domain/inventory mismatch and bundle ID spoofing (SIVT), and property-level considerations (SIVT) | 2024-04-24 (authored 2024-01) | digital measurement vendors under MRC audit | measurement standard (interim update) | [149] |
| US (MRC; international adoption encouraged) | MRC Cross-Media Audience Measurement Standards (Phase I Video) | Where ads display to a TV (linear or OTT) the 100% pixel criterion may be assumed; 2 continuous seconds still apply; cross-media input requires 100% pixels for 2 continuous seconds; refers to the desktop and mobile viewability guidelines | 2019-09 | cross-media video audience measurement incl. OTT/CTV | measurement standard | [150] |
| US (MRC; international adoption encouraged, section 6.3) | MRC Digital Advertising Auction Transparency Standards | Ad auctioneers (any company that evaluates bids, selects winners and sets prices, even in a multi-stage auction) must disclose auction type, bid-conversion variables incl. deal discounts and tech fees, reserve-price use (reported pre- and post-auction; floors should not be per-buyer), winner-determination variables incl. demand-source priority, and deals that supersede an auction; complete SupplyChain and ads.txt/sellers.json validation for OpenRTB; advance change disclosure; voluntary, with MRC accreditation audit | 2026-01 | all digital ad auctions (search, social, retail media, CTV, programmatic) | disclosure and reporting standard | [151] |
| Global (TAG, US-based) | TAG Certified Against Fraud | Company-level seal: 100% of monetizable transactions filtered for IVT by a qualified third party; data-centre IP filtering; annual January recertification | programme launched 2016 (current guidelines undated on the page) | member companies in the supply chain (publishers may rely on a CAF-certified SSP) | certification programme | [152] |
| Global (TAG, US-based) | TAG Certified Against Malvertising | Company-level seal against malvertising; requirements in the CAM Guidelines PDF (not opened); annual January recertification | programme described as partnered since 2014 (guidelines undated) | member companies | certification programme | [153] |
| Global (TAG, US-based) | TAG Brand Safety Certified | Company-level seal under 'an industry regulated framework for brand safety'; requirements in the Brand Safety Program Guidelines PDF (not opened); none listed on the page | undated | member companies (a 'brand safety framework for sellers') | certification programme | [154] |
| Global (TAG, US-based) | TAG Certified for Transparency | Company-level seal for 'harmonizing campaign log-file data across the entire supply chain'; guideline PDF dated 2023 (not opened); TAG TrustNet is a DLT initiative powered by Fiducia, not a seal | announced 2022-11; guidelines 2023 | digital advertising organisations | certification programme | [155] |
| United States (E.D. Va.) | United States v. Google LLC, Memorandum Opinion Dkt 1410 (liability) | Finds monopolisation of the worldwide open-web display publisher ad server and ad exchange markets and unlawful DFP-AdX tying; finds First Look, Last Look, sell-side dynamic revenue share and Unified Pricing Rules anticompetitive; orders no remedy | 2025-04-17 (entered) | Google DFP and AdX; open-web display | court finding (found) | [78] |
| United States (E.D. Va.) | United States v. Google LLC, Order Dkt 1857 and Memorandum Opinion Dkt 1858 (remedies) | Rejects AdX divestiture, open-sourcing of DFP auction logic and contingent DFP divestiture; accepts most behavioural remedies as modified (no DFP-AdX tie; no First Look or Last Look; deprecate UPR for all indirect transaction types; AdX bids to rival ad servers and Prebid; DFP-Prebid integration; data sharing; non-discriminatory bidding; buy-side rules); 6-year term; effect 60 days after entry of a judgment; implementation within 15 months | decided 2026-09-02; unsealed 2026-09-16; nothing in force | Google DFP and AdX; indirect transactions (open and private auctions) for open-web display; worldwide reach | remedies decision (decided) | [156] |
| United States (E.D. Va.) | [Proposed] Final Judgment Dkt 1861-1 (joint, with bracketed disputes) | Would bar First Look and Last Look for Indirect Demand (VI.I), make Header Bidding Trafficking available within 6 months, deprecate UPR and any uniform-floor requirement or fee discount (VI.J), require AdX-Prebid and AdX-rival-ad-server interoperability within [6][12] months (V.A-B), a DFP-Prebid server-to-server integration within 12 months (V.D) and a PG/preferred-deal routing connection within 15 months at publishers' option (V.E, V.H); expires on the sixth anniversary of the Effective Date | proposed 2026-10-02; Effective Date = 60 days after entry, tolled during stays | solely Indirect Demand (open and private auctions) for open-web display; excludes AdSense, AdMob, DV360 | proposed judgment (proposed) | [79] |
| United States (E.D. Va.) | Docket status, United States v. Google LLC 1:23-cv-00108 | No final judgment entered and no notice of appeal on the RECAP docket copy; last known filing 2 October 2026 | not in force (as of docket update 2026-10-29) | same | docket status (nothing ordered) | [80] |
| United States (S.D.N.Y. MDL 1:21-md-03010) | Opinion and Order on issue preclusion (Castel, J.), Dkt 146 in 1:23-cv-05177 | Precludes Google from relitigating the separate ad server and ad exchange markets, their worldwide scope, the DFP-AdX tie and the conduct of Unlawful Tying, First Look, Last Look, Dynamic Revenue Share and Unified Pricing Rules in the publisher movants' cases; otherwise denied | ordered 2025-10-27 (interlocutory) | publisher and advertiser plaintiffs in the MDL | interlocutory order (ordered) | [131] |
| United States (S.D.N.Y. MDL 1:21-md-03010) | Stipulation and Order Regarding Issue Preclusion (competitor plaintiffs), MDL entry filed in 1:26-cv-00140 as Dkt 80 | Extends the same preclusion to the competitor actions by stipulation, preserving appeal rights | ordered 2026-06-25 | OpenX, Magnite, PubMatic, Equativ/Sharethrough, Index Exchange, Kargo, Sovrn, TRUSTX actions | stipulated order (ordered) | [132] |
| United States (S.D.N.Y. MDL 1:21-md-03010) | Civil Case Management Plan and Scheduling Order (competitor actions) | Fact discovery due 10 November 2026; expert discovery due 30 April 2027; amended pleadings only with leave | ordered 2026-06-17 (entered 2026-06-18) | eight competitor actions listed in the order | scheduling order (ordered) | [132] |
| European Union | Commission Decision of 5 September 2025, AT.40670 Google - Adtech and Data-related practices (Art. 102 TFEU) | Fine EUR 2.95 billion; orders Google to end self-preferencing of AdX in DFP ad selection and in Google Ads/DV360 bidding and to cease its conflicts of interest along the ad-tech supply chain; 60 days to propose measures; preliminary view that only divestment of part of its services would suffice | decided 2025-09-05; non-confidential version published 2026-01-15; remedy for conflicts of interest not decided | EEA-wide publisher ad server and programmatic ad buying tool markets; AdX conduct | prohibition decision (decided; remedy pending) | [134] |
| European Union | Google's compliance plan in response to AT.40670 (as described on Google's blog) | Google proposes immediate product changes (publishers may set different minimum prices for different bidders in Google Ad Manager), conflict-of-interest and interoperability measures, and no break-up; says it will appeal | proposed 2025-11-14 | Google Ad Manager, Google Ads, DV360 in the EU | company proposal (proposed) | [135] |
| European Union | Commission answer E-000132/2026 (Executive Vice-President Ribera) | Commission 'has not reached a final conclusion' on a structural remedy; consulting market participants, including Proxistore, on Google's compliance plan | found 2026-03-12 (no decision) | AT.40670 remedy | regulator statement (found) | [137] |
| European Union (General Court) | Case T-794/25 Google and Alphabet v Commission (action for annulment) | Annulment action against the AT.40670 decision; lodged 20 November 2025; OJ application notice 12 January 2026; pending | pending | AT.40670 decision | court action (pending) | [136] |
| United Kingdom | CMA Competition Act 1998 investigation into Google's ad-tech conduct (case 51145 per the brief; header-bidding case 51134 combined March 2023) | Statement of objections issued 6 September 2024; timetable phase December 2025 to July 2026 (estimate) 'Considering next steps, whilst monitoring international developments'; no decision | open (page last updated 2026-05-19) | Google ad-tech conduct in the UK | investigation (open) | [138] |
| United Kingdom | CMA strategic market status designation: Google general search and search advertising | Designation 10 October 2025; publisher conduct requirement 3 June 2026; fair ranking and data portability conduct requirements 17 June 2026; user choice consultation 23 September 2026; no ad-tech requirement | in force (designation 2025-10-10; conduct requirements June 2026) | search and search advertising only | SMS designation (in force; no ad-tech scope) | [157] |
| United Kingdom | CMA strategic market status designation: Google mobile platform | Designation 22 October 2025 covering mobile operating systems, native app distribution and mobile browser and browser engine; final commitments 1 April 2026; steering conduct-requirement consultation 30 June 2026; no ad-tech requirement | in force (designation 2025-10-22) | mobile platform only | SMS designation (in force; no ad-tech scope) | [158] |
| United States (California) | CPPA regulations package: CCPA Updates, Insurance, Cybersecurity Audits, Risk Assessments and Automated Decisionmaking Technology (ADMT); Data Broker Registration and Accessible Deletion Mechanism (DROP) regulations | Adopted September 2025 (ADMT package) and November 2025 (DROP); CCPA regulations and data-broker/DROP regulations stated as effective 1 January 2026 | 2026-01-01 | businesses subject to the CCPA; registered data brokers | regulation (in force) | [159] |
| United States (federal) | FTC final consent order, In the Matter of Mobilewalla, Inc. | Bans collecting consumer data from online real-time bidding advertising exchanges for purposes other than participating in those auctions; bans use or sale of sensitive location data; Commission vote 4-1 | 2025-01-14 (final order) | one respondent (Mobilewalla) | consent order (ordered) | [160] |
| United States (California) | California Consumer Privacy Act (2018) | comprehensive consumer privacy law (rights, opt-outs, data-broker rules per the chart) | 2020-01-01 | state | statute (in force) | [85] |
| United States (Virginia) | Virginia Consumer Data Protection Act, SB 1392 (2021) | comprehensive consumer privacy law | 2023-01-01 | state | statute (in force) | [85] |
| United States (Colorado) | Colorado Privacy Act, SB 190 (2021) | comprehensive consumer privacy law | 2023-07-01 | state | statute (in force) | [85] |
| United States (Connecticut) | Connecticut Data Privacy Act, SB 6 (2022) | comprehensive consumer privacy law | 2023-07-01 | state | statute (in force) | [85] |
| United States (Utah) | Utah Consumer Privacy Act, SB 227 (2022) | comprehensive consumer privacy law | 2023-12-31 | state | statute (in force) | [85] |
| United States (Texas) | Texas Data Privacy and Security Act, HB 4 (2023) | comprehensive consumer privacy law | 2024-07-01 | state | statute (in force) | [85] |
| United States (Oregon) | Oregon Consumer Privacy Act, SB 619 (2023) | comprehensive consumer privacy law | 2024-07-01 | state | statute (in force) | [85] |
| United States (Montana) | Montana Consumer Data Privacy Act, SB 384 (2023) | comprehensive consumer privacy law | 2024-10-01 | state | statute (in force) | [85] |
| United States (Delaware) | Delaware Personal Data Privacy Act, HB 154 (2023) | comprehensive consumer privacy law | 2025-01-01 | state | statute (in force) | [85] |
| United States (Iowa) | Iowa Consumer Data Protection Act, SF 262 (2023) | comprehensive consumer privacy law | 2025-01-01 | state | statute (in force) | [85] |
| United States (Nebraska) | Nebraska Data Privacy Act, LB 1074 (2024) | comprehensive consumer privacy law | 2025-01-01 | state | statute (in force) | [85] |
| United States (New Hampshire) | New Hampshire SB 255 (2024) | comprehensive consumer privacy law (state name matched by the chart's alphabetical order) | 2025-01-01 | state | statute (in force) | [85] |
| United States (New Jersey) | New Jersey SB 332 (2024) | comprehensive consumer privacy law (state name matched by the chart's alphabetical order) | 2025-01-15 | state | statute (in force) | [85] |
| United States (Tennessee) | Tennessee Information Protection Act, HB 1181 (2023) | comprehensive consumer privacy law | 2025-07-01 | state | statute (in force) | [85] |
| United States (Minnesota) | Minnesota Consumer Data Privacy Act, HF 4757 (2024) | comprehensive consumer privacy law | 2025-07-31 | state | statute (in force) | [85] |
| United States (Maryland) | Maryland Online Data Privacy Act, SB 541 (2024) | comprehensive consumer privacy law | 2025-10-01 | state | statute (in force) | [85] |
| United States (Indiana) | Indiana Consumer Data Protection Act, SB 5 (2023) | comprehensive consumer privacy law | 2026-01-01 | state | statute (in force) | [85] |
| United States (Kentucky) | Kentucky Consumer Data Protection Act, HB 15 (2024) | comprehensive consumer privacy law | 2026-01-01 | state | statute (in force) | [85] |
| United States (Rhode Island) | Rhode Island Data Transparency and Privacy Protection Act, H 7787 (2024) | comprehensive consumer privacy law | 2026-01-01 | state | statute (in force) | [85] |
| European Union | Regulation (EU) 2016/679 (GDPR), Articles 4(1), 4(7) and 26(1) as interpreted in Case C-604/22 | A TC String is personal data where it can by reasonable means be associated with an identifier such as an IP address; a sectoral organisation proposing consent rules can be a joint controller for the string without extending automatically to participants' subsequent processing | judgment 2024-03-07 (GDPR in force earlier; application date not taken from an opened source) | EU-wide; TCF consent strings in OpenRTB | preliminary ruling (decided) | [161] |
| European Union (Belgium) | Belgian Market Court judgment on the APD's IAB Europe/TCF decision (as reported by IAB Europe) | Per IAB Europe: APD decision of February 2022 annulled; IAB Europe joint controller solely for the creation and use of TC Strings; action-plan execution remains suspended | 2025-05-14 (per the party's statement; judgment not opened) | IAB Europe and TCF participants | national court judgment (decided; party-reported) | [162] |
| European Union | IAB Europe TCF v2.3 | Industry consent framework; v2.3 launched April 2025; participants had until 28 February 2026 to adopt it | 2026-02-28 (adoption deadline) | TCF participants | industry framework (practice) | [163] |

### What the record changes, and what it does not

The record bears on Google's ad server and exchange, and it orders no change in Google's US conduct until a judgment is entered [79, 156, 78] [80]. For the H4 register its findings code as observed advantages, for Google's exchange alone ([chapter 4](#access)) [78] [79, 156, 78]. If a judgment is entered in the proposed form, publishers using DFP would regain floor control by demand source and gain Prebid routes [79]. Google's own documentation says its Ad Manager curation beta lets curators bundle Private Auctions and Preferred Deals into packages [24, 290]. On this paper's reading, a Private Auction package would fall under the First Look, Last Look and floor provisions, and a Preferred Deal package only under the optional route [79] [79].

The record says nothing about curated deals as such [79, 156, 78]. The three Virginia documents never mention curators, curated deals or third-party deal IDs [79, 156, 78]. The one use of "curate" is testimony that advertisers expect their demand-side platforms (DSPs) to "curate or prioritize one ad exchange over another", which is buyer-side supply-path selection ([chapter 8](#marketplaces)) [79, 156, 78]. The record sets no default for the priority, floors or exclusivity of curated deals on any exchange [79, 156, 78]. The privacy rules limit data collection and sale, and the one bid-stream order opened binds one company [85] [160].

### Coding Q2

Q2 is a registered coding question with no verdict. A court or regulator document settles each step, press summaries are context, and a step whose docket entry could not be opened stays unresolved. No US competition remedy is coded in force until a judgment is entered.

- **Found:** the Virginia liability findings of 17 April 2025 [78, 156]; preliminary objections by the Commission in June 2023 and the CMA on 6 September 2024 [134, 133] [138, 157, 158].
- **Decided:** the Virginia remedies of 2 September 2026 [156, 80]; the Commission's fine of 5 September 2025 [134, 133]; the ruling in C-604/22 [161].
- **Proposed:** the joint judgment of 2 October 2026 and Google's EU plan, which the Commission's answer of 12 March 2026 confirms [79] [135, 136] [137].
- **Ordered:** the New York preclusion orders and schedule, all interlocutory, and the Mobilewalla order [131, 132] [132] [160].
- **In force:** no US or UK ad-tech competition remedy; nineteen state privacy laws and California's 2026 rules. The Commission's decision is coded decided, and its remedy for the conflicts of interest is undecided [80] [138, 157, 158] [85] [159] [137].
- **Pending or open:** the nine private suits, T-794/25, the Commission's remedy and the CMA case [132] [135, 136] [137] [138, 157, 158].
- **Unresolved:** Virginia entries after 29 October 2026, the motion's outcome, the stipulation's list, the publisher complaints, the Commission's post-2019 finding, the Belgian judgment's text and three 2026 state effective dates [80] [132] [131, 132] [162, 163] [85] [83, 81].

With no judgment entered and no curator named in the three Virginia texts, the record establishes Google's own advantages and nothing about curated deals [80] [79, 156, 78].



## 10. Agents on the sell side


*Has a software agent made a sell-side selection decision in production use that someone other than the author documents? PubMatic claims 30 live agent campaigns by April 2026, and Magnite names one buy on LG's home screen. Only one account read says what the agents decided: PubMatic's release of 9 June 2026 says agents on its platform "autonomously selected CTV inventory" for Havas's Movistar campaign in Spain. Every account comes from the transacting parties and none is independently measured, so H7 codes as supported with qualification, on one platform-level record that names no specific agent.*

> **In short.**
> AdCP (the Ad Context Protocol) stood at v3.2.8, v3.1.31 and the pre-release v3.3.0-beta.3 at the cutoff, and its README lists Curation as "Coming soon", so curation has no schema [175] [164]. A seller choosing for a buyer is still representable, through the Media Buy walkthrough's seller-planned path [97, 176]. The AdCP registry lists 34 agents, test endpoints among them; a listing records an endpoint and says nothing of use [165]. The Prebid Sales Agent is alpha software targeting AdCP 3.1.1, and IAB Tech Lab's AAMP 3.0 (Agentic Advertising Management Protocols) carries OpenProposal as a draft [171, 178, 97] [170, 168, 179, 92]. Named deployments come from the vendors and from trade press relaying them: PubMatic claimed 30 live end-to-end agent campaigns in April 2026 and names customers in several releases, and Magnite names one buy [307] [172]. Of the accounts read, only PubMatic's release of 9 June 2026 says what the agents decided: for Havas's client Movistar, part of Telefónica, agents on PubMatic's platform "autonomously selected CTV inventory" in Spain [313]. The accounts of PubMatic's December 2025 campaign for Geloso Beverage Group and of H/L's human-approved buy through Magnite's Seller Agent document agent-led activation but not what the agents selected [93] [312] [172]. No result in the sources screened is measured by a party without a stake [93, 120, 172, 168, 171]. H7 codes as supported with qualification, and no deployment is a multi-publisher curated deal ID [93, 120, 172, 168, 171].

> **Note on interests.**
> The author's interests bear directly on this chapter. He co-leads the Signals & Measurement working group of AgenticAdvertising.org, which publishes AdCP, and maintains an open-source AdCP signals adaptor. No Fluff Advisory operates two agents in the AdCP registry, one typed signals and one sales; both are counted in the registry's 34 entries and excluded from the evidence [165]. Samba, his employer since May 2025 (TV measurement and ad data; owner of Semasio), appears as Samba TV on AdCP's member logo strip [165, 167]. He was executive vice president at Verve Group from September 2022 to August 2024, and earlier worked at Lotame and Hearts & Science. His essays on AdCP argue that a signals agent delivering segments to sales agents is a sell-side decisioning mechanism. Here that argument is a rival framing and never evidence, and no test he built counts as production use. None of these interests earns any evidence weight ([chapter 13](#method)).

### What the chapter counts as an agent

Of the six fields in the registered definition, feedback is the one no record describes; the most any gives is a success metric [172] [93] [120] [87]. The six are the decision, inputs, authority, permitted actions, execution location and feedback. Protocol support is recorded separately, per protocol and version, in five states: schema representation, implemented connector, demonstrated interoperability test, documented production use and demonstrated transaction. Membership and announcements earn no state ([chapter 13](#method)).

AdCP does not define a sales agent. Its documentation names roles by function: media buying agents "discover inventory and execute campaigns", signals agents "find and activate audiences", and orchestrators "make sure humans approve what matters" [97, 176]. "Sales agent" appears only in the phrase "sales-agent endpoints" and in the Prebid Sales Agent's name [97, 176], so the table codes each agent on what its own records say it does.

| Agent | Decision | Inputs | Authority | Permitted actions | Where it runs | Feedback |
|---|---|---|---|---|---|---|
| PubMatic's platform agents, for Butler/Till | Planned and implemented a CTV (connected TV), online and mobile buy | A brief entered through Butler/Till's enterprise Claude interface; adults 21 and over, geotargeted | Not stated | Plan and implement the campaign | PubMatic's Activate platform, called AgenticOS by March 2026 | Results claimed by the parties; no mechanism described [93] [77] |
| PubMatic's AgenticOS agents, for Havas | Selected CTV inventory, set frequency and brand-safety controls and optimised bids | A natural-language brief processed through Claude; the client's guardrails | Within the client's guardrails; no approval step stated | Select inventory, set controls, bid | PubMatic's Activate platform | Results claimed by the parties against their own targets [313] |
| Magnite's Seller Agent, with H/L's buyer agent | An agent-led media plan for LG Home Screen inventory | H/L's brief: NFL viewers in market for a car | LG and H/L approved the plan before launch | Set up and activate the buy | Inside Magnite's ClearLine, over MCP (Model Context Protocol) | A success metric, cost per reach; no figures [172] [87] |
| Chalice AI's models in OpenX custom decisioning | Value each video impression in real time | The impression's likelihood of driving a high-value action | Always on; no approval step stated | Bid | A container inside OpenX's infrastructure | Not described [120] |
| InMobi Buyer Hub | Inventory discovery and deal building in natural language | The trader's requests | "traders in control of every decision" | Discover inventory, build deals, optimise campaigns | Not stated | Not described [174, 165] |
| Stagwell's Media Machine | Buy-side planning and buying; decisions not itemised | Not stated | "humans firmly in control of every critical decision" | Connection to Stagwell Curate stated in the future tense | Connected to Google, Meta, Microsoft, TikTok and The Trade Desk platforms | Not described [183, 5] |
| Kargo's Project KERA | Turns a single campaign brief into an executed media plan | A campaign brief | From "full end-to-end execution" to "human-in-the-loop workflows" | Plan, create, activate and optimise | Not stated | Not described [173] |
| Apostra's seller agent | Proposes formats, combinations, sponsorships or bespoke options | The seller's business context | The seller keeps pricing, packaging and relationships | Propose to buyers | Apostra's platform, "built on AdCP" | Not described [54] |

The fields separate who sets a rule from where it runs. Chalice's models apply the advertiser's vendor's valuation inside OpenX, so the rule owner sits with the buyer and the execution host with the exchange [120]. Buyer Hub, the Media Machine and KERA are buyers' tools, with people in control or the product in closed beta [174, 165] [183, 5] [173]. Wherever authority is stated, people hold it, except for Chalice's always-on bidding agents and KERA's fully automated option: LG and H/L approved Magnite's agent-led plan, and Apostra's sellers keep pricing and packaging [172] [54] [120] [173].

### AdCP: sellers can plan, curation has no schema

AdCP can represent a seller choosing inventory for a buyer, and it has no schema for curation [164] [97, 176]. On 3 November 2026 its repository's latest tag on each line was v3.2.8 (28 October, marked Latest), v3.1.31 (30 October) and v3.3.0-beta.3 (31 October, a pre-release) [175]. The publisher calls AdCP 3.2 generally available, and its documentation names release 3.2.1 as that release while linking 3.2.8 schemas [167, 97].

The README's protocol surfaces include Media Buy, Signals and Trusted Match, and list Curation, "Media inventory curation", as "Coming soon" [164]. Selection by the seller is representable all the same. The Media Buy walkthrough offers two paths: the buyer selects and prices a seller's published products, or "The seller should curate, price, or negotiate a plan from the brief" [97, 176]. In the second path the seller's draft proposal is the media plan, which is the kind of decision H7 asks about [97, 176].

Signals sit in the same protocol [164]. In the 3.2.8 walkthrough a buyer activates a signal on DSPs (demand-side platforms) or "on the sales agent directly", which "records the activation internally" [88]. The page documents the route; no deployment of it was examined [88].

AgenticAdvertising.org's registry API returned 34 agents on 3 November 2026: 27 typed sales, 4 signals, and one each typed buying (Swivel), creative (Celtra) and unknown [165]. Some are test hosts by name: AdCP Test Agent, InMobi Exchange ADCP NonProd, goTom Sandbox and idm by briefable(demo) [165]. Six publisher storefronts, WebMD and TheWrap among them, run on Optimera's Prebid Seller Gateway, and InMobi lists a sales agent on a production-named host [165]. A self-registered listing reaches state 2 at most, and it records no compliance, volume or use [165]. It is no census either: PubMatic, Yahoo and Kargo appear on AdCP's member logo strip without a registry entry, and Magnite and Index Exchange have none [165, 167].

The publisher's own transaction claim is the oldest [167, 177, 166]. AgenticAdvertising.org says that "On October 16, 2025, the first-ever agent-to-agent media buy was executed", with "real money, real inventory from LG Ads" and human oversight [167, 177, 166]. AdExchanger's launch report, a day earlier, describes a hypothetical press demo and quotes PubMatic: "aspects of an overall campaign are being tested and executed" [167, 177, 166]. No record of the 16 October buy from any other party was found in the pages opened and searches logged [167, 177, 166]. The Scope3 post that search snippets cite for it now shows only Apostra placeholder text [89]. Magnite dates its own first seller-agent test to December 2025, with Scope3 as the buyer agent [90]. It quotes MiQ: "In December we set live one of the first AdCP buys working with Magnite's seller agent" [90]. Neither statement names a customer or spend, and neither claims an industry first [167, 177, 166] [90].

### The Prebid Sales Agent and AAMP: connectors and a draft

Neither the Prebid Sales Agent nor AAMP documents a production deployment in the sources screened by the cutoff [171, 178, 97] [92, 170, 179]. The Prebid repository labels its code "Status: alpha. A pre-1.0 implementation of a pre-release protocol (AdCP 3.1.1), under active development" [171, 178, 97]. Its package, adcp-sales-agent 2.0.0, pins the adcp==6.6.0 Python SDK (software development kit), and its latest commit on main is dated 28 October 2026 [171, 178, 97]. Its 3.1.1 target trails the 3.1 line, which had reached v3.1.31 [175] [171, 178, 97]. AdCP's documentation calls it "a community example", and docs.prebid.org gives it no version, status or deployment [171, 178, 97]. On this record it is an implemented connector, state 2 [171, 178, 97].

IAB Tech Lab announced AAMP 3.0 on 22 September 2026 with OpenProposal, "a new specification that defines how sellers can represent advertising products" [170, 168, 179, 92]. Public comment, first set to close on 22 October, now runs to 10 November 2026 [170, 168, 179, 92]. The OpenProposal repository says: "This is a draft. Nothing here is a ratified IAB Tech Lab standard. Do not build production systems against it expecting stability" [170, 168, 179, 92]. AAMP 2.3 was released on 30 July 2026 [170, 168, 179, 92]. AAMP 2.0 dates to April 2026, day unconfirmed: the month rests on SDK commits of 13 and 17 April, while the v2.0 GitHub releases are dated 8 July [91].

The reference code is a connector [169, 180, 181, 306]. The Seller Agent SDK, v2.4.2 of 18 August 2026, presents itself as a system "for publishers and SSPs to automate programmatic deal negotiation, booking, and distribution" (SSP, supply-side platform) [169, 180, 181, 306]. Its README has a curator role, and its line "Agent Range is pre-registered to curate deals" describes sample configuration and documents no deployment [169, 180, 181, 306]. The core orchestration repository, last pushed on 28 January 2026, has an MCP layer that "Returns mock responses" [169, 180, 181, 306]. Release notes label buyer-agent v2.1.0 "Verified end-to-end transactions" and v2.1.1 "Live-proven deal negotiation", with no counterparty named [91]. As IAB Tech Lab's own labels, they earn no state beyond a connector [91].

AAMP's production language names no adopter [92, 170, 179]. The standard page says AAMP 2.3 takes agentic buying "beyond demos and first live implementations" to capabilities in production stacks, and names none [92, 170, 179]. Its only named platform is Google Ad Manager, as the destination of a demo [92]. The 3.0 release's one pilot is IAB Australia's call for publishers to test workflows in a sandbox "before any live spend is involved" [92, 170, 179]. PMG's April 2026 release on integrating AAMP into its Alli operating system was not opened [92, 170, 179].

### Vendor agents: where the production claims are

PubMatic's is the earliest to name a customer [93, 120, 172, 168, 171]. Its release of 15 December 2025, read in a syndicated copy, announced an "AdCP-Enabled Agentic AI Campaign" with agency Butler/Till for Clubtails, a Geloso Beverage Group brand, "running through PubMatic's Activate platform" [77]. Marketing Dive reported on 17 March 2026 that the campaign "ran in December 2025" across CTV, online and mobile, geotargeted to adults 21 and over [93]. It says the campaign was "planned and implemented completely by artificial intelligence agents" on PubMatic's platform, by then named AgenticOS, from a brief entered through Butler/Till's enterprise Claude interface [93]. Its headline calls the campaign a test [93]. Citing "press materials shared with Marketing Dive", it reports buy-side supply chain costs reduced "by 5.5x", a multiple it does not explain [93] [93]. The same materials claim 40% more impressions than planned, 98% video completion and under 1% of inventory failing DoubleVerify standards [93]. None of these figures is independently verified, and the December release gave none [93]. Activate is PubMatic's own buying route, which this paper lists among the comparators ([chapter 8](#marketplaces)) [223, 20].

PubMatic's own releases go further. Its AgenticOS launch release of 5 January 2026 says that for Clubtails the platform "autonomously recommended campaign tactics, executed the buy, and optimized performance in real time within predefined parameters", without saying which inventory was chosen or by which agent [312]. On 27 April 2026 it claimed 30 live end-to-end agent campaigns in five countries and named Amnet's campaign for Interbev in France and Abovo Maxlead's CTV campaign in the Netherlands, with no dates or decisions [307]. On 9 June 2026 it announced a campaign with Havas for Movistar, part of Telefónica, in Spain [313]. The release says agents working across the platform "autonomously selected CTV inventory, applied frequency controls and brand safety parameters, and optimised bids in real-time", through Activate and on Spain's CTV publishers [313]. It reports a CPM 18% below target and impressions 23% above it, gives no campaign dates and quotes Havas and Telefónica; no other party measures the result [313]. It names no single agent among AgenticOS's more than 20, one of which handles inventory marketplace and curation [313]. The agent stream did not screen these releases; the final live sweep read them ([chapter 13](#method)).

Magnite's record moves from tests to a named buy [93, 120, 172, 168, 171]. Its blog of 6 January 2026 says its seller agent, embedded in SpringServe, lets publishers "engage in AdCP tests", with LG Ad Solutions among those testing [90] [172]. Its release of 16 September 2026 says partners "are using Magnite Orchestration" but names no executed transaction and no protocol [184]. On 14 October 2026 LG Ad Solutions and Magnite announced that LG "is making its Home Screen inventory available for agentic buying through Magnite's Seller Agent" [172]. Independent agency H/L was the first buyer, for the Northern California Toyota Dealers Association, in a campaign timed to the fall football season [172]. The release does not say what the seller agent decides [172]. AdExchanger's report the same day, built on interviews with H/L and Magnite, says H/L's buyer agent and Magnite's seller agent "set up and activated" the brief inside ClearLine over MCP [87]. LG and H/L approved the agent-led plan before it went live; the success metric was cost per reach, and no spend, impressions or dates were published [172] [87]. Neither account names AdCP, so the protocol link rests on Magnite's January blog, and LG's home screen is one publisher's inventory [172] [90].

OpenX and Chalice AI announced on 16 June 2026 that, for Hyundai, Chalice's models "operate as always-on bidding agents within OpenX's infrastructure" [120]. They run under OpenX's Custom Decisioning, also called containerisation, and value each video impression by its likelihood to drive a high-value action [120]. The vendors claim 20% lower cost per high-value action and 67% lower CPMs (cost per thousand impressions) [120]. The comparator is "the next-best alternative" in the body and "second-best-on-plan" in the sub-head, and no period, agency, sample, method or campaign geography is given [120]. [Chapter 7](#quality) treats these figures as vendor claims [119, 120, 121, 125].

The rest stop short of a named sell-side decision [93, 120, 172, 168, 171]. InMobi opened Buyer Hub to agencies on 31 August 2026, after opening it to curators, and names no customer, exchange or DSP [174, 165]. Stagwell launched the Media Machine on 19 June 2026 with "over 20 intelligent agents" and no client named [183, 5]. A July release says its agents "will connect directly to Stagwell Curate" ([chapter 8](#marketplaces)) [183, 5]. Kargo announced Project KERA, a buying and creative engine, in closed beta with Hershey's among its test partners, no protocol and no results [173]. Apostra, formerly Scope3, says "Today, campaigns are live on Apostra across markets around the world", naming six markets and no customer, publisher or transaction [54]. Microsoft's record is unresolved [185, 186]. This review read its May 2025 post as positioning, with no agent executing transactions; the verifier did not reopen it, and a September 2026 Curate post was read only by its listing summary [185, 186]. No Index Exchange agent announcement was found in the searches logged [165, 167].

The figure sets each protocol and vendor interface against the five states. A cell reads "reported" only where a party other than the publisher or vendor names the deployment. Two cells do: PubMatic's campaign, which the vendors' own release calls AdCP-enabled, and Magnite's LG transaction [93] [172] [77] [87]. PubMatic's later releases are its own, so they add no "reported" cell [313].

![Named agent deployments rest on vendor and trade-press accounts; no protocol state is independently observed.](figures/fig-protocol-states.svg)

*Figure fig-protocol-states. **Named agent deployments rest on vendor and trade-press accounts; no protocol state is independently observed.** Each protocol and vendor agent interface found, against the five protocol states of shared rule 12: schema representation, implemented connector, demonstrated interoperability test, documented production use and demonstrated transaction. A cell is "reported" only where a party other than the publisher or vendor names the deployment. Units: state. Scope: AdCP (tags to 3.3.0-beta.3), AAMP 2.x and the 3.0 draft, the Prebid Sales Agent, and vendor agent interfaces. Sources [164], [165], [166], [93], [167], [168], [169], [170], [92], [171], [172], [120], [54], [173], [174], [5]. Caveat: The author co-leads AdCP’s Signals & Measurement working group and maintains an open-source signals adaptor; neither earns a state. Production claims rest on vendor releases and trade press, not audit. Data: data/figure-data.json#protocols*

The instrument lists each observable screened, one row per observable. In it the Prebid Sales Agent shows as beta, because its status vocabulary has no "alpha" [171, 178, 97]. Its PubMatic row uses AgenticOS, the March 2026 name, for a campaign the December 2025 release placed on Activate [77]. Its note that the AdCP wording sits in an unopened release predates the verifier's read of a syndicated copy [93] [77]. Its AAMP 2.0 row cites the standard page, which gives no date; the April month rests on the SDK commits [91].

**AI and agentic capability map**

| Platform / product | Capability | Status | Date | Independently observable? | What the record supports | Src |
|---|---|---|---|---|---|---|
| AdCP | Release tag v3.2.8 (3.2 line, marked Latest) | live | 2026-10-28 | GitHub release tag | schema representation of Media Buy, Signals, Trusted Match surfaces | [175] |
| AdCP | Release tag v3.1.31 (3.1 maintenance line) | live | 2026-10-30 | GitHub release tag |  | [175] |
| AdCP | Release tag v3.3.0-beta.3 | beta | 2026-10-31 | GitHub release tag |  | [175] |
| AdCP | Media Buy surface (get_products, create_media_buy, get_media_buy_delivery) | live | 2026-10-28 | README table | schema representation only | [164] |
| AdCP | Signals surface (get_signals, activate_signal) | live | 2026-10-28 | README table | declared author interest (Signals WG co-lead) | [164] |
| AdCP | Curation surface (media inventory curation) | announced | 2026-11-03 | README table row | no schema representation at the cutoff | [164] |
| AdCP | Seller-planned path: seller curates, prices or negotiates a plan from the brief | live | 2026-11-03 | docs Media Buy page | represents a seller-side selection decision in a proposal | [176] |
| AdCP | 3.2 general availability statement | live | 2026-11-03 | homepage banner and docs intro | no GA date given; docs release 3.2.1 lags tag 3.2.8 | [97] |
| AdCP (AgenticAdvertising.org) | First agent-to-agent media buy, real money, LG Ads inventory | unconfirmed | 2025-10-16 | publisher's homepage claim | no third-party record opened; no buyer, agency or spend named | [167] |
| AdCP (AgenticAdvertising.org) | Membership count | live | 2026-11-03 | homepage | membership earns no capability credit | [177] |
| AdCP registry | Agents listed, total | live | 2026-11-03 | registry API count field | includes test, demo, sandbox and non-prod endpoints and two author entries | [165] |
| AdCP registry | Agents typed sales | live | 2026-11-03 | registry API, 27 entries | derived count | [165] |
| AdCP registry | Agents typed signals | live | 2026-11-03 | registry API, 4 entries | one of the four is the author's (excluded) | [165] |
| AdCP registry | Agents typed buying (Swivel), creative (Celtra), unknown | live | 2026-11-03 | registry API, 1 + 1 + 1 entries |  | [165] |
| InMobi Exchange | InMobi Sales Agent, production-named host (an InMobi production endpoint) and NonProd twin | live | 2026-11-03 | registry entries | endpoint only; no production use stated | [165] |
| Equativ | AdCP sales agent endpoint (Equativ's discover endpoint) | live | 2026-11-03 | registry entry | endpoint only | [165] |
| Optimera Prebid Seller Gateway | Hosted sales agents for WebMD, TheWrap, Bustle Digital Group, Carpenter Media Group, Trusted Media Brands, Mediatonik | live | 2026-11-03 | registry entries (6 publishers + gateway) | relationship to prebid/salesagent code not established | [165] |
| Prebid Sales Agent | AdCP Media Buy sales agent; AdCP 3.1.1 via adcp==6.6.0; package 2.0.0 | beta | 2026-10-28 | README, pyproject, latest commit 1c202df2 | open-source connector; no named deployment; the README says alpha, which this instrument folds into beta (verifier correction) | [171] |
| Prebid Sales Agent | docs.prebid.org page | unconfirmed | 2026-11-03 | docs page | AdCP docs: community example, not a maintained reference implementation | [178] |
| AAMP 3.0 | OpenProposal object model specification | announced | 2026-09-22 | press release; repo tag v3.0-draft-1 (2026-09-24) | public comment extended to 2026-11-10 | [170] |
| AAMP 2.3 | Release | live | 2026-07-30 | press-release index title and date | release text not opened | [179] |
| AAMP 2.0 | Release (added 5 new transaction types) | live | 2026-04 | standard page | April 2026 per the v2.0 tagged commits (13 April buyer agent, 17 April seller agent); the v2.0 GitHub releases are dated 8 July 2026 and the standard page gives no date (verifier correction) | [169, 180] |
| AAMP | Seller Agent SDK reference implementation v2.4.2 | live | 2026-08-18 | GitHub release; last commit 2026-10-28 | README documents a curator role and FreeWheel ad-server mode | [169] |
| AAMP | Buyer Agent SDK reference implementation v2.4.2 | live | 2026-08-18 | GitHub release; last commit 2026-10-28 |  | [180] |
| AAMP | Core Orchestration (agentic-direct), A2A/MCP for OpenDirect 2.1 | unconfirmed | 2026-01-28 | README; last push | no releases; untouched since January 2026 | [181] |
| AAMP 3.0 | IAB Australia seller-side pilot in a sandbox | beta | 2026-09-22 | press release quote | expressions of interest opened | [170] |
| AAMP 2.3 | Capabilities integrated into production media buying stacks | unconfirmed | 2026-10-27 | standard page | no adopter named | [92] |
| AAMP | PMG integrates AAMP standards into Alli operating system | announced | 2026-04-07 | press-release index | release not opened | [179] |
| IAB Tech Lab Agent Registry | Registry of agents (tools portal) | unconfirmed | 2026-11-03 | the IAB Tech Lab agent registry (tools.iabtechlab.com) | JavaScript application; contents unread | [182] |
| OpenX custom decisioning (OpenXBuild) + Chalice AI | Always-on bidding agents inside OpenX for Hyundai, impression-level valuation | live | 2026-06-16 | press release | customer named; period, agency and method absent; 20% / 67% are vendor claims | [120] |
| InMobi Buyer Hub | Natural-language inventory discovery and deal building; agency access | live | 2026-08-31 | product page | no named deployment, exchange or US statement; earlier opened to curators | [174] |
| Stagwell The Media Machine | Agentic media operating system, over 20 agents | live | 2026-06-19 | press release | no client or pilot named; humans in control of every critical decision | [183] |
| Stagwell Curate | Buying agents connecting to the curated marketplace | announced | 2026-07-16 | press release | marketplace itself 'available immediately' | [5] |
| Magnite Orchestration / Seller Agent | Partners using Orchestration; audio, high-attention display, TV extensions | live | 2026-09-16 | press release | no executed transaction named; AdCP/AAMP not named | [184] |
| Magnite Seller Agent + LG Ad Solutions Home Screen | H/L activates LG Home Screen inventory for the Northern California Toyota Dealers Association | live | 2026-10-14 | press release | no protocol named; decision not described; single-publisher inventory | [172] |
| Magnite Showcase | Packaging of premium inventory for discovery by buyer agents (per fact-check note) | unconfirmed | 2026-11-03 | three searches; 16 September 2026 release does not mention it | product name unverified | [184] |
| Kargo Project KERA | Agentic media buying and creative engine, closed beta; Hershey's a beta partner | beta | 2026-04-07 | reproduced release (ExchangeWire PressBox) | beta start date and other partners unconfirmed; no protocol named | [173] |
| Apostra (formerly Scope3) | Campaigns live on Apostra, built on AdCP | live | 2026-09-23 | company blog post | no customer, volume or transaction named | [54] |
| PubMatic AgenticOS + Butler/Till (Geloso Beverage Group) | Campaign planned and implemented by AI agents on the SSP's platform; CTV, online, mobile | live | 2025-12 | Marketing Dive report of press materials | customer, period, decision and context named by the transacting parties; the vendors' 15 December 2025 release (syndicated copy) places the campaign on PubMatic's Activate and calls it AdCP-enabled; figures unverified (verifier correction) | [93] |
| Microsoft Advertising | Agentic positioning; Invest media buying to end 28 February 2026; Curate continuing | announced | 2025-05-14 | blog post | no agent executing transactions described; no AdCP/AAMP connector found | [185] |
| Microsoft Curate | Post 'The Power of Curation' (programmatic curation) | unconfirmed | 2026-09-10 | author page listing | full post not opened | [186] |
| Index Exchange | Agent announcement or registry entry | unconfirmed | 2026-11-03 | registry (no entry); four searches | absence bounded to searches logged | [165] |
| AdExchanger (AdCP launch report) | Partial tests, human in the loop | announced | 2025-10-15 | trade report | written the day before the claimed first buy | [166] |

### Coding H7

H7 is supported, with qualification. The registered rule codes it supported when production evidence from a party other than the author identifies a customer or deployment, a period, an executed sell-side selection decision and an operational transaction context. It codes contradicted when every named deployment is documented as withdrawn or as never having executed a decision. It codes unresolved when only announcements, demos, schema support or author-built tests exist in the sources screened.

| Record | Customer | Period | Executed sell-side decision | Transaction context | Who says so | Independently measured? |
|---|---|---|---|---|---|---|
| PubMatic's AgenticOS agents, Havas | Movistar (Telefónica), through Havas | By 9 June 2026; dates not given | Agents "autonomously selected CTV inventory" and set controls; no agent or rule named | Live CTV activation on Activate across Spanish CTV publishers | PubMatic, quoting Havas and Telefónica | No [313] |
| PubMatic's platform agents, Butler/Till | Geloso Beverage Group (Clubtails) | December 2025 | Agents "recommended campaign tactics, executed the buy" and optimised; what they selected is not stated | Live CTV, online and mobile campaign on Activate | PubMatic and Butler/Till; Marketing Dive relays their press materials | No [93] [312] [77] |
| Magnite's Seller Agent, H/L | Northern California Toyota Dealers Association, through H/L | Fall 2026 football season; no dates | Seller and buyer agents set up and activated an agent-led plan, approved by LG and H/L; what the seller agent selected is not stated | LG Home Screen inventory in ClearLine | Magnite and LG; AdExchanger interviews H/L and Magnite | No [172] [87] |
| Chalice AI in OpenX | Hyundai | Not stated | Impression-level valuation and bidding | Curated video activation on OpenX | OpenX and Chalice AI | No; comparator undefined [120] |
| AgenticAdvertising.org's first buy | Not named | 16 October 2025 | Not described | LG Ads inventory, "real money" | The protocol's publisher | No [167, 177, 166] |
| Magnite and MiQ AdCP tests | Not named | December 2025 | Not described | Seller agent in SpringServe; Scope3 as buyer agent in the first test | Magnite, quoting MiQ | No [90] |
| Other AgenticOS campaigns | Interbev, through Amnet; Abovo Maxlead's client not named | Not stated | Not described | France and the Netherlands | PubMatic | No [307] |

One record meets all four elements on its face: PubMatic's Havas release names a customer, is dated within the period, states an executed selection and places it in a live CTV activation, and it comes from a party other than the author, which is what the rule asks [313] [93, 120, 172, 168, 171]. Its selection claim is made for the platform's agents as a group, so it shows that selection happened on the platform, not which agent made it or by which rule [313]. The Geloso and LG records name a customer, a period and a context and attest agent-led activation; neither says what an agent selected, so on the rule's own words they fall short of an executed selection decision [93] [312] [172]. Agent-led activation dates to December 2025, and the earliest documented selection to a release of 9 June 2026 [93] [313]. Over the period the record changes in kind: by October 2026 a product named as a seller agent transacts with a buyer agent, under human approval [172].

Every account starts with the transacting parties: PubMatic's releases are its own, Marketing Dive relays press materials, and AdExchanger interviews the agency and the exchange [313] [93] [87]. No result is measured by a party without a stake in the sources screened by the cutoff [93, 120, 172, 168, 171]. No record publishes the rule by which an agent selected, so even the Havas decision is attested and its content unknown [313]. The PubMatic records do not separate the agency's Claude brief from the choices of the agents on PubMatic's platform, and AgenticOS runs through PubMatic's own buying platform, so its selection happens on the supply side for one buyer, not as a deal ID sold to many [313] [93]. No deployment is documented as a multi-publisher curated deal ID, and LG's home screen is one publisher's inventory [93, 120, 172, 168, 171] [172].

The other records fall short. OpenX's names a customer, a decision and a context, and no period [120]. PubMatic's April count and its Amnet and Abovo Maxlead names give no period or decision [307]. The publisher's 16 October 2025 buy names no buyer, agency or spend, so it does not count [167, 177, 166]. Magnite's December 2025 tests name no customer either [90]. No named deployment is documented as withdrawn in the sources screened by the cutoff, so the contradicted rule is not met [93, 120, 172, 168, 171]. The rule asks only for a party other than the author, and vendors are such parties; read to require a party without a stake, the same evidence would code unresolved [93, 120, 172, 168, 171]. A record from a party without a stake, showing what the agent selected and how delivery was measured, would remove the qualification ([chapter 12](#next)).



## 11. The Deal Quotient


*How much of what a curation product does can a buyer establish from its public record, and which products document the most? Among exchange curation platforms, 4 products sit in the Strong band, 3 in Mid and none in Front rank; curated deal products and managed client marketplaces have too few well-documented rows to carry bands. Every band changes under at least one of the two fills tested for the missing cells, so a band describes what the research found in public by 3 November 2026. Only 7 of the 21 scored products carry a band, and three in four counted cells rest on vendor marketing or announcements [314].*

> **In short.**
> The Deal Quotient (DQ) summarises how much of a curation product's capability is documented in public, scored within its cohort; it says nothing about delivery, price or business quality. 21 products entered the matrix and 7 fell below the admission rule of four counted cells. Only exchange curation platforms have enough band-eligible rows to carry bands: 4 are Strong, 3 Mid and none Front rank, and the rest of that cohort are thin records. The field DQ is 47 for exchange curation platforms, 41 for curated deal products and 34 for managed client marketplaces, and the three cannot be compared. No readable vendor page the sweep opened discloses a periodised curated-deal volume, no row has a counted supply-path cell, and every quantified quality claim found is a vendor assertion [15, 256, 223, 266] [119, 120, 121, 125]. Every banded row changes band when its missing cells are filled with 1, with 4 or with either, so the bands describe the public record on 3 November 2026. Removing Lotame, the one admitted interest-marked row, moves the curated-deal field DQ by under half a point. Of the 122 counted cells, 92 rest on grade-D evidence, and on grade-C evidence or better only four products would be admitted and no cohort would carry bands [314]. One scoring pass set every score, and no independent second scorer checked it.

> **Note on interests.**
> The author has worked since May 2025 at Samba (TV measurement and advertising data), which owns Semasio. He was executive vice president at Verve Group from September 2022 to August 2024, and worked earlier at Lotame and at Hearts & Science. He co-leads the Signals & Measurement working group of AgenticAdvertising.org (AdCP, the Ad Context Protocol) and maintains an open-source AdCP signals adaptor, and No Fluff Advisory operates two agents listed in the AdCP registry. None of these earns any evidence weight. Lotame's Curated Marketplaces is the only interest-marked row admitted to the matrix. Verve's Brand+ Marketplace and Semasio's curated deals were excluded by the admission rule, not by choice: each had fewer than four counted cells. Samba's PMPs (private marketplace deals) are listed without a score, because Samba's site mentions them in one homepage sentence that names no platform, deal ID, channel or fee [268]. Samba and Lotame are among the named data partners of Magnite's ClearLine Activate [108] and among the launch vendors of Index's data ecosystem for Index Marketplaces [188]. Interest marks never change a score.

### What the quotient is

The DQ summarises the publicly documented capability of each scored product within its cohort, drawn from the capability matrix: 12 dimensions scored 0 to 5 against anchors fixed before scoring ([chapter 13](#method)). It does not measure how a deal performs, what it costs a buyer or how sound the operator's business is. A product with a thin public record scores low whatever its engineering.

The DQ follows the construction of the author's Contextual Quotient (CQ), with the same Depth, Reach and Standing structure, shrinkage and band offsets. That makes the two read alike; it does not validate the DQ. One part differs. The DQ's field mean, F, is the arithmetic mean of the Raw scores of the cohort's admitted rows. The author's Contextual Quotient and Native Quotient (NQ) pooled group scores instead. A DQ therefore cannot be set against a CQ or an NQ, and a DQ in one cohort cannot be set against a DQ in another.

### How it is built

Each row's counted cells are averaged within three groups. Depth covers what the deal is: selection transparency, composition disclosure, data provenance, quality evidence and, for exchange curation platforms only, auction mechanics. Reach covers where it works: channel execution, activation routes, scale and US availability. Standing covers whether a buyer can trust and trade it: pricing transparency, independent assurance and supply-path integrity. Raw is 0.45 × Depth + 0.25 × Reach + 0.30 × Standing, renormalised over the groups present. Cells marked n/e (no evidence) or n/a (outside the cohort's mask) are skipped; none counts as zero. Only cells a verifier checked count.

A sum would punish a thin record by counting unscored dimensions as zero, and a plain average would reward missing evidence. The DQ instead pulls Raw toward the field mean in proportion to how thin the row is: Shrunk = (k × Raw + K × F) / (k + K), where k is the row's number of counted cells and K, the prior strength, is 4. The DQ is 100 × Shrunk / 5. K is a heuristic set by the author: the dimensions are different capabilities, so no statistical estimate of it exists, and the tests below rerun the index with other values.

Bands come from the unrounded offset 20 × (Shrunk − F), which equals the distance in DQ points from the unrounded field DQ. Front rank starts at +12, Strong at +1 and Mid at −8, with Narrow below. Three thresholds decide who appears and who is banded. A product enters the matrix with at least four counted cells. A row is band-eligible with at least 6 counted cells and at least one in every group its cohort's mask applies. A cohort shows bands only if 6 or more of its rows are eligible. Because of the shrinkage, a product scoring 5 on every applicable dimension would reach 87 among exchange curation platforms, 84 among curated deal products and 82 among managed client marketplaces; the ceiling sits below 100.

### Who entered the matrix

21 products entered: 11 exchange curation platforms, 7 curated deal products and 3 managed client marketplaces. On the vendor sweep's reading of their own pages, the register moved YMax, whose operator-built packages activate through the buyer's DSP (demand-side platform) as standard deal IDs, to curated deal products [11]. It moved Seedtag's NeuroX Curation, a self-serve interface to Seedtag's own exchange, to exchange curation platforms [128].

7 products fell below the admission rule, each with fewer than four counted cells. Among exchange products, InMobi's Buyer Hub page names no exchange, DSP or deal-ID mechanism [174]. Sovrn's buyer page lists curated packages under an offer heading, with no named platform [277]. The sweep read Nexxen's Curated Marketplace page as stating no channel, DSP, fee, scale figure, status or geography; its verifier did not re-open the page, so that reading is unverified [254]. Verve's Brand+ Marketplace page was also read by the sweep and not re-checked [21]. Among curated deal products, Experian's curation page names no channel, fee or figure and no specific platform [255]. Dstillery's deal-ID FAQ is dated 25 July 2022 but names Microsoft Curate, so its current status cannot be settled from the page [61]. On the sweep's unverified reading, Semasio's homepage lists a HypertailPMP service with no deal-ID mechanism, channel or fee [270]. An exclusion records a public record too thin to score and says nothing about how the product performs.

### The results

Only exchange curation platforms carry bands, and no product in any cohort reaches Front rank.

![Read the band, not the rank: no product clears the front rank line, and 14 rows are not banded.](figures/fig-quotient.svg)

*Figure fig-quotient. **Read the band, not the rank: no product clears the front rank line, and 14 rows are not banded.** The Deal Quotient for the products with enough evidence to band, one section per cohort. The Deal Quotient follows the construction of the author’s Contextual Quotient, so the two read the same way; that does not validate it. Depth, Reach, Standing group means are combined 45/25/30 and shrunk toward the cohort’s field mean (the mean Raw score of its admitted rows) with a prior strength of 4; bands come from the unrounded distance to that field score, and only verified or qualified cells count. A band needs at least 6 counted cells with one in every applicable group, in a cohort with at least 6 such rows. Across the reweightings, positions move by up to 4 places within a cohort, and 4 of the 7 banded products keep their band under every weighting. DQ measures publicly documented capability within a cohort, not product performance. Units: DQ points on a 0–100 scale, pulled toward the cohort field mean; shown rounded, plotted unrounded. Scope: 7 banded products of 21 in the capability matrix, scores as of 3 November 2026. Caveat: Single scorer. The index inherits every limit of the dimension scores and is sensitive to the weights; the bands, not the positions, are the finding. Data: data/dq.csv*

The field DQ of exchange curation platforms is 47. Strong starts at 47.7, Front rank at 58.7 and Mid at 38.7. 7 of the cohort's 11 rows are band-eligible, and their DQs run from 41 to 52. Maestro by Equativ, the highest, sits about seven points below the Front rank line.

| Product | DQ | Band | Counted cells | Band with n/e filled at 1 / at 4 | Where the record is thin |
|---|---|---|---|---|---|
| Maestro by Equativ | 52 | Strong | 6 of 12 | Mid / Strong | The curation page states no DSP, fee amount or US availability, and its volume figures carry no period [15] |
| PubMatic Connect | 51 | Strong | 10 of 12 | Mid / Strong | PubMatic's own Activate is the only buying endpoint named for curated deals [223, 222, 119, 12] |
| Commerce Grid (Criteo) | 51 | Strong | 6 of 12 | Mid / Strong | Two DSPs have documented deal sync; the demand figures on the product page are company-wide [56, 16, 266, 188] [266] |
| Index Marketplaces | 48 | Strong | 10 of 12 | Mid / Strong | No buyer-facing list of the sellers or domains in a package is documented [198, 200, 187] |
| Microsoft Curate | 47 | Mid | 6 of 12 | Mid / Strong | No buyer-facing list of sellers or domains; the deal name is the only buyer-facing element documented [110, 111] |
| ClearLine (Magnite) | 44 | Mid | 8 of 12 | Mid / Strong | Auction type, priority, floors, exclusivity and sampling are unknown, and no charge is disclosed [107, 213, 214] [212, 62] |
| TL Direct (TripleLift) | 41 | Mid | 6 of 12 | Narrow / Strong | "No TripleLift curation fees" is stated with nothing on TripleLift's exchange take, and no DSP is named as a route [117] [130] |

Vendor documentation and vendor marketing carry most of these cells, capping them at 4 and 3 respectively. The matrix holds no 5. The only 4s are Index's channel execution and Microsoft Curate's auction mechanics, where Microsoft documents every curated deal as an Open Auction that never overrides seller floors [110, 111].

The rest of the cohort are thin records, each with four counted cells and a missing group. Google Ad Manager Curation has the highest DQ in the cohort, 53, and its missing group is Reach; its help page still marked the curation settings Beta on 3 November 2026 [59]. Seedtag's NeuroX Curation, at 41, also lacks a Reach cell. OpenXSelect (49) and FreeWheel Curation Hub (38) have no counted Standing cell, and the pages opened state no fee for either [17, 10] [16, 260].

Curated deal products have 2 band-eligible rows of 7, and managed client marketplaces 1 of 3, so neither cohort shows bands. Their eligible rows read "Cohort too small", the others "Thin record". Curated deal products span 39 (Comscore-Certified Deal IDs) to 43 (Adelaide's high-attention PMPs and Integral Ad Science's IAS Curation), a range in which single cells decide the order. Among managed client marketplaces, the highest DQ, 40, belongs to WPP Media's Premium Marketplace, whose newest page found dates from 9 February 2023 and does not say whether it is still sold [18]. Stagwell Curate, announced on 16 July 2026 as available immediately to Stagwell's clients, scores 36 and is the cohort's only eligible row [5]. dentsu Curate scores 26 on a 2020 release about Asia-Pacific markets [6].

**The capability matrix** (0–5 evidence-bounded scores; n/e = insufficient evidence; n/a = outside the cohort’s mask; unverified cells are marked and counted as n/e; availability, evidence grade and verification status per cell are in data/capability-matrix-long.csv; one section per cohort; banded products ordered by DQ, the Deal Quotient, within their cohort, and unbanded rows listed A to Z)

*A — Exchange curation platforms*

| Vendor | Product | Group | DQ (band) | Selection | Composition | Provenance | Quality | Auction | Channels | Routes | Scale | US | Pricing | Assurance | Supply path | Counted |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Equativ | Maestro by Equativ | exchange | 52 (Strong, provisional: Mid with n/e = 1, Strong with n/e = 4) | 3 | 3 | n/e | 1 | 3 | n/e | 3 | n/e | n/e | 3 | n/e | n/e | 6/12 |
| PubMatic Connect | PubMatic Connect | exchange | 51 (Strong, provisional: Mid with n/e = 1, Strong with n/e = 4) | 3 | 3 | 3 | 2 | 2 | 3 | 1 | n/e | 3 | 3 | 3 | n/e | 10/12 |
| Criteo | Commerce Grid | exchange | 51 (Strong, provisional: Mid with n/e = 1, Strong with n/e = 4) | 3 | 1 | 3 | n/e | 2 | n/e | 3 | n/e | n/e | 3 | n/e | n/e | 6/12 |
| Index Marketplaces | Index Marketplaces | exchange | 48 (Strong, provisional: Mid with n/e = 1, Strong with n/e = 4) | 3 | 1 | 3 | 1 | 3 | 4 | 3 | n/e | 3 | 3 | 1 (unconfirmed) | n/e | 10/12 |
| Microsoft Curate | Microsoft Curate | exchange | 47 (Mid, provisional: Mid with n/e = 1, Strong with n/e = 4) | 3 | n/e | 3 | 1 | 4 | n/e | 1 | n/e | n/e | 3 | n/e | n/e | 6/12 |
| Magnite ClearLine | ClearLine | exchange | 44 (Mid, provisional: Mid with n/e = 1, Strong with n/e = 4) | 3 (beta) | n/e | 3 | n/e | 1 (beta) | 2 | 1 | n/e | 3 | 1 | 3 | n/e | 8/12 |
| TripleLift | TL Direct | exchange | 41 (Mid, provisional: Narrow with n/e = 1, Strong with n/e = 4) | 3 | n/e | 3 | n/e | 1 | 3 | n/e | n/e | 1 | 1 | n/e | n/e | 6/12 |
| FreeWheel (Comcast) | FreeWheel Curation Hub | exchange | 38 (thin record · no standing) | 3 | n/e | 1 | n/e | 1 | n/e | 1 | n/e | n/e | n/e | n/e | n/e | 4/12 |
| Google | Google Ad Manager Curation | platform | 53 (thin record · no reach) | 3 (beta) | n/e | 3 (beta) | n/e | 3 (beta) | n/e | n/e | n/e | n/e | 3 (beta) | n/e | n/e | 4/12 |
| OpenX | OpenXSelect | exchange | 49 (thin record · no standing) | 3 | 3 | 1 | n/e | n/e | n/e | n/e | n/e | 3 | n/e | n/e | n/e | 4/12 |
| Seedtag | Seedtag NeuroX Curation | exchange | 41 (thin record · no reach) | 3 | n/e | 3 | 1 | n/e | n/e | n/e | n/e | n/e | 1 | n/e | n/e | 4/12 |

*B — Curated deal products*

| Vendor | Product | Group | DQ (band) | Selection | Composition | Provenance | Quality | Auction | Channels | Routes | Scale | US | Pricing | Assurance | Supply path | Counted |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Adelaide | Adelaide High-Attention PMPs | verification | 43 (thin record) | 3 | n/e | 3 | 2 | n/a | n/e | 3 | n/e | n/e | 1 | n/e | n/e | 5/11 |
| Audigent SmartPMP | SmartPMP | data | 40 (cohort too small) | 3 | n/e | 3 | 1 | n/a | n/e | 3 | n/e | 1 (unconfirmed) | 2 | 1 (unconfirmed) | n/e | 7/11 |
| Comscore | Comscore-Certified Deal IDs | verification | 39 (thin record · no reach) | 3 (unconfirmed) | 3 (unconfirmed) | 3 (unconfirmed) | 1 (unconfirmed) | n/a | n/e | n/e | n/e | n/e | n/e | 1 (unconfirmed) | n/e | 5/11 |
| Integral Ad Science | IAS Curation | verification | 43 (thin record · no standing) | 3 | n/e | 3 | 1 | n/a | n/e | 2 (beta) | n/e | n/e | n/e | n/e | n/e | 4/11 |
| Lotame Curated Marketplaces | Lotame Curated Marketplaces | data | 40 (cohort too small) | 3 | 1 | 3 | 2 | n/a | n/e | 2 (beta) | n/e | 3 (unconfirmed) | 1 | 1 (unconfirmed) | n/e | 8/11 |
| Multilocal | Multilocal curation | data | 41 (thin record · no standing) | 3 | n/e | 3 | 1 | n/a | n/e | 2 (beta) | n/e | 1 | n/e | n/e | n/e | 5/11 |
| Yieldmo | YMax | exchange | 42 (thin record · no standing) | 3 | 3 | 3 | 2 | n/a | n/e | 1 | n/e | n/e | n/e | n/e | n/e | 5/11 |

*C — Managed client marketplaces*

| Vendor | Product | Group | DQ (band) | Selection | Composition | Provenance | Quality | Auction | Channels | Routes | Scale | US | Pricing | Assurance | Supply path | Counted |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| dentsu | dentsu Curate | agency | 26 (thin record · no reach) | 1 (unconfirmed) | 1 (unconfirmed) | 1 (unconfirmed) | 1 (unconfirmed) | n/a | n/e | n/e | n/e | n/e | 1 (unconfirmed) | n/e | n/e | 5/11 |
| Stagwell Curate | Stagwell Curate | agency | 36 (cohort too small) | 3 | 1 (unconfirmed) | 3 | n/e | n/a | 3 | n/e | n/e | 1 | 1 | n/e | n/e | 6/11 |
| WPP Media | WPP Media Premium Marketplace | agency | 40 (thin record) | 1 (unconfirmed) | 3 (unconfirmed) | n/e | n/e | n/a | n/e | n/e | n/e | 3 (unconfirmed) | 2 (unconfirmed) | n/e | n/e | 4/11 |

### Where the record is empty

Scale and supply-path integrity have no counted cell in any row. Independent assurance and channel execution are n/e for most rows, and quality evidence, where scored, never rises above 2. Across the matrix, 120 of 252 cells are n/e, 10 are n/a and 122 carry a score.

None of the readable vendor pages the sweep opened discloses a periodised curated-deal spend or impression figure. The only product-level volumes are Maestro's 30K+ deals and 50 billion impressions, with no period, and Multilocal's 21+ SSP (supply-side platform) curation marketplaces [15, 256, 223, 266]. No scale threshold, the lowest being $50 million or 10 billion impressions in a stated 12-month window, could be applied to any row.

Supply-path integrity anchors need an observed sample of a deal's inventory checked against ads.txt, sellers.json and SupplyChain, and the research recorded none; vendors describe their own adherence instead, as PubMatic's console page does for Connect [222, 225]. The files themselves define no curator role: a curator appears only where it is paid as a seller, and a data or selection provider outside the payment flow cannot appear at all [143, 142, 141]. The independent observations found tie to no scored product: Adalytics documented ads served to declared data-centre bots through PMP deal IDs on several exchanges, with examples from March 2022 to November 2024 [69], and its Magnite examples predate ClearLine curation [69].

Where assurance is scored, the best is certification of the operating company and the rest is self-attestation. Seals from TAG (the Trustworthy Accountability Group) certify companies, and none certifies a deal or a product [308, 152, 155, 154, 153]. TAG's 2026 release lists PubMatic and Magnite among certified companies without naming their seals [242, 231] [242, 243]. Microsoft's MRC (Media Rating Council) accreditation covers paid search clicks, which leaves Curate outside it [219].

In the sources screened, every quantified quality or performance claim found for a cohort product is a vendor assertion, except one commissioned study of platform economics; none names an independent measurer, published method, sample or period [119, 120, 121, 125]. No independent measurement with a published method was found that tests a named curated deal's quality promise against the same inventory in the open auction [310, 101, 311]. [Chapter 7](#quality) holds the detail. Channel execution is n/e for most rows because pages advertise environments without documenting any as live; Microsoft's pages, for one, document no live execution per environment, and its About page states no advertised scope [110, 220].

The DQ in this paper therefore rests mainly on how selection, data sources, auction rules and charges are documented. For Commerce Grid, Maestro, Microsoft Curate and TL Direct, the whole Standing group, 30% of the weight, is a single pricing cell. A Strong band is no evidence of clean supply, product-level assurance or proven quality, because no row documents any of the three.

### How stable the bands are

Weights and prior strength move a few bands. Missing evidence moves all of them.

The index was recomputed under four other weightings, with Raw and the field mean recomputed each time: equal thirds, Depth-led (60/20/20), Reach-led (30/45/25) and Standing-led (30/20/50). 4 of the 7 banded rows keep their band under all four: Commerce Grid, Maestro, ClearLine and PubMatic Connect. Under the Depth-led and Standing-led weightings, Index Marketplaces falls to Mid and Microsoft Curate rises to Strong. TL Direct falls to Narrow under Standing-led. Positions move by up to 3 places among exchange curation platforms and 4 among curated deal products.

Removing the shrinkage (K = 0, which reports Raw) changes 2 bands: Microsoft Curate rises to Strong and TL Direct falls to Narrow. At K = 2 no band changes. At K = 8, 1 band changes, as Index Marketplaces falls to Mid.

Each counted cell was then deleted in turn, with the field mean and every other row held fixed. Among exchange curation platforms, 47 of 68 deletions change a row's band or status, and most follow from the thresholds. Commerce Grid, Maestro, Microsoft Curate and TL Direct have exactly six counted cells, so any deletion leaves them without a band (24 cases). The thin records have four, so any deletion takes them out of the matrix (16 cases). The other changes are switches: Index Marketplaces falls to Mid when any of five of its cells is deleted, and ClearLine rises to Strong when its score of 1 for auction mechanics or for pricing is deleted. Leaving out a low score raises a quotient: deleting Microsoft Curate's activation routes score of 1 would lift its DQ from 47 to 52, and leave it with five cells and no band.

The missing-evidence scenarios work the other way. One row at a time, with the field mean and every other row frozen, each banded row's n/e cells were filled with 1 and then with 4. These are scenarios and bound nothing. 7 of the 7 banded rows change band under one fill or both, and each carries the ± mark. PubMatic Connect and Index Marketplaces, with ten counted cells each, still drop to Mid when their two n/e cells, scale and supply-path integrity, are filled with 1. No row reaches Front rank at a fill of 4. Only the hypothetical envelope, which fills every n/e cell with 5, lifts the rows with the most unknowns (Commerce Grid, Maestro, Microsoft Curate and TL Direct) above the line.

### What the bands rest on

The bands summarise a thin record, and four facts belong beside them [314]. Only 7 of the 21 scored products carry a band. Of the 122 counted cells, 92 rest on grade-D evidence (vendor marketing, a press release or a case study), 26 on grade C (vendor technical documentation) and four on grade A or B. In four banded exchange rows, Commerce Grid, Maestro, Microsoft Curate and TL Direct, pricing transparency is the only counted Standing cell, so one cell carries the group's full 30% weight. All seven banded rows move under at least one missing-evidence scenario, as the previous section shows.

A stricter cut shows how much rests on grade D. Counting only cells graded C or better, four products clear the admission rule (Maestro, Google Ad Manager Curation, Index Marketplaces and Microsoft Curate), no curated deal product or managed client marketplace is admitted, and no cohort carries bands [314]. This is a diagnostic, not the registered method, and the published bands stand.

The admission rule lets live and Beta products in, scored from their documentation like any other row, and three admitted rows rest on availability that is not confirmed [314]. Google marks its curation settings Beta, Stagwell Curate is live on its operator's claim, and WPP Media's Premium Marketplace has no page newer than February 2023 [59] [5] [18]. Without the Beta Google row the exchange field DQ falls from 46.7 to 45.4 and Microsoft Curate moves from Mid to Strong; the other two sit in a cohort with no bands [314].

### The interest-marked row and the single scorer

Lotame sells PMPs for web and mobile and customised connected TV (CTV) marketplaces built from unnamed SSPs [13]. Its row is admitted with eight counted cells, scores 40 and is one of the 2 eligible curated deal products. Removing it from the cohort's field mean moves the curated-deal field DQ from 41.32 to 41.75. The cohort carries no bands, so no band can change. The other cohorts hold no interest-marked row, and their field DQs are unchanged. Two exchange rows, Index Marketplaces and ClearLine, carry a separate relationship mark: their public records name Samba, the author's employer, and Lotame among their data partners [108] [94]. Both marks appear in the matrix and in the score and matrix exports.

One scoring pass set every score. Verifiers checked the facts behind the cells, nobody re-scored them, and no separate reviewer checked the conflicted admission and scoring decisions ([chapter 13](#method)).

### The profiles

7 products have deep profiles, each with a SWOT (strengths, weaknesses, opportunities and threats) whose lines cite their sources: ClearLine, Index Marketplaces, Microsoft Curate, PubMatic Connect, Audigent's SmartPMP, Lotame's Curated Marketplaces and Stagwell Curate. Lotame was profiled as the ownership test case for H5, the paper's test of supplier independence: Publicis Groupe acquired 100% of Lotame Solutions, Inc. in April 2025 [41]. Its interest mark applies to the profile as it does to the row.

### Audigent SmartPMP (active: an operating brand of Experian ('a part of Experian'))

*Not stated in the sources opened: Experian's release of 4 December 2024 names the company only as 'Audigent', and Audigent's pages carry no copyright line or legal entity name (CUR-AUDIGENT-S01, CUR-AUDIGENT-S02, CUR-AUDIGENT-S07).*

Audience-data deal IDs from the company Experian bought in December 2024; routes come from partners' records and fees only as charge types.

**SWOT (2026-11-03).** Audigent, owned by Experian since December 2024, sells SmartPMP as audience data and inventory in one deal ID with exchange and DSP routes on record, but no source gives a fee amount, seller list, per-segment provenance or independent quality evidence, and Digiday reported on 11 August 2026 that Experian had retired the Audigent brand into Experian Marketing Services.

- **Strengths.** Exchange operators' records name Audigent: Google Ad Manager's curation help page lists 'Audigent/Experian' twice; Index Exchange and Nexxen releases of 2024–2025 name it a curation or data partner. (record) [59, 187, 188, 189] Experian posts of August 2025 describe deal-ID activation of Audigent packages in DV360, with over 250 deals via Index, and pre-built deals in Amazon DSP's Inventory Hub. (vendor claim) [190, 191] Its data inputs are named: publisher and data partners including Condé Nast, TransUnion, Warner Music Group and Fandom, Comscore Predictive Audiences, and Experian deterministic signals in DV360 packages. (vendor claim) [116, 192, 190] Owned by Experian plc since 4 December 2024; Experian said the deal adds 'sell-side distribution to its existing demand-side capabilities'. (record) [38]
- **Weaknesses.** Audigent discloses charge types, not amounts: 'PMP pricing is determined by bid floors and curation fees only', with no rate, base, payer or period in any source reviewed. (vendor claim) [116, 192] No seller list, domain list or seller count for any identified deal was found; composition wording is product-wide: publisher verticals and 'hundreds, or even thousands' of sources. (vendor claim) [193, 194, 190] Every performance figure is a vendor case claim (CPC, CTR, VCR, advertiser KPI) with no sample, dated period or independent measurer; none covers MFA, viewability or IVT. (vendor claim) [122, 123] No source gives per-segment provenance (collection method, permissions, legal basis), and no certification, audit or supply-chain sample covering SmartPMP was found. (vendor documentation) [193, 116, 195]
- **Opportunities.** Google's November 2024 launch said 'other DSPs' were 'coming in the months ahead', so the Google route could reach buyers beyond DV360 and Yahoo. (inference) [196] Nexxen's Curated Marketplace reached general availability in October 2025 with Audigent as a named client, a GA exchange route beside Google's Beta curation feature. (inference) [189, 59] Experian signals pre-loaded in DV360 packages (August 2025) and the acquisition's stated aim of 'sell-side distribution' give owner data a deal-ID route to buyers. (inference) [38, 190]
- **Threats.** Experian co-sells Audigent's audiences with 'Experian's syndicated audiences' (2025); no source shows whether SmartPMP stays a separate service or merges into Experian's own deals. (inference) [189, 190] Two routes rest on platform features still pre-GA when documented, Google Ad Manager curation (Beta) and Index's Data Vendor Ecosystem (early access), so access and terms may change. (inference) [59, 188] Claims to scale 'without cookies or device IDs' sit beside homepage counts of 1.7+ billion mobile advertising IDs; with segment identifiers unstated, privacy scrutiny could test those claims. (inference) [190, 116]

- **What is sold.** SmartPMP: curated deal IDs that combine Audigent's audience data with publisher inventory from SSP partners. The product page calls it 'a first-to-market, breakthrough data + inventory product' and 'the intelligent marriage of premium first party data sets and premium inventory pre-packaged together as a single, easy-to-buy deal ID' (CUR-AUDIGENT-S01). There are two variants, both register aliases: ContextualPMP ('Cookieless & futureproofed PMPs', which 'isolate hundreds, or even thousands of contextually relevant inventory sources across over 20 industry-leading SSPs', CUR-AUDIGENT-S04) and CognitivePMP ('Comscore Predictive Audiences'). The site navigation also offers a Comscore CognitivePMP Marketplace ('Search 1000+ Comscore CognitivePMPs™', CUR-AUDIGENT-S03). Experian's posts of August 2025 describe the same offer under other names: 'curated audience and contextual inventory packages' in DV360, delivered through Index Marketplaces (over 250 deals, CUR-AUDIGENT-S09), and 'Off-the-shelf deals in Inventory Hub on Amazon DSP', plus custom deal libraries (CUR-AUDIGENT-S10). Those posts do not use the name SmartPMP, so mapping their packages to this row is an inference. Audigent also sells audience segments that are not deal IDs (for example, in Nexxen DSP's segment builder, CUR-AUDIGENT-S11), and this row does not score them.
- **Who buys it.** Advertisers and agencies who buy through a DSP. The SmartPMP page speaks to 'digital media buyers' and says a deal is 'dropped into a campaign line item' (CUR-AUDIGENT-S04). The DV360 route has buyers 'add the deal ID to your line item' (CUR-AUDIGENT-S09). On Google Ad Manager, the deal ID 'can be subscribed to within the agency seat' (CUR-AUDIGENT-S13). One named advertiser appears: Alcon, in an undated vendor case page (CUR-AUDIGENT-S06). On the supply side, publishers and data partners supply data and inventory: Audigent's 'verified, opt-in data drives monetization' for partners including Condé Nast, TransUnion, Warner Music Group, Penske, a360 Media and Fandom (CUR-AUDIGENT-S02).
- **How it earns.** From curation fees carried in the deal price. The only charge-type statement is 'PMP pricing is determined by bid floors and curation fees only.' (CUR-AUDIGENT-S02, CUR-AUDIGENT-S03), and the DV360 post says Audigent 'integrates premium partner data into a single price' with 'no additional data management platform (DMP) fees' (CUR-AUDIGENT-S09). No amount, rate, base, payer, collector or period is stated. On the Google Ad Manager route, the platform documents the mechanism: 'Curation partners define the fee of the data segment or curated package', and Google pays the partner from the bid, 'taken out of the bid price before it enters the auction' (CUR-AUDIGENT-S13). How Audigent shares revenue with the publishers and data partners whose data it monetises is not disclosed. Experian does not break out Audigent's revenue (CUR-AUDIGENT-S08).
- **Supply and data dependencies.** Supply: Audigent says it has 'partnered with premium SSP platforms' and is 'Now integrated with over 20 of the industry's leading SSPs', naming none (CUR-AUDIGENT-S01, CUR-AUDIGENT-S03). Exchange records name three: Google Ad Manager (curation launch of 4 November 2024; help page partners 'Audigent/Experian (Audience)' and 'Audigent/Experian (Contextual)', CUR-AUDIGENT-S12, CUR-AUDIGENT-S13), Index Exchange (Marketplaces early adopter, 23 January 2024; Data Vendor Ecosystem early launch partner, 7 October 2025; host of the DV360 packages, CUR-AUDIGENT-S14, CUR-AUDIGENT-S15, CUR-AUDIGENT-S09) and Nexxen (Curated Marketplace client, 1 October 2025, CUR-AUDIGENT-S16). Data: Audigent's first-party audience data from publisher and data partners (Condé Nast, TransUnion, Warner Music Group, Penske, a360 Media, Fandom, CUR-AUDIGENT-S02); Experian's deterministic signals and The Weather Company's weather data in the DV360 packages (CUR-AUDIGENT-S09); Comscore Predictive Audiences for CognitivePMP (CUR-AUDIGENT-S01, CUR-AUDIGENT-S03); and Audigent's Hadron ID identity suite (CUR-AUDIGENT-S07). On the Google route, a curation partner may receive 'URL, App ID, Pub ID, GPID, user geography at the metro level, and TCF consent string' as contextual signals, plus identifiers including Device ID and the Ad Manager third-party cookie where the publisher allows data sharing (CUR-AUDIGENT-S13).
- **Differentiation.** Packages audience data and inventory into one deal ID (vendor's description), with three variants: audience (SmartPMP), contextual and cookieless (ContextualPMP) and Comscore predictive (CognitivePMP). Draws its audiences from named media-owner and data partners (Condé Nast, TransUnion, Warner Music Group, Penske, a360 Media, Fandom) and, since the acquisition, from Experian's deterministic signals (DV360 packages), per vendor pages. Is listed on three exchange records (Google Ad Manager, Index Exchange, Nexxen) and reaches three named DSPs (DV360, Yahoo, Amazon DSP's Inventory Hub). Among the cohort B seeds screened by stream F, the Amazon DSP route is the only one stated for this product. Appears twice in Google Ad Manager's partner list, as 'Audigent/Experian (Audience)' and 'Audigent/Experian (Contextual)'.
- **Risks.** Channel overlap inside one owner: Experian sells its own Curated Deals, Deal Engine and Deal Marketplaces alongside Audigent, and Experian executives already pitch 'Audigent's first-party audiences' with 'Experian's syndicated audiences' (CUR-AUDIGENT-S16). Rationalisation or rebranding is possible (inference, not a documented plan). Platform dependency: Google Ad Manager curation is still Beta ('might not be available in your network') and Index's Data Vendor Ecosystem was in early access at its 7 October 2025 announcement, so two of the three exchange routes were documented at pre-GA status. Fee opacity: a buyer cannot reconcile the curation fee, data costs and exchange or DSP take rates from public records. On Google's route, the fee is taken out of the bid before the auction, so the media portion of the bid is net of an undisclosed amount. Data-consent scrutiny: in-flight enrichment with audience data from a credit-bureau owner and publisher partners rests on 'opt-in' assertions without a documented consent route (inference about regulatory exposure; no enforcement action was searched for or found). Claim tension: 'never 3rd party cookies' and 'without cookies or device IDs' sit beside homepage statistics of '1.7+ billion mobile advertising IDs' and a '13.3+ billion 1st-party device graph' (see counterevidence).
- **Best fit.** Buyers who want audience- or context-enriched PMPs activated as a single deal ID in DV360, Amazon DSP or Yahoo without onboarding data DSP-side. Buyers whose own measurement will judge outcomes after purchase and who do not need pre-purchase composition or fee disclosure.
- **Poor fit.** Buyers who need seller and domain lists, a fee schedule with base and collector, or a supply-chain sample before paying a curation fee. Buyers who require independently measured MFA, viewability, IVT or attention outcomes for the deal; no such evidence was found. Buyers who need documented US production use or contractual post-purchase controls from public records.
- **Limits.** Audigent's own pages name no exchange, DSP, channel, fee amount, seller list or geography. Every named route comes from a platform's or the owner's announcement. The August 2025 route posts describe 'curated audience and contextual inventory packages' and 'curated deals' without the name SmartPMP, so attributing those routes to this row (rather than to Audigent's offer in general) is an inference within the register's alias set. Fees are disclosed only as charge types ('bid floors and curation fees'), with no amount, rate, base, payer, collector or period. Provenance stops at source type and named partners; per-segment collection method, intended use, permissions and legal basis are undocumented on the pages read. Every performance figure is a vendor case claim without a sample, period or measurer, and none addresses a registered quality dimension. No curated-deal spend or impression figure exists in a 12-month window; the scale figures found are counts or company-wide statistics. No product-scoped assurance and no supply-chain or IVT evidence were found in the sources screened. Post-purchase controls (change notification, audit rights, exportability, remedies; brief Q5) are not documented in any source opened.

| Dimension | Availability | Evidence | Observed |
|---|---|---|---|
| Selection transparency | live | D | The selection logic is named but not specified. It is an audience match of Audigent's first-party segments to 'premium curated publisher inventory' (SmartPMP), contextual relevance across sources on 'over 20 industry-leading SSPs' (ContextualPMP) and Comscore Predictive Audiences (CognitivePMP), with packages that 'optimize consumer-safe data with premium inventory supply at scale' and DV360 deals paired with 'hand-vetted inventory'. No page read states a threshold, an inclusion or exclusion rule, a quality filter or an update cadence. In anchor terms, this is logic named without thresholds, from vendor marketing. [193, 192, 194, 190] |
| Composition disclosure | unconfirmed | D | No evidence found of a seller list, domain list, seller count or seller categories for any SmartPMP, ContextualPMP, CognitivePMP or package. Searched: the SmartPMP, ContextualPMP, CognitivePMP marketplace and homepage pages, both case pages, Experian's three August 2025 posts and the Google, Index and Nexxen records, all on 3 November 2026. The composition wording is limited to 'premium curated publisher inventory', 'some of the biggest names in entertainment, sports, and lifestyle' and, for DV360, 'hand-vetted inventory' from 'top-tier publishers'. The homepage figure of 476,080 web/mobile sites is company-wide and unperiodised. No identified deal's contents were observed; the DSP and exchange interfaces are authenticated and were not accessed. [193, 116, 190] |
| Data provenance | live | D | Provenance is documented at the level of source type and named providers only. The pages state '100% 1st party audience data, never 3rd party cookies' and 'verified, opt-in' data. They name the partners Condé Nast, TransUnion, Warner Music Group, Penske, a360 Media and Fandom, say that Experian's deterministic signals and The Weather Company's data are pre-loaded in the DV360 packages, and name Comscore Predictive Audiences for CognitivePMP. No page gives per-segment collection method, intended use, permissions or legal basis. ContextualPMP's 'beyond GDPR, CCPA, and other legislation' is an assertion, not a documented basis. The Platform Privacy Policy linked from the homepage was not read. [193, 116, 194, 190, 239] |
| Quality evidence | live | D | The only quantified claims are vendor case pages. One reports '12% - 67% lower CPC', '1.6x - 2.5x higher CTR' and '12.5% - 42% higher VCR' against 'traditional DSP-level data to open exchange inventory', with no advertiser, date, sample, period, measurer or method. The Alcon case is a single-advertiser before-after against Alcon's target KPI, tracked by Alcon's floodlight tag. None of these addresses a registered quality dimension (MFA, viewability, IVT, suitability, attention, carbon). The remaining statements ('surpassing performance targets', 'quality over quantity', 'start closer to KPIs') have no measurement. In anchor terms, these are claims without qualifying measurement or a vendor study without method. [122, 123, 191, 187] |
| Auction mechanics | unconfirmed | E |   |
| Channel execution | live | D | The SmartPMP page advertises no channel, and workstream U recorded its scope as 'not stated' (CUR-U channel_scope), so the anchor percentage cannot be computed against U's frozen scope. Experian's DV360 post of 20 August 2025 advertises the packages 'across connected TV (CTV), display, and online video (OLV)' and documents them as launched ('over 250 ready-to-run deals'). The Alcon case documents display ('display creatives'), and the case-study page reports a video metric (VCR). On the DV360 post's own scope, three of three environments are announced as available in one vendor post, with no per-channel deal list. [193, 190, 123, 122] |
| Activation routes | live | D | Three named DSP endpoints have an activation description for Audigent deals. Display & Video 360: Google's 4 November 2024 launch says curation partners' deals are available 'via Display & Video 360 and Yahoo', and Experian's 20 August 2025 post says to search Audigent in DV360's Marketplace and 'add the deal ID to your line item'. Yahoo DSP: covered only by Google's launch-day statement for all partners. Amazon DSP: 'Off-the-shelf deals in Inventory Hub on Amazon DSP', 25 August 2025. Nexxen DSP carries Audigent audience segments, not deal IDs, so it is not counted. Index's 'preferred DSPs' and Audigent's own 'DSP integrations' name no endpoint. All route records are platform or owner announcements (marketing grade), not help-centre documentation. Stream F's table graded the Google and Index records C; by the grade definitions, a blog post and a press release are D. [196, 190, 191, 239, 187] |
| Scale | live | D | No evidence found of curated-deal spend or impressions in a stated 12-month window. Searched: the four Audigent pages, both case pages, Experian's acquisition release and FY25 results, and the three August 2025 posts. The figures found are counts and company-wide statistics. They are: '1000+ Comscore CognitivePMPs' (navigation, unperiodised); 'over 250 ready-to-run deals' in DV360 (August 2025); 'hundreds of pre-built curated deals' in Amazon's Inventory Hub (August 2025); 'over 1,000 ready-to-activate segments' in Nexxen (segments, not deals); '100,000+ campaigns each month' (company-wide); and a homepage stat block (4.0+ billion 1st-party IDs, 476,080 web/mobile sites, 400,593+ audience segments; unperiodised). All are context only under the anchor. [116, 192, 190, 191, 239] |
| US availability and coverage | unconfirmed | D | No page read states that SmartPMP is available in the US. The US nexus is an office address ('Audigent NYC', 200 Varick St., New York) and the Costa Mesa, Calif. dateline of Experian's acquisition release. Experian's three August 2025 route posts state no geography. The Alcon case names a customer but no geography and no dated period, so no named US customer in a stated period is documented. In anchor terms, the US appears only by address in marketing pages. [193, 116, 123, 38, 190, 191] |
| Pricing transparency | live | D | Audigent names charge types only: 'PMP pricing is determined by bid floors and curation fees only'. Beside that sit general assertions: 'Removing detrimental bid shading practices', 'more fairly priced PMP alternatives', a 'single price' with 'no additional data management platform (DMP) fees' in DV360, and 'reduces overall buyer costs' on Amazon. No amount, rate, base, payer, collector or period is disclosed, and no Deal Sync API curfeetype was found. On the Google route, the platform documents that the partner defines the fee and that Google pays it 'taken out of the bid price before it enters the auction'. That is a third-party mechanism, recorded as an attribute. [116, 192, 193, 190, 191, 59] |
| Independent assurance | unconfirmed | D | No evidence found of any certification, accreditation or audit naming Audigent, SmartPMP or its deals. Searched: the four Audigent pages (no TAG or other seal on the homepage), Experian's acquisition release and three August 2025 posts, a web search for Audigent TAG seals and TAG's 6 March 2026 certifications news page (body not returned; no company named), all on 3 November 2026. The TAG Registry was not queried. Comscore's and Google's platform roles are inputs and routes, not assurance of this product. [116, 38, 195] |
| Supply-path integrity | unconfirmed | D | No evidence found: no sample of SmartPMP or package supply was examined against ads.txt, sellers.json or SupplyChain, and no Audigent page or route post states an IVT filtering method or measurer. Searched: the four Audigent pages, both case pages, Experian's three August 2025 posts, and the Google, Index and Nexxen records, all on 3 November 2026. Index is credited with 'efficiency and transparency' in the DV360 post, but that describes the exchange, not a deal sample. Whether Audigent appears as a node in any deal's SupplyChain object is not documented. [193, 116, 190] |

Performance evidence: SmartPMPs drove '12% - 67% lower CPC', '1.6x - 2.5x higher CTR' and '12.5% - 42% higher VCR' than 'applying traditional DSP-level data to open exchange inventory', 'Through rigorous testing'. (vendor assertion, sponsor Audigent (a part of Experian), self-published) [122]; 'Audigent exceeded Alcon's target KPI by 27%', 'reducing the cost per online eye exam by 87% from the start of the optimizations'. The page also reports 'a +1.5 share gain', 'revenue grew +32% in a second segment' and '+1.9' share gain in the following quarter. (vendor assertion, sponsor Audigent (a part of Experian), self-published; client Alcon) [123]; Audigent's initial Index Marketplaces campaigns 'have yielded promising results, surpassing performance targets'. (vendor assertion, sponsor Index Exchange release quoting Audigent (both interested parties)) [187]; Audigent's Amazon DSP deals prioritise 'quality over quantity through premium web inventory' and the approach 'reduces overall buyer costs'. (vendor assertion, sponsor Experian Marketing Services (owner)) [191]; In DV360, 'Exclusive audience data is paired with hand-vetted inventory, so you start closer to KPIs from day one.' (vendor assertion, sponsor Experian Marketing Services (owner)) [190]

### Index Marketplaces (active)

*Index Exchange Inc. (the name in the Index Exchange Inc. v. Google LLC docket card read by workstream I and its verifier; Index's own terms page names only 'Index Exchange')*

The only exchange product that shows the arithmetic of a curated bid in its own documentation, with fee disclosure left optional for the partner.

**SWOT (2026-11-03).** Index Marketplaces publicly documents its fee lines, floor rule, net-of-fee auction and data-vendor list, but publishes no fee amounts, buyer-facing seller list, periodised volume or independently measured quality.

- **Strengths.** The knowledge base documents three fee lines with bases, collectors and worked examples: partner fee, data-vendor fee and Index revenue share, all removed from the bid before auction. (vendor documentation) [109, 197] Index documents first- or fixed-price packages, a floor at the higher of package and publisher floor plus fees, and net-of-fee bidding; fixed-price packages carry priority. (vendor documentation) [198, 14, 199] The knowledge base lists 49 active data vendors (21 October 2026) with data types and markets, and documents vendor-set segment fees with optional caps. (vendor documentation) [94, 197] The package builder documents partner-set targeting across Web, In-App and CTV and names Display & Video 360, The Trade Desk and Amazon as buyer-seat DSPs. (vendor documentation) [14]
- **Weaknesses.** No fee amounts are published and the partner chooses whether its fee and vendor fees are disclosed, so public documentation gives no deal's all-in cost. (vendor documentation) [109, 197] No page opened documents a buyer-facing seller or domain list for a package, and a blank publisher list automatically adds new publishers without documented buyer notice. (vendor documentation) [198, 14, 200] Every performance figure found is Index's own case-study or release claim without a named measurer or method; none measures MFA, viewability or invalid traffic for a Marketplace deal. (vendor claim) [201, 202, 203, 204] Assurance is company-level: undated TAG seals and logos on its exchange-quality page, TAG Platinum recorded under 2019, none scoped to Index Marketplaces; no IVT measurer is named. (vendor claim) [19, 205]
- **Opportunities.** The Data Vendor Ecosystem went from seventeen named vendors at early access (7 October 2025) to 49 active by 21 October 2026, widening what partners can package. (inference) [188, 94] Index Cloud (closed beta, 2026) could widen sell-side decisioning: in Index's 6 October 2026 case, Chalice AI's model fed Marketplaces a targeting key for dentsu. (inference) [202, 205] Index said in October 2025 that the Data Vendor Ecosystem 'will also extend to media owners in 2026', opening a publisher-side outlet for the same segments. (inference) [188]
- **Threats.** Media owners are opted in to Standard Marketplaces by default and can opt out, so a package's supply can shrink without documented notice to the buyer. (inference) [109, 206] Under the undisclosed fee model media owners see neither media spend nor partner and vendor fees, which could push publishers wary of curation fees to opt out. (inference) [109, 206] Index's 'does not charge data fees' covers only its own charge; partner and vendor fees and Index's revenue share still come off the bid, inviting buyer scrutiny. (inference) [188, 109, 197]

- **What is sold.** A self-service deal-building platform on Index Exchange's supply: a Marketplace Partner (an agency, data provider, curator or retail media network) creates Marketplace Packages that bundle inventory from one or more of Index's '3,000+ Media Owners', adds targeting and third-party or its own registered segments, sets a first-price floor or a fixed price and its own fee, and assigns the package to a DSP and optionally a buyer seat under a single deal ID. Data vendors sell segments into packages through the Data Vendor Ecosystem; media owners are opted into Standard Marketplaces by default.
- **Who buys it.** Two sides. Marketplace Partners (agencies, data providers, retail media networks, curation companies such as Multilocal, SWYM.ai and Chalice AI in Index's own materials) use the platform to build and sell packages; advertisers and agencies buy the resulting deal IDs in their DSPs (Display & Video 360, The Trade Desk, Amazon and Xandr are named in the knowledge base). Index's case studies name Butler/Till, PMG and dentsu as agencies; the brands are unnamed.
- **How it earns.** Index earns its revenue share on Marketplace transactions, negotiated directly with each media owner on Standard Marketplaces (15% in the knowledge base's worked example, not a schedule). The partner's Marketplace Owner Fee and each data vendor's Marketplace Vendor Fee are deducted from the buyer's bid before the auction and paid to them; Index says it 'does not charge data fees' on the Data Vendor Ecosystem and that there is 'no incremental expense for partners using Marketplaces'. For Standard Marketplaces Index handles all payments.
- **Supply and data dependencies.** Supply: Index Exchange's own media owners (web, in-app, CTV), opted in to Standard Marketplaces by default with an opt-out; Custom Marketplaces may need publisher opt-in. Data: 49 active vendors in the Data Vendor Ecosystem on 21 October 2026 (among them Acxiom, Adelaide, Audigent, Comscore, DoubleVerify, Dstillery, Experian, IAS, IRIS.TV, LiveRamp, Lotame, Peer39, Samba TV and The Weather Company), plus segments a partner registers itself; vendors must grant permission before a partner can target their segments. Demand: the buyer's DSP seat. Partner models can run inside Index Cloud (a separate product), as in the dentsu and Chalice AI case.
- **Differentiation.** Among the cohort A products workstream E screened, Index's knowledge base is the only one that publishes the arithmetic of a curated bid (partner fee, Index share, net bid to the media owner) and of the floor sent to DSPs. The knowledge base names four DSPs with documented package functionality (buyer-seat requirement), where workstream E found no cohort A product naming more than one DSP in text on the pages it opened (CUR-E-F38). Data vendors are listed publicly by name (49 on 21 October 2026) with data types and markets, and segment fees are documented as a vendor-set CPM or percentage with an up-to cap and per-vendor stacking rules. Index documents that Marketplace bids compete net of fees and must clear the publisher's floor plus fees, a stated default of no floor advantage; priority depends on the auction type the package's creator chooses, and Index defines a fixed price auction as having priority over all other auction bid types (corrected in the final live sweep, 4 November 2026). A partner can run its own decisioning model inside Index's infrastructure (Index Cloud) and feed Marketplaces a targeting key, a sell-side decisioning route documented in one named agency case.
- **Risks.** Control of Index Exchange is not established from a filing or registry; an undisclosed owner or a future transaction could change the product's independence coding (H5). Supply depends on default media-owner opt-in to Standard Marketplaces with a standing opt-out, so package supply can shrink without notice to the buyer (inference from S01 and S12). Index reserves the right to update targeting on deals using LiveRamp Data Marketplace segments, a post-purchase change right held by the exchange with no documented buyer notification (Q5). Under the undisclosed fee model the partner and vendor fees are undisclosed to media owners and optional to buyers, so the curated deal's fee stack can be opaque to both ends of the trade. The Xandr route depends on Microsoft Invest, whose media buying Microsoft said would end on 28 February 2026; execution after that date is not confirmed. Independent observation of Index Exchange PMP deal IDs serving declared data-centre bots (CUR-B-F29) is a reputational and supply-path risk for deal-based trading on the exchange, though not tied to this product.
- **Best fit.** A data company, agency or curator that wants to package Index supply with its own or third-party segments and set its own fee without building exchange integrations. A buyer on Display & Video 360, The Trade Desk or Amazon DSP that wants a single deal ID spanning several Index media owners across web, in-app and CTV. A curator that needs documented fee and floor arithmetic before contracting, since Index publishes both.
- **Poor fit.** A buyer that needs to verify the sellers and domains in a deal before purchase from public documentation. A buyer that needs independently measured quality (MFA, viewability, IVT) against the open auction on the same inventory. A buyer that needs a published fee schedule or a guaranteed all-in cost. Audio or DOOH campaigns, which Index Marketplaces neither advertises nor documents.
- **Limits.** No fee amounts are published: the 5% partner fee, $1.00 CPM, 15% Index revenue share and vendor fee figures are worked-example values, so the all-in cost of any Marketplace deal is unknown (H2a unresolved). No buyer-facing list of the media owners or domains in a package is documented, and no example deal was found (H6 composition: no public disclosure). Request sampling, deal-health throttling and the treatment of 'otherwise specified' Custom Marketplace mechanics are not documented. No periodised Index Marketplaces volume is public; the 'hundreds of millions of impressions daily' figure is a multi-region platform rate. All performance figures are Index case-study claims without comparator, measurer or method. The knowledge base documents the partner's tool; it does not document what a buyer sees about a package's rules, data or sellers before or after purchase. Interest marks: Samba TV (the author's employer since May 2025) and Lotame (a former employer) are among the data vendors; the author has no recorded relationship with Index Exchange. Interest marks do not change any record here.

| Dimension | Availability | Evidence | Observed |
|---|---|---|---|
| Selection transparency | live | C | Index's knowledge base documents the selection logic a Marketplace Partner configures per package: publisher or deal targeting, domain and app-bundle include and exclude lists of up to 10,000 entries, up to four include segment rule sets and one exclude, device type, Web or In-App environment, format, geography, dayparting, a minimum viewability probability, VCR and CTR (beta), video, streaming-TV content and IAB categories, with at least one non-daypart dimension required. Thresholds are chosen by the partner deal by deal and are not observable for any identified deal; no update cadence is documented and the buyer's view of a package's rules is not stated. [14, 198] |
| Composition disclosure | live | D | No page opened documents a buyer-facing list of the media owners, sellers or domains inside a Marketplace Package: the knowledge base says packages 'bundle inventory from multiple Media Owners' and that publishers, buyers, DSPs and partners can build reports in the Index UI, without saying whether reports break down by seller or domain. The product page asserts 'Real-time dashboards and audit logs' and 'Total transparency' into inventory quality, and the launch release promises media owners visibility of who buys their inventory; no identified example deal was found, so only a general assertion of transparency is on record. [200, 198, 109, 187] |
| Data provenance | live | C | The knowledge base lists 49 active Data Vendors (21 October 2026) by name with data types (demographic, behavioural, contextual, purchase, brand safety, media quality, attention, identity and others), channels and countries of availability, and a data-usage-conditions page restricts LiveRamp Data Marketplace segments to US users; source type and provider are therefore named at vendor level. Collection method, intended use, permissions and legal basis are not documented per segment on the pages opened, segment descriptions sit in the authenticated UI, and the provenance of segments a partner registers itself is not described. [94, 240, 197, 14] |
| Quality evidence | live | D | Index's own materials carry outcome claims (conversion rate up 56% and cost per conversion down 26% for a US financial-services brand, August to November 2024; cost per booking down 41% and bookings up 14% for a hotel brand in Europe; 9x brand preference on streaming TV for PMG) and a 25% GPID accuracy figure, none with a comparator, measurer or method, and no named quality dimension (MFA share, viewability, IVT, suitability, attention) is measured for any Marketplace deal. The independent Adalytics observation of Index Exchange PMP deal IDs serving declared bots (workstream B, CUR-B-F29, examples 2022 to 2024) concerns exchange deals with 2020-era names and is not an evaluation of an Index Marketplaces quality claim. [201, 202, 203, 204, 19] |
| Auction mechanics | live | C | The knowledge base documents the auction type per package as partner-set First (floor in USD, minimum $0.10) or Fixed; DSPs are sent the higher of the package floor and the highest publisher floor grossed up for fees, and a fixed-price deal below publisher floor plus fees is ineligible to transact. 'Unless otherwise specified' Marketplace bids compete net of all fees and win only if 'higher than any other net bid', which documents price-based competition with no stated priority; buyer eligibility is a DSP plus an optional, recommended buyer seat, media owners are opted in to Standard Marketplaces by default, and request sampling or deal-health throttling is not documented. [14, 199, 198, 109, 206] |
| Channel execution | live | C | Advertised scope: streaming TV, display, video and mobile app (launch release, 23 January 2024; streaming TV and CTV on the 2026 product page), which maps to web display, web video, mobile in-app and CTV. The package builder documents targeting for Web and In-App environments, CTV device types, Banner, Native and Video formats and streaming-TV content, so all four advertised environments are documented as available in the tool; Index's case studies describe display (US, 2024) and streaming-TV executions without stating environment-level results. Audio and DOOH are neither advertised nor documented. [14, 187, 200, 201, 203] |
| Activation routes | live | C | The package-creation page names four DSPs for which 'a Buyer Seat ID is required': Display & Video 360, The Trade Desk, Amazon and Xandr; Microsoft said on 14 May 2025 it would stop supporting Microsoft Invest (Xandr) media buying from 28 February 2026 (workstream E's verifier, CUR-E-F38), so that route's current status is unconfirmed. The releases assert 'preferred DSPs' (2024) and 'reach all DSPs' (2025) without names, and the full DSP list in the UI is not on the page. [14, 188, 187] |
| Scale | live | D | no evidence found: searched the product page, the releases of 23 January 2024, 7 October 2025 and 7 January 2026, the knowledge base and three Index case studies. The only figures are 'hundreds of millions of impressions daily' for the platform across four regions (7 October 2025; a daily rate with no window, unit perimeter or US split), '3,000+ Media Owners' (a supply count) and campaign volumes in one case study, none a curated-deal spend or impression total in a stated 12-month window. [188, 198, 202] |
| US availability and coverage | live | C | US availability is documented in the knowledge base: LiveRamp Data Marketplace segments 'can only be used to target users in the United States', geography targeting covers countries, states, metro codes and postal codes, and 20 of the 49 active data vendors list the USA among their markets. US production use appears only in Index's own case study (grade D): SWYM.ai for Butler/Till's client, a 'major US financial services brand', 'across multiple markets in the US', August to November 2024, brand unnamed. [240, 94, 14, 201] |
| Pricing transparency | live | C | The knowledge base documents each charge type, base and collector without a schedule: a Marketplace Owner Fee set by the partner as a percentage of media spend or a fixed USD CPM (account default, deal-level or media-owner-level), a Marketplace Vendor Fee set by each data vendor as a fixed CPM or percentage of media with an optional up-to CPM cap (highest per vendor, summed across vendors), and an Index revenue share negotiated with each media owner (15% in a worked example), all removed from the buyer's bid before the auction, with Index handling payment on Standard Marketplaces. Amounts are illustrative only, the partner chooses whether owner and vendor fees are disclosed in reporting (media spend is always disclosed to buyers), and Index's 'does not charge data fees' covers only Index's own data charge. [109, 197, 14, 188] |
| Independent assurance | live | D | Index's exchange-quality and about pages display TAG Brand Safety Certified, TAG Certified Against Malware and TAG Certified Against Fraud seals, 'TAG Platinum' status, an IAB Gold Standard logo and HUMAN, IAS and The Media Trust logos, all at company level, undated and with no scope covering Index Marketplaces. TAG's own registry was not opened in this profile, so the seals are vendor-stated; no audit of a Marketplace package's composition or fees was found. [19, 205] |
| Supply-path integrity | live | D | no evidence found: no observed sample of Marketplace deal inventory was available; searched the product page, the knowledge base, the exchange-quality page and the workstream records. Index asserts that it screens and audits 'every app and site on the exchange', prohibits MFA sites and helped develop ads.txt and sellers.json, without naming an accredited IVT measurer for Marketplace deals; the independent Adalytics report (CUR-B-F29, examples March 2022 to November 2024) observed Index Exchange PMP deal IDs serving declared data-centre bots, an exchange-level observation not tied to Index Marketplaces. [19] |

Performance evidence: SWYM.ai's ScaLE engine running through Index Marketplaces for Butler/Till's client, a 'major US financial services brand', August to November 2024: conversion rate up 56%, cost per conversion down 26%, SSPs used down 90%, domains down 52%. (vendor assertion, sponsor Index Exchange (editorial blog); partner SWYM.ai; agency Butler/Till) [201]; dentsu, Chalice AI and Index Cloud for a global hotel brand in Europe over two months: cost per booking down 41% and hotel room bookings up 14%, with Chalice's model scoring close to half a billion pages before the bid and supplying a targeting key to Index Marketplaces. (vendor assertion, sponsor Index Exchange (editorial blog); partners dentsu and Chalice AI quoted) [202]; PMG, IRIS.TV and Index on streaming TV for an outdoor lifestyle brand: 9x brand preference, 10x favourability and 3x consideration. (vendor assertion, sponsor Index Exchange (editorial blog); partners PMG and IRIS.TV quoted) [203]; Adelaide AU in Index Marketplaces, combined with Index's GPID support, with 'testing indicating improvements in accuracy by 25%'. (vendor assertion, sponsor Index Exchange (release)) [204]; Audigent's initial Marketplaces campaigns were 'surpassing performance targets' (launch release, 23 January 2024). (vendor assertion, sponsor Index Exchange (release)) [187]; Exchange-wide quality assertions: Index screens and audits 'every app and site on the exchange', 'all inventory is brand-safe, all traffic is real and valid', and it prohibits MFA sites. (vendor assertion, sponsor Index Exchange (product page)) [19]

### Lotame Curated Marketplaces (active (operating as the Lotame brand of Epsilon, a Publicis company))

*Lotame Solutions, Inc. until 1 October 2025, when it was merged into Epsilon Data Management, LLC and 'ceased to exist as a separate legal entity' (Lotame Services Privacy Notice, last modified 19 November 2025); Lotame now describes itself as 'a division of Epsilon Data Management, LLC'. Product pages still carry '©2026 Lotame Solutions, Inc.'.*

Curated PMPs and CTV marketplaces on Lotame data, now inside Publicis's Epsilon; no exchange, DSP or fee named on its own page.

**SWOT (2026-11-03).** Lotame, now a division of Publicis's Epsilon, sells managed multi-SSP PMP and CTV packages built on its audience data, but its pages name no SSP, DSP, fee, seller list, US customer or volume.

- **Strengths.** Lotame's ©2026 capability page offers CTV marketplaces and web and mobile PMPs that package its audience data with inventory from several SSPs, with Lotame handling launch and pacing. (vendor documentation) [13] Google's November 2024 Ad Manager curation launch named Lotame among eight curation partners whose packages, as a group, were 'currently available' via Display & Video 360 and Yahoo. (vendor documentation) [196] Onetag (October 2025), Index Exchange's Data Vendor Ecosystem (October 2025, early access) and Magnite's ClearLine Activate (September 2026) list Lotame as a data supplier to their tools. (vendor documentation) [207, 188, 108] Lotame's November 2025 privacy notice documents, company-wide, its collection methods, identifier sources, licensed TV viewership data for CTV, claimed EU consent basis and TCF vendor ID 95. (vendor documentation) [208]
- **Weaknesses.** Lotame's pages name no SSP, DSP, deal ID, seller or domain list; only general transparency assertions, on the capability page and PHD case, describe composition. (vendor documentation) [13, 125] No Lotame fee is published with an amount or base; on Google's curation route a curator may hide its package fee from buyers, and Lotame's setting is undocumented. (vendor documentation) [13, 209, 24] Every quality figure is a vendor assertion without method: early-trial gains of 20%, 15% and 21%, and a PHD view-through rate above 70% against an unsourced 60% benchmark. (vendor claim) [209, 125] The only documented customer case, PHD EMEA (published 25 February 2025), ran in the UK, Spain and Singapore before Publicis ownership; no US customer or deal was found. (vendor documentation) [209, 210, 125]
- **Opportunities.** In March 2025 Publicis said Lotame would be accessible to all Publicis clients and teams, an in-group buyer pool; no such client of the curated product is documented. (inference) [40] Google said in November 2024 that Programmatic Guaranteed and other DSPs would follow Display & Video 360 and Yahoo, which could widen the routes open to Lotame packages. (inference) [196] Licensed smart-TV viewership data, documented in the November 2025 privacy notice, gives the advertised CTV marketplaces a data input; no live CTV deal is on record yet. (inference) [208, 13] Three SSPs added Lotame data to their own tools between October 2025 and September 2026, letting Lotame segments reach curated deals that other parties build. (inference) [207, 188, 108]
- **Threats.** Agencies outside Publicis may treat a Publicis-owned curator as a rival's asset; the only documented agency buyer, Omnicom's PHD EMEA, bought before the April 2025 acquisition. (inference) [41, 210, 125] Onetag, Index and Magnite now carry Lotame data in their own curation tools, so deals using Lotame segments need not be Lotame Curated Marketplaces. (inference) [207, 188, 108] Lotame's only documented curator route, Google Ad Manager Curation, is still labelled Beta and its current help page names no DSP, so the route's continuity is undocumented. (inference) [196, 24] Lotame Solutions, Inc. ceased to exist on 1 October 2025, yet product pages and the ISO 27001 summary still name it; contracting entity and certificate continuity are unconfirmed. (inference) [208, 13, 211]

- **What is sold.** Operator-built curated deals: 'Lotame's Curated Marketplaces', described as a 'package of data and vetted inventory' made 'available to DSPs for scaled audience targeting via private marketplace (PMP)', with 'customized connected TV marketplaces, along with PMPs for web and mobile' (capability page, ©2026). Lotame says it 'curates highly correlated, quality inventory from leading SSPs using validated audience insights', offers 'Hand-selected supply partners and inventory' and provides 'operational support to ensure smooth launch and pacing'. Launched on 14 August 2024 as 'Lotame Curation', with custom audiences 'using cookieless publisher data' and insights from Lotame Analytics. No deal ID list, transaction-type default or self-serve interface is documented, so the product is a managed curated-deal service (decision-tree step 2), not a deal-building tool.
- **Who buys it.** Agencies and advertisers buying through DSPs. The one documented customer case is PHD EMEA (Omnicom) for an unnamed global CPG drinks brand in the UK, Spain and Singapore (case study dated 25 February 2025, before Publicis's April 2025 acquisition). Publicis said at the announcement that Lotame would be 'accessible to all Publicis clients and teams'; no named Publicis client of the curated product and no US buyer was found.
- **How it earns.** Not disclosed for the curated product: no curation fee, data fee, margin, CPM or revenue share is stated on any Lotame page opened, and the 2024 release says only that the offering 'Cuts down operational workload and data/tech fees'. On the one documented exchange route, Google Ad Manager Curation (Beta), a curator enters a package fee 'as either percentage of media or CPM' and can mark it 'visible or hidden from buyer', and Google 'will handle billing'; whether and how Lotame uses that field is not documented. Company-wide, Lotame's privacy hub lists data licence and data-marketplace buyer terms (incoming and outgoing data licence agreements, 'Data Marketplace Buyer Managed Service Terms'), and Publicis describes Lotame as running 'one of the largest end-to-end data marketplaces in the world'; Lotame's revenue is not disclosed (a private company before April 2025; no segment line for it was found in the Publicis filing opened).
- **Supply and data dependencies.** Supply: partner SSPs, unnamed on Lotame's own pages; the only exchange records naming Lotame as a curation partner are Google Ad Manager (4 November 2024 partner list) and, as a data supplier rather than a curator, Onetag (2 October 2025), Index Exchange's Data Vendor Ecosystem for Index Marketplaces (early access, 7 October 2025) and Magnite's ClearLine Activate (9 September 2026). Buying endpoints: 'DSPs' unnamed on Lotame's page; Display & Video 360 and Yahoo named by Google for curation partners' packages at launch. Data: per the Services Privacy Notice, identifiers from 'Business customers, data partners, and graphing partners', behavioural data from data partners and Lotame's own inferences, collected through pixels, tags, cookies (Lotame PID, Panorama ID), SDKs, APIs and file transfers, with television viewership data licensed 'from third parties that offer smart TVs and video streaming devices' for CTV targeting; the PHD case segment was 'collected via loyalty / credit cards and online actions across publisher sites'. Hosting: Lotame's Spherical platform on Amazon Web Services (security summary). Corporate: Epsilon Data Management, LLC within Publicis Groupe.
- **Differentiation.** A data company's curated product: the deals are packaged around Lotame's own audience data and analytics (Lotame Data Exchange, Lotame Analytics), with company-wide scale claims of 109 countries, more than 1.6 billion IDs and over 100 data sources (acquirer and partner statements, unperiodised). Multi-SSP by design on its own description ('from leading SSPs'; 'multiple supply-side platforms' in the PHD case), unlike single-exchange curation tools in cohort A; the SSPs are not named. Sold as a managed service with 'operational support to ensure smooth launch and pacing', not a self-serve deal builder. Advertises CTV marketplaces alongside web and mobile PMPs, with licensed TV viewership data documented in its privacy notice for CTV targeting. Lotame both curates and supplies data to other curators' tools (Onetag's Smart Curation Platform, Index Marketplaces' Data Vendor Ecosystem, Magnite's ClearLine Activate, and Google Ad Manager Curation as a partner), so Lotame data in a deal does not by itself make it a Lotame Curated Marketplace. Owned by an advertising holding company (Publicis, through Epsilon) since April 2025, and positioned at the announcement as 'accessible to all Publicis clients and teams'.
- **Risks.** Holding-company ownership: buyers at agencies outside Publicis may treat a Publicis-owned curator as a competitor's asset; the one documented case agency (PHD EMEA, Omnicom) bought before the acquisition, and no post-acquisition non-Publicis buyer is documented (inference about buyer behaviour, not a finding). Fee opacity: no Lotame charge is disclosed, and the Google Ad Manager route lets a curator hide its package fee from the buyer; a buyer cannot reconcile the deal-path charges from public records (H2b, Q3). Unverifiable quality promises: 'No MFAs' and the trial and case figures have no measurer or method; a buyer cannot establish that the promise was met (H3). Composition opacity: no seller or domain list is documented, and as a non-payment provider Lotame cannot appear in SupplyChain 1.0, so a buyer depends on the partner SSPs for traceability (H6). Sensitive data exposure: the Services Privacy Notice lists 'Racial and ethnic data (U.S. only)' and 'Political, religious, and philosophical beliefs (U.S. only)' among behaviour categories; whether such segments are ever applied to curated marketplaces is undocumented, which leaves the question to contract terms (no finding of use). Contracting-party drift: Lotame Solutions, Inc. ceased to exist as a separate legal entity on 1 October 2025 while product pages still carry its copyright line, so buyers should confirm which entity contracts and bills (Q5).
- **Best fit.** Agencies and advertisers that want audience-data-led PMP packages assembled across several SSPs and delivered to their existing DSP, with the curator handling launch and pacing. Multi-market campaigns in EMEA and APAC, where the only documented case ran and where Publicis said Lotame would drive Epsilon's expansion. Publicis clients and teams, to whom Publicis said Lotame would be accessible.
- **Poor fit.** Buyers who require a published fee schedule, a seller and domain list, or independent quality measurement before paying for a curated deal. US buyers who need documented US production use: no US deal or customer was found. Buyers whose DSP is not Display & Video 360 or Yahoo and who need a documented activation route: no other endpoint is named for the product. Buyers who want to build and edit their own deals on an exchange (a cohort A tool), since this is a managed operator-built product.
- **Limits.** Lotame's own pages name no SSP, DSP, deal ID, fee, seller list, US customer or volume figure, so most dimensions rest on platform records or on company-wide documentation rather than on the curated product's own disclosure. The only documented customer case is outside the US (UK, Spain, Singapore), predates the Publicis acquisition, and used the agency's own allow-lists as the inventory rule. The fullest provenance record (the Services Privacy Notice) is company-wide, not per curated deal or segment. Buyer-facing deal consoles (Google Ad Manager, DSP deal libraries) are authenticated and were not accessed, so fee visibility and composition on screen are unknown rather than absent. The certifier's record for Lotame's ISO 27001 certification did not render, so its scope and validity are unverified. Interest mark: the author is a former Lotame employee; no separate reviewer checked this profile's coding.

| Dimension | Availability | Evidence | Observed |
|---|---|---|---|
| Selection transparency | live | D | The page names the selection logic only in general terms: Lotame 'curates highly correlated, quality inventory from leading SSPs using validated audience insights' and offers 'Hand-selected supply partners and inventory', with 'No MFAs' as an exclusion promise; no threshold, scoring rule, update cadence or definition of 'quality' or 'hand-selected' is documented. In the one documented case the inventory rule was the agency's own allow-list ('customized using PHD EMEA's carefully vetted inventory "allow" lists') plus performance optimisation, so the selection rule there belonged partly to the buyer. Against the anchors: logic named without thresholds, in vendor marketing. [13, 125] |
| Composition disclosure | live | D | No seller list, domain list, seller count, seller categories or authenticated composition interface is documented for any Lotame curated marketplace; the page names no SSP and the PHD case names neither the SSPs ('multiple supply-side platforms (SSPs)') nor the publishers. The only composition language is a general assertion: the case's 'Better Control & Transparency' section says the marketplaces were 'transparently optimized based on performance'. Searched: the capability page, the launch release, the case page and PDF, Google's launch post and curation help page, and the Onetag, Index and Magnite records, on 3 November 2026. [13, 125] |
| Data provenance | live | C | Product records name a source type only ('cookieless publisher data', 'AI-powered consumer insights', 'market-level consumer data and unique audience insights from its Analytics suite'); one case segment is defined with its collection route ('Data is collected via loyalty / credit cards and online actions across publisher sites') and 'No modeling was done'. The Services Privacy Notice (vendor documentation, last modified 19 November 2025) documents, company-wide rather than per segment, the collection methods (pixels, tags, cookies, SDKs), the source categories ('Business customers, data partners, and graphing partners'), licensed TV viewership data for CTV, and the claimed legal basis (consent; TCF vendor ID 95). No named data provider, intended-use or permission record is given per curated deal or segment; the score is for documentation, never a conclusion of legal compliance. [13, 209, 125, 208] |
| Quality evidence | live | D | Every quality claim is a vendor assertion without qualifying measurement: early-trial '20% better campaign efficiencies, improved video completion rates by 15%, and 21% higher viewability' (no comparator, sample, period, geography or measurer), a PHD case 'View-Through Rate consistently above 70%, beating industry benchmarks of 60%' (benchmark unsourced, period and measurer unstated) and 'projected cost savings of up to 33%', and the unmeasured promises 'No MFAs. No waste.' and 'Higher-than-your-DSP match rates'. No vendor or commissioned study with method and comparator, and no independent study, was found in the sources screened by 3 November 2026. [13, 209, 125] |
| Auction mechanics | unconfirmed | E |   |
| Channel execution | unconfirmed | D | Advertised scope (workstream U): CTV ('customized connected TV marketplaces'), web and mobile ('PMPs for web and mobile'), with web not split into display and video and mobile not split into web and in-app. No record opened documents any of these environments as live for a named Lotame curated deal: the PHD case names PMPs across unnamed SSPs without an environment, and the 2024 trial's 'video completion rates' implies a video environment without saying web or CTV (inference). Google's launch post names transaction types and DSPs, not environments. [13, 209, 125] |
| Activation routes | beta | D | Lotame's own page names no buying endpoint ('available to DSPs'), which is n/e by the anchor. Two endpoints are named only by a partner platform: Google's 4 November 2024 launch post lists Lotame among the supported curation partners whose packages were 'currently available for Preferred Deals and Private Auctions via Display & Video 360 and Yahoo', on Ad Manager Curation (Beta). Onetag's release names no DSP ('for DSPs'), and the Index and Magnite records describe Lotame as a data supplier to those platforms' tools, not as a route for Lotame-built deals. [13, 196, 24, 207] |
| Scale | unconfirmed | E | No evidence found: searched Lotame's capability page, launch release, case page and PDF, Publicis's release and half-year note, and the Google, Onetag, Index and Magnite records for curated-deal spend or impressions in a stated 12-month window. The figures found are company-wide or unperiodised and are context only: 'spanning 109 countries', 'more than 1.6 billion IDs' and 'over 4,000 of the world's leading brands and publishers' (Publicis, 6 March 2025), and an LDX taxonomy 'built from more than 50,000 attributes' and 'over 100 data sources' (Onetag). [207, 40] |
| US availability and coverage | unconfirmed | D | The August 2024 launch release says the offering is 'available in North America, EMEA and APAC', which names a region that includes the US but not the US itself; the current capability page states no geography. No named US customer, US deal or US campaign period was found: the only case is in the UK, Spain and Singapore, and the privacy notice's 'U.S. only' data categories describe Lotame's data, not the curated product's availability. Against the anchors this is a regional marketing statement only. [13, 209, 125] |
| Pricing transparency | live | D | No charge Lotame controls or collects is disclosed with an amount, rate, base, payer, collector or period; the only pricing statement is the 2024 general assertion that the offering 'Cuts down operational workload and data/tech fees', and the current page carries no price. On the Google Ad Manager route the curator's fee is entered 'as either percentage of media or CPM' and can be marked 'visible or hidden from buyer', but Lotame's setting is not documented, and no Deal Sync API curfeetype declaration for Lotame was found. Third-party charges (SSP and DSP take rates; Index's statement that it 'does not charge data fees') are attributes, not Lotame's disclosure. [209, 13, 24] |
| Independent assurance | unconfirmed | C | Lotame's undated security summary (Public V2.1.1) says it runs an Information Security Management System 'certified to meet the ISO 27001 standard', covering the Spherical platform and corporate systems; the certification announcement now redirects to Schellman's certificate directory entry 1571242-15, which did not render (access error), so the certifier's statement of scope, standard version and validity was not read. Information-security certification of the operating company does not cover the composition, fees or invalid-traffic filtering of the curated marketplaces. The PHD case quotes the brand's 'CDAP (Consumer Data Assessment Process)' review of Lotame's 'data sourcing, consent, and activation practices', which is an advertiser's compliance review reported by the vendor, not an independent audit. No TAG seal or product audit was found in the sources screened. [241, 211, 125] |
| Supply-path integrity | unconfirmed | E | No evidence found: no observed sample of Lotame curated-deal inventory with ads.txt, sellers.json or SupplyChain records was available, no request-duplication record was found, and no invalid-traffic filtering is documented for the curated marketplaces ('No MFAs' and 'brand-safe' are not IVT filtering). Lotame operates no exchange, and as a non-payment data and selection provider it cannot appear as a SupplyChain 1.0 node (brief, H6), so traceability would rest on the partner SSPs' files. Searched: Lotame's capability page, case page and PDF, Google's curation help page and launch post, 3 November 2026. [13, 24] |

Performance evidence: '20% better campaign efficiencies' in early trials of Lotame Curation (vendor assertion, sponsor Lotame (self-published release); author's former employer) [209]; 'improved video completion rates by 15%' in early trials (vendor assertion, sponsor Lotame) [209]; '21% higher viewability' in early trials (vendor assertion, sponsor Lotame) [209]; PHD EMEA case: 'a View-Through Rate consistently above 70%, beating industry benchmarks of 60%' (vendor assertion, sponsor Lotame (case study); agency quoted) [210, 125]; PHD EMEA case: 'projected cost savings of up to 33%' (forecast, sponsor Lotame (case study)) [210, 125]; 'No MFAs. No waste. No headaches.' (vendor assertion, sponsor Lotame (capability page)) [13]; 'Higher-than-your-DSP match rates to find your full audience on any device' (vendor assertion, sponsor Lotame (capability page)) [13]; 'Reach your most valuable customers in a streamlined, carbon efficient manner' (vendor assertion, sponsor Lotame (capability page)) [13]

### Magnite ClearLine (active)

*Magnite, Inc.*

Magnite's self-serve buying seat with curation built in; announced in beta in October 2025, with no published fee, auction rule or composition disclosure.

**SWOT (2026-11-03).** ClearLine adds buyer- and curator-built deal packages to Magnite's self-service CTV buying seat, but no page opened documents general availability, fees, auction defaults, deal composition, a DSP route or curated-deal volume.

- **Strengths.** The FY2025 10-K describes ClearLine as self-service direct access to CTV inventory with integrated curation, and says the Magnite Curator Marketplace within it launched in 2025. (record) [212] Magnite's September 2026 ClearLine Activate blog names 21 data and clean-room partners and six publishers; buyers apply audiences through line items, not new deal IDs. (vendor documentation) [108] On 14 October 2026 AdExchanger reported H/L's buyer agent activating LG Home Screen inventory with Magnite's seller agent within ClearLine over MCP; LG and H/L approved the plan. (vendor claim) [87, 172] Magnite, the Nasdaq-listed SSP behind ClearLine, reported FY2025 revenue of $714.0 million and net income of $144.6 million; CTV was 45.4% of its Contribution ex-TAC. (record) [212]
- **Weaknesses.** No page opened dates general availability: the 1 October 2025 release said curation was 'Coming soon in beta'; the FY2025 10-K says the Curator Marketplace launched in 2025. (record) [107, 212] No curation fee, curator margin, data fee or Magnite take on ClearLine curated deals is published; the 10-K describes its take rate qualitatively and states no rate. (record) [107, 108, 212] No auction type, priority, floor, exclusivity, request sampling or seller and domain disclosure is documented for ClearLine curated deals on any page opened. (vendor documentation) [107, 108, 213] No DSP is named as a route for ClearLine-built deals, and no page opened discloses curated-deal volume, a named curated-deal buyer or a measured quality result. (record) [212, 214, 107, 108, 215, 213]
- **Opportunities.** Culture Hive's July 2026 packages and Kinective Media by United Airlines' October 2025 plan to scale curated offerings in ClearLine point to curators adopting the tool. (inference) [213, 107] Magnite said in 2023 that DSPs would remain agencies' primary route to its video supply; a documented DSP route for ClearLine-built deals would reach buyers outside the seat. (inference) [62, 212] After a December 2025 seller-agent test, H/L's October 2026 agentic buy ran within ClearLine on single-publisher inventory; agent-led buying of multi-publisher curated packages is not yet documented. (inference) [90, 87] LG Ad Solutions' three-year extension of 14 October 2026 cites its expanded use of ClearLine, keeping LG's CTV supply within reach of ClearLine packages. (inference) [172]
- **Threats.** ClearLine lets buyers skip DSPs for some transactions while two unnamed buyers indirectly contributed about 44% of Magnite's 2025 revenue, so direct buying risks friction with concentrated demand. (inference) [212, 62] Adalytics documented Magnite-transacted PMP deal IDs, none identified as ClearLine deals, serving ads to declared data-centre bots as late as November 2024; buyers may discount deal packaging. (inference) [69] The filings say take rates run lower on CTV and on reserve auctions, including private marketplaces, which may cap Magnite's fee on curated CTV deals run that way. (inference) [212, 214] Culture Hive took its score to Index Exchange about a month before ClearLine, per PPC Land, so curators can carry the same signal to rival SSPs. (inference) [213]

- **What is sold.** A self-service buying and curation tool on Magnite's inventory. ClearLine gives agencies and advertisers 'direct access to premium CTV inventory and integrated curation capabilities' (FY2025 10-K); the Magnite Curator Marketplace inside it lets buyers and curators build 'custom deal packages for buyers that include curated pools of premium inventory that can be enriched with the curator's first-party or third-party data' (10-K). ClearLine Activate (CLA), an activation variant and alias of this row, lets buyers apply third-party, advertiser first-party and permissioned publisher first-party audiences 'through line items and demand tags' inside their CLA seat instead of requesting new audience-specific deal IDs (blog, 9 September 2026).
- **Who buys it.** Agencies and advertisers buying directly ('For ClearLine, advertisers and agencies use a self-service programmatic solution to directly access premium', 10-K); curators who package Magnite inventory with their own signals (the release of 1 October 2025 names 'buyers and curators'; Culture Hive announced CRS-informed packages in ClearLine on 21 July 2026, per PPC Land). Named users: launch partners Camelot, GroupM and MiQ (17 April 2023); independent agency H/L for the Northern California Toyota Dealers Association (October 2026, agentic buy of LG Home Screen inventory through ClearLine). No named buyer of a ClearLine curated deal was found.
- **How it earns.** At company level: 'Generally, our revenue is based on a percentage of the ad spend that runs through our platform'; IO campaigns are booked on the full ad spend; some fixed CPMs and fixed monthly fees; 10% of FY2025 revenue was reported gross (14% in 2024, 18% in 2023). The take rate is defined as 'our fee as a percentage of advertising spend', 'varies by publisher and transaction type', is lower for reserve auctions (direct deals and PMPs) and CTV and higher for managed service. No charge specific to ClearLine, ClearLine curation, curators or data partners is disclosed on any page opened, and the payer of any ClearLine fee is not stated.
- **Supply and data dependencies.** Supply: Magnite's seller base; the CLA blog names publisher partners LG Ad Solutions, Paramount, A+E Global Media, Scripps, AMC Networks and Wonder/Tastemade; ClearLine curation is 'built on the same underlying infrastructure as Magnite's SpringServe' (Magnite's CTV ad server and SSP; SpringServe is LG Ad Solutions' 'exclusive CTV video ad server' under the October 2026 extension). Data: Magnite Access audiences; onboarding and marketplace partners Experian, LiveRamp, TransUnion, Throtle and Cadent; direct third-party partners Acxiom, Comscore, Captify, Lotame, Nielsen, OpenAP, Samba, TruthSet, Genius Sports and TargetSmart; clean rooms and CDPs Snowflake, Databricks, Hightouch, Growthloop, Optable and Adobe; Cognitiv's deep learning models (January 2026); curator signals such as Culture Hive's Cultural Relevance Score; technology from the Streamr.ai acquisition (September 2025) for AI and agentic workflows. Demand: two buyers of advertising inventory indirectly contributed about 44% of Magnite's 2025 revenue (not named in the passage read). Samba (author's employer) and Lotame (former employer) appear in the partner list: interest marks.
- **Differentiation.** ClearLine curation runs on 'the same underlying infrastructure as Magnite's SpringServe', so 'curated campaigns are executed closer to the impression' (vendor assertion), on an exchange whose CTV Contribution ex-TAC was $304,192 thousand in FY2025, 45.4% of the three channels' total (derived from the 10-K). ClearLine is also a direct buying seat: since 2023 agencies can buy Magnite supply 'without the need for a DSP across certain kinds of inventory transactions' (GroupM quote), so a curated package can be activated where it is built. CLA names a large roster of data, clean-room and publisher partners and claims permissioned publisher first-party data inside the platform. Magnite pairs ClearLine with agentic tooling: a seller agent embedded in SpringServe (first test December 2025) and an executed agent-to-agent CTV buy inside ClearLine for H/L in October 2026.
- **Risks.** Channel conflict (inference): ClearLine gives buyers a route to Magnite supply outside DSPs while two buyers indirectly contributed about 44% of Magnite's 2025 revenue; the filing passage read does not name them. Fee opacity: with no published curation or data fees, a buyer cannot reconcile the charges in a ClearLine curated deal (H2b, Q3). Status ambiguity: beta announcement versus a 'launched' filing statement leaves the product's maturity undocumented. Supply-quality exposure: independent observations of Magnite-transacted PMP deals reaching declared bots (2023 to 2024) show that deal-ID packaging alone did not exclude such traffic on Magnite; whether ClearLine curation changes this is undocumented. Pending litigation: Magnite's damages suit against Google (filed 16 September 2025) is unresolved on the record opened.
- **Best fit.** CTV-focused agency buyers who want to buy Magnite streaming supply directly and apply named third-party, advertiser or publisher first-party audiences without requesting new deal IDs. Curators with their own signal (for example Culture Hive's score) who want to package Magnite inventory and let buyers transact it inside ClearLine. Buyers testing agentic CTV activation through a seller agent with human approval of the plan.
- **Poor fit.** Buyers who need a published fee schedule or a documented curator margin before paying for curation. Buyers who need documented auction type, priority, floor and exclusivity defaults for curated deals. Buyers who must see seller and domain lists or independent quality measurement for a named deal. Buyers who must activate curated deals through a specific DSP, since no DSP route is documented for ClearLine-built deals.
- **Limits.** Curation inside ClearLine was announced as 'Coming soon in beta' (1 October 2025); the 10-K says the Curator Marketplace launched in 2025, and no page opened dates general availability. No fee, curator margin, data fee or Magnite take on ClearLine curated deals is published; the 10-K's take-rate statements are corporate and unquantified. No auction type, priority, floor, exclusivity or request-sampling rule is documented for ClearLine deals. No buyer-facing composition disclosure (sellers or domains) is documented. No DSP route is named for deals built in ClearLine; the documented route is ClearLine's own self-service seat. No ClearLine-specific volume, revenue or deal count is disclosed; the filings give company-wide and channel figures only. No Magnite help-centre or developer documentation for ClearLine was located; the record rests on releases, blogs, filings and reporting.

| Dimension | Availability | Evidence | Observed |
|---|---|---|---|
| Selection transparency | beta | A | The curation release of 1 October 2025 says buyers and curators using ClearLine 'can define deal terms, pricing, and targeting directly' and package deals with first- and third-party audiences from Magnite Access, and the FY2025 10-K says the Magnite Curator Marketplace builds 'custom deal packages for buyers that include curated pools' of premium inventory 'enriched with the curator's first-party or third-party data'; the Cognitiv integration (6 January 2026) adds model-based bidstream enrichment, and Culture Hive (July 2026, via reporting) selects inventory into its packages with its Cultural Relevance Score. The logic is named (buyer- or curator-defined targeting, audience match, curator data, model scores) and is owned by the buyer or curator, but no threshold, rule definition or update cadence is documented on any page opened. [107, 212, 215, 213] |
| Composition disclosure | beta | D | No evidence found: searched the curation release (1 October 2025), the ClearLine Activate blog (9 September 2026), the ClearLine launch release (17 April 2023), the Cognitiv release (6 January 2026), the FY2025 10-K (characters 0 to 300,000 of 347,401) and the Q2 2026 10-Q, Magnite's Responsible Advertising page and PPC Land's Culture Hive report; none states whether a buyer can see the sellers or domains in a ClearLine curated deal. The CLA blog's six named publisher partners (LG Ad Solutions, Paramount, A+E Global Media, Scripps, AMC Networks, Wonder/Tastemade) are supply partners of the audience workflow, not the disclosed composition of an identified deal. [107, 108, 212, 62, 213] |
| Data provenance | live | D | The CLA blog (9 September 2026) names the data by type (third-party, advertiser first-party and 'permissioned publisher first-party data') and by provider: onboarding and marketplace partners Experian, LiveRamp, TransUnion, Throtle and Cadent; direct third-party partners Acxiom, Comscore, Captify, Lotame, Nielsen, OpenAP, Samba, TruthSet, Genius Sports and TargetSmart; clean rooms and CDPs Snowflake, Databricks, Hightouch, Growthloop, Optable and Adobe; the curation release adds Magnite Access audiences and the 10-K the curator's own first- or third-party data. No page documents per-segment collection method, intended use, permissions or legal basis, and the inputs of Cognitiv's models and Culture Hive's score are not described. Samba (author's employer) and Lotame (former employer) are in the list: interest marks, no capability credit from names alone. [108, 107, 212] |
| Quality evidence | beta | D | No quality claim with a named metric (MFA share, viewability, IVT, suitability, attention or carbon) and no measurement was found for ClearLine curated deals: the curation release, the Cognitiv release, the CLA blog and the 2023 launch release carry only 'premium' and benefit statements such as 'significantly increases spend going towards working media' (2023), and PPC Land reports no outcome data for the Culture Hive packages. The only independent observation opened that touches Magnite deals (Adalytics, examples October 2023 to November 2024) predates ClearLine curation, names no ClearLine deal and is recorded under supply-path integrity, not as evidence behind a ClearLine claim. [107, 108, 215, 62, 213] |
| Auction mechanics | beta | D | No evidence found: the curation release, the CLA blog, the 2023 launch release, the Cognitiv release, the FY2025 10-K, the Q2 2026 10-Q and PPC Land's Culture Hive report document no auction type, priority, floor treatment, exclusivity or request sampling for ClearLine curated deals; buyers and curators 'define deal terms, pricing, and targeting' and Culture Hive's buyers use 'the deal structures the platform already supports', neither naming a transaction type. The filings' statements that reserve auctions ('direct deals or private marketplaces with select buyers') carry a lower take rate are company-level fee statements, not deal mechanics; all five attributes are unknown. [107, 108, 214, 213] |
| Channel execution | live | D | Advertised scope: the FY2025 10-K describes ClearLine as 'direct access to premium CTV inventory and integrated curation capabilities', the 2023 launch release as 'premium video inventory', and the curation release only as Magnite's 'premium omnichannel footprint' (streaming TV in executive quotes; Magnite's boilerplate lists CTV, online video, display and audio for the company), so the product's scope is CTV and video with an undefined omnichannel extension. Documented live: CTV, through LG Home Screen inventory activated via ClearLine for H/L in fall 2026 (release and AdExchanger, 14 October 2026), and premium video via the Cognitiv statement; no web display, mobile in-app, audio or DOOH execution through ClearLine curation is documented. Read as CTV-only, one of one advertised environment is documented live; read as Magnite's four channels, one or two of four. [212, 62, 107, 172, 87, 215] |
| Activation routes | live | A | One named endpoint is documented: ClearLine's own self-service buying seat ('advertisers and agencies use a self-service programmatic solution to directly access premium', 10-K; buyers 'activate approved audiences directly within their CLA seat', blog), which H/L's buyer agent also used over MCP for the LG campaign (AdExchanger, 14 October 2026). No DSP is named as a route for deals built in ClearLine on any page opened; the 2023 launch release says 'DSPs will remain the primary method' for agencies buying Magnite's premium video generally, and Cognitiv's 'Dynamic Deal run through the DSP of your choice' describes Cognitiv's own platform, an unsupported any-DSP statement. [212, 108, 87, 62, 215] |
| Scale | beta | A | No evidence found of ClearLine curated-deal spend or impressions in any stated 12-month window: the FY2025 10-K and Q2 2026 10-Q disclose company-wide revenue ($713,953 thousand in FY2025) and channel figures (CTV Contribution ex-TAC $304,192 thousand in FY2025; CTV revenue $100,076 thousand in Q2 2026) but no ClearLine, curation or deal volume, and the curation release's '109 million US ad-supported households' is an unperiodised reach statement. These figures are context only under the dimension's rule. [212, 214, 107] |
| US availability and coverage | live | D | US availability is implied rather than stated: the New York-datelined curation release quotes Magnite's Group SVP, US Revenue on streaming TV 'where they can directly reach 109 million US ad-supported households', and the 2023 launch named launch partners Camelot, GroupM and MiQ. US production use is documented for ClearLine buying: the LG/Magnite release of 14 October 2026 names H/L as the first buyer to activate LG Home Screen inventory with Magnite's agentic capabilities for 'the Northern California Toyota Dealers Association', 'Timed to the fall football season', and AdExchanger reports the agents worked 'within ClearLine'; that buy is single-publisher inventory, not a curated deal, and the only curation-specific US record is a New York curator's July 2026 package announcement (via reporting) with no named buyer. [107, 62, 172, 87, 213] |
| Pricing transparency | beta | D | No charge for ClearLine or its curation is disclosed on any page opened: the curation release has buyers and curators set deal 'pricing' but names no curation fee, data fee or Magnite take on curated deals, the CLA blog names no data-partner charge, and the 2023 launch release offers only that ClearLine 'significantly increases spend going towards working media', a general assertion. At company level the FY2025 10-K defines the take rate as 'our fee as a percentage of advertising spend', says it 'varies by publisher and transaction type' and is lower for reserve auctions and CTV, and reports revenue 'generally' as a percentage of ad spend, without a rate, a payer statement for ClearLine or any curated-deal fee. [107, 108, 62, 212] |
| Independent assurance | live | B | TAG's release of 5 March 2026 lists 'Magnite', 'Magnite + DV / MDV+' and 'Magnite CTV / MCTV' among the companies certified in 2026 across its four seal programmes but does not state which seal each holds or whether it was self-attested or independently validated; the TAG Registry, which records both, was not opened. That is certification of the operating company at most, without ClearLine or curation scope; Magnite's Responsible Advertising page names no certification, and no audit of ClearLine composition or fees was found. [242, 243] |
| Supply-path integrity | unconfirmed | D | No observed sample of a ClearLine curated deal was found or taken, so authorisation (ads.txt, sellers.json, SupplyChain) and request duplication are unobserved for this product; Magnite self-describes inventory quality guidelines that 'address inventory authorization' and 'a tapestry of industry, proprietary, and third-party technology' to eliminate non-human traffic 'in real-time', naming no measurer. Adalytics' observational report documents Magnite-transacted PMP deal IDs (for example 'MGNI-MD-9-5198', 17 March 2024, and 'MGNI-AP-5083', 4 October 2023 and 24 November 2024) delivering ads to declared data-centre bots; those deals predate ClearLine curation, are not identified as ClearLine deals, and the authors note that 'not all bot traffic is fraudulent'. [243, 69] |

Performance evidence: In the ClearLine curation release, Mike Laband (Group SVP, US Revenue) speaks of streaming TV 'where they can directly reach 109 million US ad-supported households' (vendor assertion, sponsor Magnite) [107]; ClearLine 'significantly increases spend going towards working media' and partners describe 'tech tax efficiencies' (2023 launch) (vendor assertion, sponsor Magnite and quoted partners) [62]; CLA's permissioned publisher first-party data is 'a differentiated capability not available through other platforms' (vendor assertion, sponsor Magnite) [108]; The Cognitiv integration lets Magnite 'enrich the bidstream' and gives buyers 'more effective ways to plan, test, and activate custom curated deals' (vendor assertion, sponsor Magnite and Cognitiv) [215]; Culture Hive's CRS-informed packages in ClearLine carry no published outcome data (vendor assertion, sponsor Culture Hive and Magnite (release), relayed by PPC Land) [213]; H/L uses 'cost per reach as its main metric of success' for the agentic LG Home Screen campaign (vendor assertion, sponsor H/L and Magnite (interviews), reported by AdExchanger) [87]; Magnite's supply covers '96% of overall omnichannel supply' (Jounce March 2025 Supply Benchmarking Report, cited in Comscore's release) (vendor assertion, sponsor Comscore release citing a client-only Jounce report) [244]; Magnite-transacted PMP deal IDs delivered ads to declared data-centre bots (examples October 2023 to November 2024) (independent measurement, sponsor Adalytics (funding not stated)) [69]

### Microsoft Curate (active)

*Microsoft Corporation (the registrant). The operating unit is Microsoft Advertising, and the predecessor business was Xandr, Inc., which Microsoft acquired on 6 June 2022 (CUR-MICROSOFTCURATE-S12, CUR-MICROSOFTCURATE-S13).*

The most fully documented curation platform in the record: open-auction packaging, net-of-fee bidding, one buyer per deal and Microsoft-controlled request sampling.

**SWOT (2026-11-03).** Microsoft Curate, Microsoft Advertising's curator tool on Monetize supply, publicly documents its auction defaults, floors and fee stack without amounts, but no source gives a buyer-facing seller list, independent quality evidence, Curate-scoped assurance, curated-deal volume or a US availability statement.

- **Strengths.** Every curated deal is an Open Auction that never overrides seller floors or ad-quality settings; by default bids compete net of fees in the seller's final auction. (vendor documentation) [110, 111] Microsoft documents each charge's type, base and deduction order: optional curator margin, BrASC or MASC charged to the curator, seller-side SASC; curation charges apply only on wins. (vendor documentation) [115, 216, 111] Curators can set Fixed, Standard or Market floors above, at or below a seller floor; a nine-row table documents seller-by-curator floor outcomes. (vendor documentation) [112] Selection inputs are documented: Microsoft universal categories, seller categories, seller and publisher lists, blocklists, segments, geography and curator-set thresholds acting as hard limits on predicted viewability or completion. (vendor documentation) [110]
- **Weaknesses.** No fee amount or maximum margin value is public; the 10% SASC and 30% MASC are worked examples; no page says whether buyers see margins or tech fees. (vendor documentation) [115, 216] The buyer sees the deal name and an optional code; no page opened documents a buyer-facing list of the sellers, publishers or domains in a curated deal. (vendor documentation) [110, 111] Its product-page figures (74% lower CPA, 25% lower CPM, +12% viewability rate) carry no comparator or method; no independent study of a Curate deal was found. (vendor claim) [121] No curated-deal volume, US availability statement or Curate-scoped assurance was found; the 10-Ks do not break out the product, and Microsoft's MRC accreditation covers paid search clicks. (record) [217, 218, 219, 121]
- **Opportunities.** With Invest media buying announced to end from 28 February 2026 and Curate and third-party DSP access kept, more Monetize demand could flow through curated deal IDs. (inference) [185] Microsoft's September 2026 post courts retail media networks, data aggregators and publishers as curators, and buyers may curate across DSPs, which could widen demand for the tool. (inference) [220, 56] If deployed, the MCP-backed Copilot workflows announced on 10 September 2026 would let curators bring account-specific intelligence into Curate; no deployment or customer is yet described. (inference) [220]
- **Threats.** Curate's About and Curated Deals pages still list Invest as a buying route after its media buying's announced end, so buyers may misjudge which DSPs reach curated deals. (inference) [56, 111, 185] Curation Optimization samples deals buyers stop bidding on, with no threshold, rate or opt-out stated, so slow deals can lose bid requests on Microsoft's call. (inference) [113] Adalytics (2024) placed Microsoft Xandr among four exchanges that appeared to relay mis-declared Forbes inventory, naming no Curate deal; curated line items cannot filter on ads.txt, inviting scrutiny. (inference) [67, 110] After the 2023 rename, Curate's documentation paths and sign-in address still carry Xandr names, and the FY2026 10-K text read omits it, leaving its weight inside Microsoft unclear. (inference) [221, 56, 218]

- **What is sold.** Access to the Microsoft Curate tool, at curate.xandr.com, lets a curator build curated deal line items on Microsoft Monetize ('Microsoft Advertising Marketplace') supply and sell them to buyers as deal IDs. A line item packages supply 'from the open exchange or through deals negotiated directly with sellers', applies the curator's targeting and assets, and sets a price type (Fixed, Standard or Market) and an optional curator margin. It is exposed to one selected buyer, under a deal name the buyer sees (CUR-MICROSOFTCURATE-S03). Microsoft describes the product as a way for curators to 'use their proprietary assets to enhance the value of a seller's inventory and create unique offerings for buyers' (CUR-MICROSOFTCURATE-S01). The purchased service is the curation platform. The curated deals curators build on it are their products, not separate rows.
- **Who buys it.** Curators buy the tool. Microsoft names 'retailers, data companies, independent trading desks, and other media companies' (CUR-MICROSOFTCURATE-S01) and, in 2026, 'enterprise data owners, data aggregators, retail media networks, publishers' (CUR-MICROSOFTCURATE-S10). Buyers 'can act as curators' to centralise targeting across DSPs (CUR-MICROSOFTCURATE-S01). Onboarding is through a Microsoft Advertising representative (CUR-MICROSOFTCURATE-S01), although the product page calls it 'a self-serve solution' (CUR-MICROSOFTCURATE-S08). Buyers of the resulting deals are platform buyer members, one per curated deal line item (CUR-MICROSOFTCURATE-S03). The 2021 launch release named adoption by Playground XYZ, Omnicom Media Group in Australia and Dentsu Asia (CUR-MICROSOFTCURATE-S15). Permutive's documentation names 'Monetize' among the SSP curation seats through which its Permutive-branded curated deals are accessed (CUR-MICROSOFTCURATE-S18). That Permutive's 'Monetize' seat is a Microsoft Curate curator account is an inference.
- **How it earns.** Microsoft charges the curator for using Curate in one of two ways, set by contract. BrASC is 'a charge for enabling the curator to use Curate's suite of packaging and audience tools and access supply', calculated on 'the buyer's media cost after buy-side charges' and applied either as an invoice fee or as a deduction from media cost. MASC is 'applied as a percentage of the curator's gross margin'. Microsoft also takes its seller auction service charge (SASC), 'quoted as a percentage of the buyer's media cost', on the seller side (CUR-MICROSOFTCURATE-S07). Curation-related charges apply 'only if the bid for the curated deal wins the impression in the seller's final auction' (CUR-MICROSOFTCURATE-S02). No rate is disclosed: the 10% SASC and the 30% MASC are worked examples. Microsoft's 10-Ks report no revenue for Curate, Monetize or the former Xandr business in the portions read (CUR-MICROSOFTCURATE-S12, CUR-MICROSOFTCURATE-S13).
- **Supply and data dependencies.** Supply comes from sellers on Microsoft Monetize, either their open-exchange supply or deals they negotiate with the curator. Sellers keep control: curated transactions 'will not override any seller account configurations, including ad quality settings or floor price settings', and sellers can block curators in Partner Center (CUR-MICROSOFTCURATE-S02). Data comes from four sources: (1) the curator's own assets ('audience data' and segments, CUR-MICROSOFTCURATE-S01); (2) Microsoft's universal categories, applied 'based on the inventory's content' when Microsoft reviews inventory; (3) seller-defined categories; (4) Microsoft's predicted view and completion rates, which enforce curator thresholds, plus 'Cross-device enhanced by Microsoft', switched on by default when a segment is targeted (CUR-MICROSOFTCURATE-S03). For demand, the documentation names Microsoft Invest and 'all major DSPs' (CUR-MICROSOFTCURATE-S01, CUR-MICROSOFTCURATE-S02), and Microsoft stopped supporting media buying through Invest from 28 February 2026 (CUR-MICROSOFTCURATE-S09). Third-party curators depend on the platform. Permutive documents Prebid.js as 'Required for Monetize, Index Exchange, and Pubmatic curation activation' (CUR-MICROSOFTCURATE-S18).
- **Differentiation.** Among the cohort A products the streams screened, this is the most complete public documentation of curated-deal auction mechanics and fee stacking (streams E and H, re-opened here): auction type, floor types with a seller-by-curator floor outcome table, net-of-fee bidding, deduction order, and the named Microsoft charges (BrASC, MASC, SASC) (CUR-MICROSOFTCURATE-S02 to S07). It documents a platform-run request-sampling mechanism for curated deals (Curation Optimization), the first case the brief names. Stream E's verifier found parallel wording in Criteo's Commerce Grid knowledge base (CUR-MICROSOFTCURATE-S06). In-flight enrichment by the exchange is on by default: 'Cross-device enhanced by Microsoft' is applied free whenever a segment is targeted (CUR-MICROSOFTCURATE-S03). Buyers can act as curators to centralise business rules across DSPs, so the same tool serves sell-side data owners and buy-side desks (CUR-MICROSOFTCURATE-S01). Microsoft announced Copilot 'MCP-backed capabilities' and 'agentic workflows' for Curate on 10 September 2026. That is an announcement, not documented production use (CUR-MICROSOFTCURATE-S10).
- **Risks.** Route risk: the documentation (ms.date 2025-10-21) still names Microsoft Invest, whose media buying Microsoft stopped supporting from 28 February 2026, and names no third-party DSP. A buyer cannot establish from public records which DSPs can activate a Microsoft Curate deal (CUR-MICROSOFTCURATE-S01, CUR-MICROSOFTCURATE-S09). Fee opacity to the buyer: the curator can edit margins 'at any time', the maximum margin and Microsoft's rates are contractual, and no page says the buyer sees either. A fee-sensitive buyer would need contractual disclosure (CUR-MICROSOFTCURATE-S05, CUR-MICROSOFTCURATE-S07). Sampling opacity: Microsoft decides when to sample a deal, and the documentation tells curators that falling bid requests reflect sampling, 'not due to the available supply' (CUR-MICROSOFTCURATE-S06). Supply-path: an independent 2024 observation placed Microsoft Xandr among exchanges that relayed mis-declared inventory, some via PMP deal IDs, and curators cannot filter by ads.txt in a curated line item (CUR-MICROSOFTCURATE-S17, CUR-MICROSOFTCURATE-S03). Disclosure and strategic-weight risk (inference): Microsoft's FY2026 10-K text read does not name Curate, Monetize or Xandr, the documentation still sits under /xandr/ paths with sign-in at curate.xandr.com three years after the 2023 rename, and Microsoft exited DSP media buying. Public records therefore give no basis to judge the product's commercial weight inside Microsoft (CUR-MICROSOFTCURATE-S13, CUR-MICROSOFTCURATE-S01, CUR-MICROSOFTCURATE-S11).
- **Best fit.** Data owners (retailers, data companies, publishers, data aggregators) that want to layer their own audiences over multi-seller Monetize supply and sell the result as deal IDs with a configurable margin (CUR-MICROSOFTCURATE-S01, CUR-MICROSOFTCURATE-S10). Trading desks and large buyers that want to keep targeting rules and inventory lists in one curation layer across DSPs, Microsoft's stated buyer-as-curator use (CUR-MICROSOFTCURATE-S01). Curators who need documented auction and fee mechanics (open-auction packaging, seller floors respected, net-of-fee bidding, win-only charges) before contracting (CUR-MICROSOFTCURATE-S02 to S05).
- **Poor fit.** Buyers who need a documented seller and domain list for each deal, or a check of it against sellers.json (no buyer-facing composition disclosure documented). Buyers who need a documented activation path in a named third-party DSP (none named; Invest withdrawn). Buyers who need independent quality evidence or scope-covering assurance for the curated deal (none found). Buyers who want priority, first look or exclusive supply by default: the documented defaults give none, though a curator can build on seller deals with 'preferred inventory access' or 'specially negotiated rates' (CUR-MICROSOFTCURATE-S01, CUR-MICROSOFTCURATE-S02).
- **Limits.** A curated deal line item has one buyer, and 'You cannot change the selected buyer after the curated deal line item is created and saved' (CUR-MICROSOFTCURATE-S03). The product page's 'assign deals to multiple buyers on any DSP' (CUR-MICROSOFTCURATE-S08) is not documented as a mechanism. ads.txt targeting is 'not supported for curated deal line items' (CUR-MICROSOFTCURATE-S03). No third-party DSP is named as a route on any Microsoft page opened, and the one named route, Microsoft Invest, lost media-buying support from 28 February 2026 by Microsoft's announcement (CUR-MICROSOFTCURATE-S01, CUR-MICROSOFTCURATE-S09). No fee amount, rate card or maximum margin value is public, and buyer visibility of the margin and tech fee is undocumented (CUR-MICROSOFTCURATE-S05, CUR-MICROSOFTCURATE-S07). Curation Optimization can reduce the impressions matched and bid requests seen on low-spend deals. No threshold, rate or opt-out is documented (CUR-MICROSOFTCURATE-S06). Onboarding goes through a Microsoft Advertising representative, and opting out of cross-device needs an account representative (CUR-MICROSOFTCURATE-S01, CUR-MICROSOFTCURATE-S03). Legacy insertion orders are not supported, insertion orders must carry an unlimited budget, and the payment model is preset to Pay Per Impression (CUR-MICROSOFTCURATE-S03). The documentation warns that 'Some clients might see different options when creating a curated deal line item', so the documented workflow is a superset of what any one curator sees (CUR-MICROSOFTCURATE-S03). No buyer-facing seller or domain list, independent quality study, scope-covering assurance, curated-deal volume or US availability statement was found in the sources opened.

| Dimension | Availability | Evidence | Observed |
|---|---|---|---|
| Selection transparency | live | C | Vendor documentation describes the curator's selection toolset in operational terms. It covers universal categories that Microsoft defines and applies 'based on the inventory's content' when it reviews inventory, seller-defined categories, and include or exclude lists of sellers and exposed publishers. It also covers inventory type (App & Web by default, App Only, Web Only), allowlists and blocklists ('by default, the AppNexus Blocklist will always be applied'), and buy-side targeting with segments and geography. Viewability or completion-rate thresholds act as a 'hard limit' on predicted rates. The curator sets the thresholds. Microsoft's prediction models, the review cadence for categories and the update cadence for a deal are not documented, and ads.txt filtering is 'not supported for curated deal line items'. The record documents a capability, not the rules of an identified deal. [110, 111] |
| Composition disclosure | live | C | The buyer is shown the deal name ('this curated deal name is exposed to the buyer') but not the line-item name, and an optional deal Code is 'typically used by external buyers for deal mapping'. No page opened documents a buyer-facing list of the sellers, publishers or domains in a curated deal. None documents an identified example deal or delivery verification either. The transparency functions documented belong to sellers and curators: sellers can block curators and report on 'is curated' and 'curator member ID', and the product page describes curator-side reporting. Buyer-side composition disclosure is therefore access unknown, not documented absence. [110, 111, 121] |
| Data provenance | live | C | Data inputs are named by type and owner, but not per segment. Curators bring 'audience data' and target segments. Microsoft applies universal categories from its review of inventory content, a non-personal contextual input with Microsoft as provider. Sellers define seller categories. 'Cross-device enhanced by Microsoft' switches on 'automatically' and 'at no cost' whenever a segment is targeted, and only an account representative can turn it off. No page opened gives a segment's source, collection method, intended use, permissions, consent route or legal basis, or the source of Microsoft's cross-device graph. In anchor terms the record names source types and providers at the level of the tool, never per segment. [56, 110] |
| Quality evidence | live | D | Every quantified quality claim is a vendor assertion without qualifying measurement: - The product page shows '74%' 'Lower CPA', '25%' 'Lower CPM' and '+12%' 'Viewability rate', footnoted '2023 Microsoft Curate first-party data', with no comparator, sample, period within 2023, geography or method. - The line-item page says to 'expect to see an average view/completion rate 10-15% higher than the threshold you set'. - Microsoft's 10 September 2026 post cites Experian figures on data-enriched PMPs that are not specific to Microsoft Curate. The one third-party observation found is Adalytics, 2024, whose funding is not stated. It places Microsoft Xandr among four exchanges relaying mis-declared, MFA-classed Forbes subdomain inventory, some via apparent PMP deal IDs. It identifies no Microsoft Curate deal. No independent or commissioned study with method and comparator was found for a Microsoft Curate deal. [121, 110, 220, 67] |
| Auction mechanics | live | C | Each mechanic is documented with its default: - Auction type: 'always set to "Open Auction" for curators'. A Dynamic CPM deal 'competes in either a first or second-price auction at the price bid by the buyer', depending on the seller. - Priority: none by default. Curated transactions 'are not guaranteed deals', never override seller floor or ad-quality settings, and compete 'in the seller's final auction with all other deal and open exchange bids, net of all applicable fees and charges'. Net-of-fee submission is a default 'unless you agree otherwise with the applicable curator'. - Floors: configurable as Fixed, Standard or Market, set 'above, at, or even below a seller floor'. A fixed-price deal on a second-price seller may clear lower than the configured price. - Exclusivity: one buyer per line item, fixed after saving, and the same supply stays biddable on the open exchange. - Request sampling: Microsoft-controlled Curation Optimization samples deals a buyer has stopped bidding on, with no threshold, rate or opt-out stated. [110, 111, 112, 113] |
| Channel execution | unconfirmed | C | Workstream U captured no advertised channel scope for Microsoft Curate: its About page states none (CUR-U-S04, re-opened here as CUR-MICROSOFTCURATE-S01). The line-item documentation offers Ad Type Banner, Video ('includes audio types as well') and Native, and Inventory Type App & Web, App Only or Web Only. On the paper's taxonomy these map to web display, web video, mobile in-app and audio, but as configurable settings, not documented live delivery. Microsoft's 10 September 2026 post adds 'Premium supply, including CTV inventory, through Microsoft Monetize' (marketing). DOOH is not mentioned, and no page documents live execution per environment. [110, 220, 56] |
| Activation routes | unconfirmed | C | The documentation names one buying route, Microsoft Invest, plus 'all major DSPs' ('Curated deals can be bought via Microsoft Invest and all major DSPs'). The product page adds 'available via all major demand-side platforms (DSPs)' and 'assign deals to multiple buyers on any DSP', and the 2026 post mentions 'integrations with major third-party DSP partners'. None of these names a DSP. Microsoft said on 14 May 2025 that it 'will no longer support media buying through our DSP, Microsoft Invest, starting on February 28, 2026'. At the cutoff, therefore, the only named route has been withdrawn by Microsoft's own announcement, and execution after that date is unconfirmed. No third-party DSP is named with documented activation functionality on any Microsoft page opened. [56, 111, 121, 185, 220] |
| Scale | live | E | No evidence found. Searched: the seven Microsoft Learn Curate and Monetize pages, the product page, Microsoft Advertising's posts of 18 June 2023, 14 May 2025 and 10 September 2026, and Microsoft's FY2022 and FY2026 10-Ks in the portions read. None discloses curated-deal spend, impressions, deal count or curator count for Microsoft Curate in any window. The 10-Ks report only Search (and news) advertising revenue growth excluding traffic acquisition costs (27% in FY2022, 12% in FY2026), a line not shown to contain Curate. The 2026 post's EMARKETER figure on private-marketplace share of RTB spend is market context, not Curate scale. [121, 217, 218, 220] |
| US availability and coverage | live | E | No evidence found. Searched: the seven Microsoft Learn pages, the product page, the three Microsoft Advertising posts and the two 10-Ks. No page states that Microsoft Curate is available in the US or names a US customer or US deals in a period. The product page speaks of a 'global marketplace'. The 2026 post makes no market claim: its only US reference is the site's region selector. The line-item page lets curators target countries without listing any. The 2021 launch release names adoption in Australia and by Dentsu Asia. Context only, not product evidence: Microsoft is a Washington corporation headquartered in Redmond, and Adalytics observed Microsoft Xandr relaying a US publisher's inventory between 2017 and 2024. [56, 121, 220, 245] |
| Pricing transparency | live | C | Every charge the product controls is documented by type, base, payer, collector and order of deduction, without amounts: - Curator margin: none by default. Set as 'The percentage of the buyer's bid (Curator Revenue)' or 'A fixed CPM deducted from the buyer's bid', editable 'at any time', and capped by a 'maximum margin rate' whose value is not stated. - Microsoft's charge to the curator: BrASC (a percentage of 'the buyer's media cost after buy-side charges', taken as an invoice fee or a deduction per contract) or MASC (a percentage of the curator's gross margin). - Microsoft's seller charge, SASC: a percentage of the buyer's media cost. Charges apply only when the curated bid wins. The only figures are worked examples (SASC 10%; MASC 30% of $0.18 stated as $0.06, which computes to $0.054). No page states whether the buyer sees the margin or the tech fee. Cross-device enrichment is 'at no cost'. [115, 216, 111, 112] |
| Independent assurance | live | A | No assurance whose scope covers Microsoft Curate or Microsoft Monetize was found. The MRC's letter of 31 January 2025 continues accreditation of 'Microsoft Advertising' for 'Desktop and Mobile Web Paid Search Clicks and Sophisticated Invalid Traffic (SIVT) functionality applied to them'. A TAG Certified Against Fraud addendum covers domain-list and data-centre IP filtering for that same scope. This is an accreditor's record of company-level assurance for a different service and does not extend to curated deals. TAG's registry was not opened; two web searches found no TAG seal or MRC record for Monetize, Xandr or Curate. No audit of a curated deal's composition or fees was found. [219] |
| Supply-path integrity | live | C | No observed sample of Microsoft Curate deals checked against ads.txt, sellers.json or SupplyChain was found or made for this profile. The documentation says three things: - ads.txt targeting is 'not supported for curated deal line items', so a curator cannot add an ads.txt filter. - The AppNexus Blocklist is always applied. - Auctions now carry 'industry IDs, schains, and, in recent years, curated deals'. No page opened describes Monetize's ads.txt enforcement, IVT filtering for curated deals, or how a curator appears in sellers.json or the schain. One independent observation exists: Adalytics, 2024, funding not stated, found Microsoft Xandr among four exchanges that appeared to relay www3.forbes.com inventory as www, some of it through apparent PMP deal IDs. It does not identify a Microsoft Curate deal, so it is recorded as counterevidence and not as a finding about this product. [110, 113, 67] |

Performance evidence: 'Curate by the numbers': '74%' 'Lower CPA', '25%' 'Lower CPM', '+12%' 'Viewability rate', footnoted '2023 Microsoft Curate first-party data'. (vendor assertion, sponsor Microsoft Advertising (self-published product page)) [121]; With a viewability or completion-rate threshold set, curators should 'expect to see an average view/completion rate 10-15% higher than the threshold you set'. (vendor assertion, sponsor Microsoft Advertising (documentation)) [110]; Microsoft's 10 September 2026 post says advertisers see '1.5x-3.0x higher CTR, and 10-30% higher video completion rate' when using data-enriched PMPs rather than the open exchange, citing Experian (June 2025). (vendor assertion, sponsor Experian (partner content), cited by Microsoft Advertising) [220]; A case study cited in the same post ('Telfast Achieves 17x Operational Efficiency Through Curated Buying', April 2025) reports one buyer 'reducing object complexity from 250 line items to 14 line items'. (vendor assertion, sponsor Microsoft Advertising) [220]; 'Buyers will tend to bid more often and with higher bids on impressions that come with a deal floor.' (vendor assertion, sponsor Microsoft Advertising (documentation)) [112]; Curation Optimization 'ensures that each auction focuses on processing the most valuable curated deals. In this way, buyers, curators, and sellers will achieve better results.' (vendor assertion, sponsor Microsoft Advertising (documentation)) [113]; Adalytics reports that 'four ad exchanges - Microsoft Xandr (fka AppNexus), Triplelift, Pubmatic, and Magnite - appear to incorrectly relay' www3.forbes.com inventory, classed by DeepSee.io as made-for-advertising, as the 'www.' subdomain, and that 'Some of this mis-declared inventory was observed transacting via what appear to be Private Marketplace (PMP) deal IDs'. (independent measurement, sponsor Adalytics Research (funding not stated; a commercial ad-analytics vendor, so independence is not established)) [67]

### PubMatic Connect (active, listed, independent)

*PubMatic, Inc.*

Curation run inside PubMatic's SSP, sold as PubMatic-built packages, self-serve deals and custom curation, with fees described only in a CEO interview.

**SWOT (2026-11-03).** PubMatic Connect runs curation inside PubMatic's own SSP through three build routes, but publishes no fee amounts, no seller list for a named deal, no named third-party DSP route, no periodised curated volume and no independently measured quality.

- **Strengths.** Connect runs inside PubMatic's own SSP, which hosts the deal, applies the data and, per its console page, 'manages the revenue share with third-party partners'. (vendor documentation) [222, 223] Buyers get three build routes under one name: PubMatic-built Auction Packages, self-serve multi-publisher deals under one deal ID in the Media Buyer Console, and staff-built custom curation. (vendor documentation) [224, 222, 225] In March 2025 PubMatic's CEO said curation gets no different floor rules; a curator wins only 'if the net CPM to the publisher is higher than the alternative'. (vendor claim) [226] Partner documents show curated deals on PubMatic enabled for CTV (Comscore with Adelaide, 10 June 2025) and built for desktop and mobile web (Adloox, 17 July 2026). (vendor documentation) [227, 12]
- **Weaknesses.** No Connect fee rate, schedule or payer is public; the CEO named a 'core SSP fee' plus a data fee 'based on the curator's fee', negotiated per partner. (vendor claim) [226, 222] No opened record shows a Connect deal's seller list or names a third-party DSP for one; PubMatic's own Activate is the only named route. (vendor documentation) [223, 228, 222, 12, 229] Every quantified quality or performance figure is a vendor or agency assertion without published method; none measures a quality dimension against a pre-curation comparator. (vendor claim) [119, 230] The 10-K and 10-Q call Connect only 'our solution that provides additional data and insights to buyers' and disclose no Connect revenue, curated-deal volume or take rate. (record) [231, 232]
- **Opportunities.** Announced for buyers and curators on 28 April 2026, the Deal Management agent already credited with over 1,000 publisher deals could bring more self-serve package building. (inference) [233] PubMatic's March 2025 claim, without stated method, of 25%+ higher eCPMs than non-curated deals gives it a publisher-side argument for staying opted into curation. (inference) [230] Adloox (July 2026) and Comscore with Adelaide (June 2025) offer data products in deals on PubMatic's SSP, a route other data owners could use without a curation desk. (inference) [12, 227]
- **Threats.** PubMatic sells its own Auction Packages on an exchange that also charges a data fee 'based on the curator's fee', competing with the curators it earns from. (inference) [226, 224] Publishers are opted into curation by default and can opt out, so package supply can shrink if publishers object, and no public record lists opt-outs. (inference) [226] For buyers the agent 'handles the rest'; the blog shows its review-and-approval step only in a publisher example, so a misread instruction could become a deal configuration. (inference) [233] With a $14.5 million FY2025 net loss and SPO rebates netted from revenue, buyer pressure on undisclosed curation fees would fall on a loss-making company. (inference) [231, 232]

- **What is sold.** Connect is PubMatic's audience and curation layer on its own SSP. On the buyer side it is sold as three things that share the Connect name: (1) PubMatic-built 'seasonal and "always on" Auction Packages' (storefront), multi-publisher packages 'transacted via a single deal ID'; (2) self-serve deal building in the Media Buyer Console, where buyers and curators 'set up and manage multiple Publisher PMPs into one Deal ID', add audience data from Connect or 'chosen providers', and since 28 April 2026 can ask PubMatic's Deal Management agent to create auction packages in natural language; (3) custom curation by PubMatic's Customer Success and 'dedicated global audience team'. On the sell side, 'Connect for Publishers' sells data owners and publishers 'access to Deal ID transactions' and a near-real-time view of their data deals. The filings describe Connect only as 'our solution that provides additional data and insights to buyers'.
- **Who buys it.** Advertisers, agencies and curators on the buy side (the buyers page says the platform 'directly connects advertisers, agencies, and curators'); named users in opened vendor material are agencies and holding-company units (Agence 79 of Havas Media Network in France; Omnicom Media Group Germany; IPG Mediabrands; Goodway Group; Publicis Groupe India) and data or measurement partners that build deals on PubMatic (Adloox via Qwarry; Comscore with Adelaide). On the sell side, publishers and data owners supply inventory and data; per PubMatic's CEO (AdExchanger, 7 March 2025) publishers are opted into curation by default and can opt out. The filings list 'data partners and curators' as one of four primary customer types.
- **How it earns.** Company level (filings): PubMatic generates revenue 'through fees charged to our publishers, which are generally a percentage of the value' of impressions monetised, reported net ('gross billings to buyers, net of amounts we pay publishers'); no numeric take rate and no Connect revenue line appear in the portions read. Curation transactions (CEO interview, AdExchanger, 7 March 2025, reporting): 'We have our core SSP fee' plus 'a data fee in the curation transaction based on the curator's fee', 'negotiated partner by partner'; the curator sets its own fee and the buyer's CPM 'is declared in the SSP when the curator sets up the deal'. The Media Buyer Console page says PubMatic 'manages the revenue share with third-party partners', so partner payments flow through PubMatic's billing. No rate, schedule or payer statement for any of these charges is public in the sources opened.
- **Supply and data dependencies.** Inventory: PubMatic's own SSP supply (company-wide 'direct access to 1,900 premium publishers' and 'over 821 billion daily ad impressions' on the buyers page; '800 billion+ daily omnichannel impressions' in March 2025), with publishers opted in to curation by default. Data: buyers' first-party data, publisher data and 'specialized partner' data ('more than 190 premium data partners' in March 2025, including commerce media networks Instacart, Intuit and Western Union; 'Access to 30+ Unique DMP/Publisher Data Partners' on an undated US package page), partner segments selectable in the console (Adloox attention segments via Qwarry at $0.15 CPM) and partner deal products enabled in the SSP (Comscore-Certified Deal IDs with Adelaide's AU, June 2025). Demand: PubMatic's own Activate or 'your DSP of choice'; no DSP is named in any opened Connect record. Identity: 'emerging identifiers' and 'any advertiser preferred user identifier'.
- **Differentiation.** Curation runs inside the SSP itself: PubMatic hosts the deal, applies the data and bills partners ('PubMatic manages the revenue share with third-party partners'), rather than a third-party curator layering data onto another exchange (vendor documentation, S08; CEO, S05). Three build routes under one name: PubMatic-built Auction Packages, self-serve multi-publisher deals in the Media Buyer Console, and custom curation by PubMatic staff (S06, S08, S10). A natural-language Deal Management agent for building auction packages, live for buyers and curators from 28 April 2026 inside AgenticOS and over MCP, with human approval of each configuration (S09). A stated policy of equal floor rules for curated and non-curated demand and net-CPM competition for the publisher ('the only way the curator can win is if the net CPM to the publisher is higher than the alternative'), on the CEO's word (S05). An open partner model: PubMatic says it focused on 'enabling the business model of curators, rather than on PubMatic as a curator' (190 curation partners, March 2025), alongside its own packages (S05).
- **Risks.** Self-dealing perception (inference): PubMatic builds and sells its own Auction Packages on the exchange it operates while also hosting third-party curators who pay it a data fee 'based on the curator's fee', so it competes with and earns from the curators it hosts (S05, S06). Default publisher opt-in (S05) means a publisher can be in curated packages without an affirmative choice, a composition and consent question if a buyer or publisher later disputes inclusion (inference). Fee pressure on curation: PubMatic's CEO accepts that curator fees are set by 'the market'; a buyer cannot reconcile the stack from public sources (S05). Company economics: FY2025 net loss of $14.5 million after net income in 2024, and more than half of activity under SPO agreements that carry rebates (S11, S13); curation revenue is not separately disclosed. Agentic configuration: the Deal Management agent drafts deal settings that a human approves; an error in a natural-language instruction becomes a deal configuration (inference; S09).
- **Best fit.** Buyers who want multi-publisher packages on one SSP under one deal ID and are willing to buy through PubMatic's console or Activate (inference from S03, S08). Data and attention partners that want to sell segments into deals built on PubMatic's exchange without running their own curation desk (Adloox via Qwarry; Comscore) (S14, S16). Agencies that value sell-side data application and PubMatic-managed billing of partner revenue shares (S08, S10).
- **Poor fit.** Buyers who need a published fee schedule, a seller list for a named deal or an independent quality audit before buying (none public; S05, S08). Buyers who need a named third-party DSP route documented for a curated deal (none named; S03, S08). Buyers who need multi-exchange reach from one curated deal: Connect packages run on PubMatic's own supply (inference from S03, S06).
- **Limits.** No public fee rate, schedule or payer statement for any Connect charge; charge types come from a CEO interview relayed by trade press (grade D). Auction type, priority, exclusivity and request sampling for Connect auction packages are undocumented in public sources; PubMatic's help documentation (help.pubmatic.com) was not reached and the console is authenticated. No identified Connect deal with a public seller or domain list; composition visibility is asserted for the authenticated console only. No DSP is named as an activation route for Connect deals; the only named route is PubMatic's own Activate. Every quality or performance figure is a vendor or agency assertion without published method; none measures a registered quality dimension against a pre-curation comparator. The filings describe Connect as 'additional data and insights to buyers' and disclose no Connect revenue, curated share or take rate. Publishers are opted into curation by default (CEO statement), so what a package contains depends on publisher opt-outs that no public record lists.

| Dimension | Availability | Evidence | Observed |
|---|---|---|---|
| Selection transparency | live | D | PubMatic's pages name selection logics for its own auction packages without thresholds or update cadence: audience data (buyer first-party, publisher and partner data), vertical and category customisation, an 'OMSDK ‘Measurable’ Package', an 'Inventory Quality Score Marketplace', a 'High Video Completion Rate Package', device-ID types and live-sports content packages (MLB, NFL and NCAA football, CTV Sports). The only thresholds found are in a partner's documentation: Adloox's attention segments in PubMatic's console are tiered 'Top 20%' and 'Top 40%' and 'updated daily'. No PubMatic document opened states what score, rate or list qualifies inventory for a named package. [228, 224, 246, 12] |
| Composition disclosure | live | D | A disclosure capability is asserted in marketing for the authenticated Media Buyer Console: an 'Inventory Discovery Tool' showing 'available inventory from the past 30 days', 'Log-Level Transaction Verification' and 'Two-Way Fee Transparency' letting buyers 'know exactly where the inventory is coming from'; the Deal Management agent accepts 'publisher lists', and Adloox's console steps include choosing or uploading publishers when building a deal. No identified example deal with a seller or domain list was found: the US package page read lists package types, not sellers, and the US CTV package page could not be read (access error). Stream E's access-unknown coding (CUR-E-F34) is superseded for the capability statement only. [222, 233, 12, 228, 229] |
| Data provenance | live | D | Source types and some providers are named: buyers' first-party data, publisher data and 'specialized partner' data; 'more than 190 premium data partners' with 'purchase-based insights' from Instacart, Intuit and Western Union (March 2025); Adloox attention segments hosted by Qwarry; Comscore-Certified Deal IDs combining Comscore signals, Proximic targeting and Adelaide's AU. No opened record documents, per segment, the collection method, intended use, permissions or legal basis, and contextual or non-personal inputs are not marked as such; the site footer states that 'PubMatic is a registered data broker in the state of Texas'. [246, 230, 228, 222, 12, 227] |
| Quality evidence | live | D | Every quantified claim found is a vendor assertion: the Agence 79 campaign (CTR five times 'the same campaign on the open market', VCR +16.7% and viewability +12.5% against the agency's targets, 'Source: Agence 79 internal data', no period, volume, measurer or method; activated through PubMatic's direct-to-supply platform, Connect not named); the March 2025 release's yield and price figures (25%+ eCPM against non-curated deals, CPM 'improve by 37%' against buy-side data application) with no sample, period or method; and a 2022 Lotame case on the Connect for Buyers page (interest mark). No independent study of a Connect deal's MFA share, viewability, IVT, suitability or attention against a pre-curation comparator was found in the sources screened. [119, 230, 246] |
| Auction mechanics | live | D | Floor treatment is documented only as a reported CEO statement: 'we don't have any different rules on floor pricing when it's a curated data set or inventory', with publishers able to push floors up or enforce them downstream of the SSP, and 'the only way the curator can win is if the net CPM to the publisher is higher than the alternative'; publishers are opted into curation 'by default' and can opt out. The buyer and DSP are chosen when an auction package deal is created (Adloox's console steps, paraphrased by the fetch tool). Auction type, priority, exclusivity and request sampling for Connect deals are not documented in any source opened; a 'priority P6' setting appears only in a publisher PG example for the Deal Management agent. [226, 12, 233] |
| Channel execution | live | D | Advertised scope (buyers page): 'CTV, video, mobile app, and display', which maps to web display, web video, mobile in-app and CTV on the paper's taxonomy (the March 2025 release lists 'CTV, mobile app, and web display'; the console page adds OTT). Dated execution records: CTV deal IDs enabled in PubMatic's SSP 'including many of the top CTV publishers' (Comscore release, 10 June 2025, partner) and a live CTV Sports package on the storefront; desktop and mobile-web deal building with an instruction to avoid CTV (Adloox documentation, 17 July 2026, partner); a programmatic video campaign in France (11 June 2026) whose environment within video is not stated. Mobile in-app appears only as listed package types ('In-App', 'OMSDK ‘Measurable’ Package') with no dated execution record. [223, 230, 222, 227, 12, 119, 224, 228] |
| Activation routes | live | D | The buyers page says 'Curated deals can be activated through PubMatic’s Activate solution or your DSP of choice', and the console page says 'PubMatic is fully integrated with all major DSPs', an unsupported 'any DSP' assertion. The only named buying endpoint with a stated activation for curated deals is PubMatic's own Activate (and the Agence 79 campaign ran on PubMatic's 'unified direct-to-supply platform'); no DSP is named in any opened Connect, auction-package, Comscore or Adloox record, the Adloox steps only say a DSP is selected at set-up. Activate is the register's comparator row and is a PubMatic route, not a third-party DSP. [223, 222, 119, 12, 227] |
| Scale | live | D | No evidence found: searched the Connect, auction-package, Media Buyer Console and buyers pages, the March 2025 release, the March 2025 AdExchanger interview, the FY2025 10-K (characters 0 to 100,000 and 200,000 to 300,000) and the Q2 2026 10-Q for a curated-deal spend or impression figure in a stated 12-month window; none was found. The only Connect-specific volume statement is AdExchanger's report that Connect 'grew by 140% year over year in Q4' (a growth rate without a base); company-wide figures (1,900 publishers, over 821 billion daily impressions, revenue) and the Deal Management agent's 'more than 1,000 AI-powered deals, transacting millions in platform spend' (publisher deals since August 2025) are context only. [226, 223, 233, 231, 232] |
| US availability and coverage | live | D | US availability is shown by vendor marketing only: the auction-package storefront has region pages under /auction-packages/us/ (a 'Video & VCR Refresh' US package was read) and US-sports packages ('curated NFL + NCAA Football auction packages', MLB), and the Comscore partner release carries a Reston, VA dateline. No opened record names a US customer with a period for a Connect deal: the IPG Mediabrands and Goodway Group quotes give no geography or period, Minute Media's quote gives none, and the dated campaigns found are in France (Agence 79) or run through Activate (the Butler/Till campaign in stream J). Stream U's 'not stated' coding is superseded only by the US region pages. [224, 228, 227, 230] |
| Pricing transparency | live | D | Charge types are named but no amount, rate or schedule is published: PubMatic's CEO describes 'our core SSP fee' and 'a data fee in the curation transaction based on the curator's fee', 'negotiated partner by partner', with the curator setting its own fee and the buyer's CPM 'declared in the SSP when the curator sets up the deal' (reporting, 7 March 2025). The console page says PubMatic 'manages the revenue share with third-party partners' and offers 'Two-Way Fee Transparency' (a general assertion); the 10-K gives the corporate basis only (publisher fees 'generally a percentage of the value'). The only priced input found is a third-party data fee (Adloox segments at '$0.15/CPM', payer and collector not stated), an attribute rather than a PubMatic charge; no Deal Sync API curfeetype declaration by PubMatic was found. [226, 222, 231, 12] |
| Independent assurance | live | B | TAG's release of 4 March 2026 lists PubMatic 'Among the companies that achieved TAG Certification Seals for 2026' without naming the seal or stating whether it was independently validated; this is certification of the operating company with no product scope, and the TAG Registry entry was not opened. The FY2025 10-K describes company fraud controls 'supported by proprietary and third-party fraud detection software, manual review, timely fraud investigations' (self-description). No audit of a Connect deal's composition or fees, and no MRC accreditation covering Connect, was found in the sources screened; Comscore's 'Certified' label is a partner's product brand, not an audit of Connect. [242, 231, 227] |
| Supply-path integrity | live | D | No evidence found on an observed sample: no Connect deal's ads.txt, sellers.json or SupplyChain records were observed, and PubMatic's sellers.json was not opened within the budget. The record holds only self-description: the console page says 'MBC ensures brand safety and quality control by adhering to industry standards such as sellers.json and ads.txt', and the 10-K describes proprietary and third-party fraud detection without naming an accredited measurer. PubMatic's own blog says curators 'don’t sit in the media payment chain', consistent with SupplyChain 1.0 having no curator node, while partner revenue shares are paid through PubMatic's billing. [222, 231, 225] |

Performance evidence: Campaign CTR five times 'the same campaign on the open market'; VCR +16.7% versus target; viewability +12.5% versus the agency's 80% target (relative or points not stated). (vendor assertion, sponsor PubMatic (release); figures attributed to 'Agence 79 internal data' (the buying agency)) [119]; Publishers opted into PubMatic's curation tools saw 'up to 5% incremental daily deal revenue uplift' (some '10%+'), 'Up to 14 percentage point increase in spending from a broader range of buyers' and 'A remarkable 25%+ increase in eCPMs compared to non-curated deals'. (vendor assertion, sponsor PubMatic) [230]; 'campaigns with data applied via Connect have seen CPM improve by 37% when compared to campaigns where the same data is applied on the buy-side'. (vendor assertion, sponsor PubMatic) [230, 225]; Panorama ID campaign with Xaxis, Xandr and PubMatic: 94% video completion rate and 83% viewability rate in Safari/Firefox against a '34%+ viewability benchmark'. (vendor assertion, sponsor Lotame (data source 'Lotame Internal Data, 4/11/22 – 5/9/22'), shown by PubMatic) [246]; Brain Nordic (publisher) raised its 'audience addressability rate to 97%' and targeting base to 70% from 35% with 'PMPs with PubMatic'. (vendor assertion, sponsor PubMatic (testimonial on the Connect for Publishers page)) [20]; Deal Management agent: 'Early tests indicated an 87% reduction in deal setup time'; more than 1,000 AI-powered deals since August 2025. (vendor assertion, sponsor PubMatic) [233]; PubMatic's CEO: 'My sense is that when we apply it on the sell side, the fees are lower than in the DSP.' (vendor assertion, sponsor PubMatic (CEO interview in AdExchanger)) [226]; Storefront testimonials (Publicis Groupe India, OMG Germany, Goodway Group) report increased viewability, win rates, efficient CPMs and exceeding an 'Audible Rates benchmark' from auction packages. (vendor assertion, sponsor PubMatic (customer quotes on its storefront)) [224]; Comscore describes Comscore-Certified Deal IDs on PubMatic as 'High-Quality, Attention-Optimized Curated Deals'. (vendor assertion, sponsor Comscore (with Adelaide and PubMatic quoted)) [227]

### Stagwell Curate (active; the product is live on the operator's claim ('available immediately', 16 July 2026))

*Stagwell Inc. (EDGAR registrant 'Stagwell Inc', CIK 0000876883, incorporated in Delaware). The product runs inside the Stagwell Media Platform, whose Global CEO is Matt Adams. No separate legal entity for Stagwell Curate or the Stagwell Media Platform is named in the sources opened (CUR-STAGWELLCURATE-S01, CUR-STAGWELLCURATE-S09).*

A holding company's own curated marketplace on FreeWheel and Magnite supply, live for network clients from July 2026, with its agent connection still in the future tense.

**SWOT (2026-11-03).** Stagwell Curate, which the agency holding company Stagwell Inc. launched on 16 July 2026 for its network's clients, names Stagwell as the rule owner but documents no fee, seller list, live buying endpoint, client or volume.

- **Strengths.** Stagwell names itself the rule owner: it applies 'its own intelligence, quality standards and performance signals upstream', and Adams says it controls which sites and apps are eligible. (vendor claim) [5, 71] Network clients access it through their existing agency teams 'with no DSP migration required', and Digiday reports that bids still execute in DSPs against the curated inventory. (vendor claim) [5, 71] Stagwell's Q2 2026 earnings presentation, an 8-K exhibit of 30 July 2026, lists 'Launched Stagwell Curate' among the quarter's highlights, so the launch is on the investor record. (record) [234] The release names its inputs: FreeWheel and Magnite as supply partners and 'supply‑quality models powered by The Trade Desk’s Sincera' for continuous evaluation with log-level data. (vendor claim) [5]
- **Weaknesses.** No curation, platform or data fee or take rate is disclosed; the releases offer only 'exclusive pricing' and a 'material reduction in ad tech fees', with no amount. (vendor claim) [5, 72] No seller list, domain list, selection threshold or quantified quality figure is public; Adams's claim that clients 'can see our buy data' has no documented interface. (vendor claim) [5, 71] The execution host is described three ways: FreeWheel 'will power' Curate (April), FreeWheel and Magnite 'will provide' publisher relationships (July), and it 'curates ad inventory in-house' (Q2 deck). (vendor claim) [72, 5, 234] No client, deal, spend, impression count or live DSP route is documented in the releases, Digiday's report or Stagwell's 2026 filings searched on EDGAR. (record) [5, 71, 234, 73, 235]
- **Opportunities.** The Media Machine, launched 19 June 2026, 'will connect directly' to Curate; a live link would give the network's own buying agents a route into its deals. (inference) [183, 5] FreeWheel's Curation Hub and Buyer Cloud 'will integrate' (April 2026), and FreeWheel says Curation Hub deals activate in any DSP; going live would give Curate a named endpoint. (inference) [72, 16] Adams argues that as DSP capabilities converge, 'differentiation shifts to supply'; if that holds, an agency-set supply filter above the client's DSP is where Stagwell can compete. (inference) [5]
- **Threats.** The group that buys for clients also selects supply and negotiates 'exclusive pricing'; with no fee recipient or principal status documented, clients may question the conflict. (inference) [5, 236] Supply runs through FreeWheel and Magnite and quality models come from The Trade Desk's Sincera, so a change in any of those relationships would change what Curate is. (inference) [5, 71, 72] OpenSincera, whose data Digiday says Stagwell uses, was announced in May 2025 as 'Free and open to all participants', so rivals can draw on the same data. (inference) [71, 237] Fifteen weeks after launch, Stagwell's Q3 2026 release and deck name 'Stagwell Machines' but not Curate, which may mean it is immaterial to the group or deprioritised. (inference) [73, 238]

- **What is sold.** Stagwell's agencies sell their clients access to a central deal marketplace and to custom-built deal packages. The release calls the product 'a new centralized deal marketplace and bespoke inventory curation platform' with two parts. The first is a marketplace of 'pre‑vetted inventory across CTV, online video, display and audio'. The second is 'Bespoke curation', which lets brands and agency teams 'build customized deal packages' 'aligned to specific audiences, contextual signals, KPIs or sustainability goals'. The release labels what the Media Machine's buying agents will reach through it as 'pre‑negotiated PMP deals and exclusive pricing' (CUR-STAGWELLCURATE-S01). The Q2 2026 earnings presentation calls it 'our new AI-powered marketplace and platform that curates ad inventory in-house' (CUR-STAGWELLCURATE-S05). It is offered through the agency relationship ('through their existing agency teams'), not as a separately listed self-serve tool. The release pairs it with Stagwell Control, 'the network’s brand safety and media quality offering', as the 'Stagwell Supply Product Suite'.
- **Who buys it.** Clients of Stagwell's network agencies, through their existing agency teams (CUR-STAGWELLCURATE-S01). Buyer eligibility, on the paper's definition, is one agency group's clients. No client, advertiser or agency inside the network is named in any source opened. Digiday quotes Adams that clients 'can see our buy data' (CUR-STAGWELLCURATE-S02). That speaks to client visibility; it is not a customer list.
- **How it earns.** Not disclosed. No source opened states a curation fee, platform fee, data fee or margin for Curate, or says who collects one. Company-wide, the FY2025 10-K describes revenue from 'agency arrangements in the form of fees for services performed, commissions, and from performance incentives or bonuses' (CUR-STAGWELLCURATE-S10). Whether Curate is charged as a separate line, folded into agency fees or commissions, or earned through negotiated pricing ('exclusive pricing') is not stated. The filings do not break Curate out (CUR-STAGWELLCURATE-S06).
- **Supply and data dependencies.** Supply: the July release names 'Partners including Freewheel and Magnite' that 'will provide access to deep, direct publisher relationships' (future tense). The April 2026 release says 'FreeWheel will power Stagwell Curate' and 'FreeWheel's Curation Hub and Buyer Cloud will integrate into Stagwell's media acquisition layer' (CUR-STAGWELLCURATE-S01, CUR-STAGWELLCURATE-S03). Digiday paraphrases Adams on concentrating spend on a 'small set of preferred SSPs' (CUR-STAGWELLCURATE-S02). Data: 'log‑level data' (source not named) and 'supply‑quality models powered by The Trade Desk’s Sincera' (CUR-STAGWELLCURATE-S01). Digiday adds 'AI agents built in Claude and data pulled from The Trade Desk's OpenSincera' for a 'domain health check' (CUR-STAGWELLCURATE-S02). OpenSincera is The Trade Desk's free ad-experience metadata application, live from 6 June 2025 (CUR-STAGWELLCURATE-S13). Buying: the client's existing DSP ('no DSP migration required'); no DSP is named. Partner names alone are not evidence of integrated functionality; the integration with FreeWheel is documented only in the future tense.
- **Differentiation.** The curator is the buyer's own agency group, not an exchange or a data company. Adams says it 'puts our intelligence above the DSP' and that Stagwell controls 'which sites and apps are eligible to send bid requests into our deals'. Clients keep their DSP ('no DSP migration required'), and Digiday reports that bids still execute in DSPs against the curated inventory. It is paired with Stagwell Control, the network's own brand-safety and media-quality offering, as the 'Stagwell Supply Product Suite'. Stagwell positions Curate against DSP-controlled marketplaces (Digiday cites OpenPath as an example of 'ubiquitous' marketplace solutions), yet it names The Trade Desk's Sincera as the source of its supply-quality models. Per stream G, it is the only buyer-built marketplace among those screened with a 2026 launch record (WPP Media's Premium Marketplace dates from 2022–2023 and dentsu Curate from 2020). That comparison comes from stream G and was not re-opened for this profile.
- **Risks.** Agency as curator (inference): the same group acts as the client's buying agent and as the party that selects supply, negotiates 'exclusive pricing' and may earn on the marketplace. No source documents principal or agent status for Curate, its fee recipient or how conflicts are managed. Partner dependence: supply runs through FreeWheel and Magnite, and the quality models come from The Trade Desk's Sincera. A change in any of these relationships would change what the product is, and the April-to-July change in partner wording shows the architecture was still moving at launch. Thin public record after launch: neither the Q3 2026 results release nor its presentation mentions Curate. That could reflect immateriality at group level or simply disclosure choice; the record does not say which. The agentic claims are forward-looking: 'will connect' is all the record holds for the Media Machine link, so any agentic positioning of Curate exceeds the documented state.
- **Best fit.** Clients already buying through Stagwell network agencies who want agency-negotiated PMP access across CTV, video, display and audio without changing DSP, and who accept agency-controlled supply selection on trust. Buyers who want an agency's own supply-quality filter applied upstream of their DSP and can audit it through their agency contract rather than from public documentation.
- **Poor fit.** Buyers outside the Stagwell network: the release offers the product only 'to clients across the Stagwell network'. Buyers who need a published seller or domain list, a fee schedule with base and collector, or independent quality measurement before paying for curated supply. Buyers looking for documented agentic or automated deal execution: the Media Machine connection is future tense.
- **Limits.** Every capability statement comes from Stagwell's releases, Stagwell's investor materials or Digiday's reporting of Stagwell executives. No product documentation, help centre, deal example or third-party observation of Curate was found. The record does not settle the execution host. The April 2026 release says FreeWheel's Curation Hub and Buyer Cloud 'will integrate'. The July release says FreeWheel and Magnite 'will provide' publisher relationships. The Q2 presentation says Curate 'curates ad inventory in-house'. No seller list, seller count, deal ID, threshold, fee, audience-data provider, DSP, client or volume is public, so H6 composition, H2b fees and dimension 8 scale cannot be coded beyond 'no public disclosure' or 'unknown'. The Media Machine's connection to Curate is stated in the future tense, and the Q3 2026 materials name neither product. No agent-executed selection on Curate is documented. US availability rests on a New York dateline and a US-listed, New York-headquartered operator. No page states it, and no US client or deal is named. The 10-Q for Q3 2026 and the FY2025 10-K financial statements were not read in full. The EDGAR full-text search bounds the absence of 'Curate' in Stagwell's 2026 filings, subject to the index's coverage.

| Dimension | Availability | Evidence | Observed |
|---|---|---|---|
| Selection transparency | live | D | Stagwell names its selection logic but publishes no threshold or update schedule. The release says Curate lets Stagwell 'apply its own intelligence, quality standards and performance signals upstream'. It describes the marketplace as 'pre‑vetted inventory' that is 'continuously evaluated using log‑level data and supply‑quality models powered by The Trade Desk’s Sincera', and says bespoke packages are 'aligned to specific audiences, contextual signals, KPIs or sustainability goals'. Digiday quotes Adams that Stagwell controls 'which sites and apps are eligible to send bid requests into our deals' and reports a 'domain health check' on OpenSincera data. In anchor terms this is logic named without thresholds (an eligibility list plus model scoring), from a release and from reporting of the operator's statements. [5, 71] |
| Composition disclosure | unconfirmed | D | No evidence found of a seller list, domain list, seller count or seller categories for the marketplace or for any bespoke package. Searched: the 16 July 2026 release, Digiday's report, the 27 April 2026 release copy, FreeWheel's Curation Hub page, Stagwell's Q2 and Q3 2026 earnings presentations, the Q3 results release and an EDGAR full-text search of 2026 filings. The only statements are general. The release promises 'enhanced transparency, measurement and accountability'. Adams told Digiday that clients 'can see our buy data; they can see how their inventory is being put together'. That is an assertion of client-side visibility with no documented interface and no example deal; capability, observed deal disclosure and delivery verification are all undocumented. [5, 71, 72, 234, 73, 238, 235, 16] |
| Data provenance | live | D | Provenance is documented only for the supply-quality input, and only at the level of source type and provider. The release names 'log‑level data' without its source and 'supply‑quality models powered by The Trade Desk’s Sincera'. Digiday says the domain health check uses 'data pulled from The Trade Desk's OpenSincera'. OpenSincera's release describes non-personal ad-experience metadata such as 'ads-to-content ratio, page weight, average ads-in-view, and ad refresh rate'. For the audiences, contextual signals and 'performance signals' used in bespoke packages, no provider, collection method, intended use, permission or legal basis is stated. That input is documented only at the 'proprietary data' level. [5, 71, 237] |
| Quality evidence | live | D | No quantified quality claim appears in the sources opened. There is no MFA share, viewability, IVT, suitability, attention or carbon figure, no measurer and no comparator. The release uses the adjective 'pre‑vetted' and promises 'enhanced transparency, measurement and accountability'. The Q2 2026 earnings presentation calls Curate an 'AI-powered marketplace', and the Q3 2026 release and presentation do not mention it. In anchor terms these are claims without qualifying measurement. [5, 71, 234, 73, 238] |
| Auction mechanics | unconfirmed | E |   |
| Channel execution | live | D | The July release advertises four environments on the paper's taxonomy: CTV, web video ('online video'), web display ('display') and audio. The April 2026 release also names 'streaming and linear TV', and linear TV is absent from the July list. Execution is claimed for all four only in the release's present tense ('The marketplace provides an always‑on, cross‑channel pool of pre‑vetted inventory across CTV, online video, display and audio'). No deal, case, partner page or filing documents execution in any one environment, and FreeWheel's Curation Hub page covers CTV only and does not name Stagwell. The sources do not say whether 'online video' and 'display' include mobile in-app. [5, 72, 16] |
| Activation routes | unconfirmed | D | No named buying endpoint is documented as live for Curate deals. The release says clients buy 'with no DSP migration required', and Digiday reports that DSPs 'will still be used to execute bids against the curated inventory' without naming one; under the anchor that is an unsupported any-DSP statement. The only named endpoint is FreeWheel's Buyer Cloud. The April 2026 joint release says it 'will integrate into Stagwell's media acquisition layer' (future tense), and FreeWheel's own page says Curation Hub deals activate 'across any demand-side platform (DSP) including Buyer Cloud' without mentioning Stagwell. The Media Machine's integrations (Google's GMP products, Meta, Microsoft & LinkedIn, TikTok, The Trade Desk) belong to a separate product, and its connection to Curate is stated in the future tense. [5, 71, 72, 183, 16] |
| Scale | live | D | No evidence found of curated-deal spend or impressions in a stated 12-month window, and no client, publisher or deal count. Searched: the July and April releases, Digiday, the Q2 2026 earnings presentation (8-K exhibit, 30 July 2026), the Q3 2026 results release and presentation (8-K, 29 October 2026) and an EDGAR full-text search of Stagwell's 2026 filings for 'Curate'. The only figures are company-wide: Q3 2026 revenue of $816 million, net revenue of $664 million and 'specialists in 45+ countries'. The anchor treats these as context only. [5, 71, 72, 234, 73, 238, 235] |
| US availability and coverage | live | D | US availability is implied but never stated. The release is datelined New York City and says Curate is available 'through their existing agency teams' to clients across the Stagwell network. The operator is a Delaware company listed on Nasdaq and headquartered at One World Trade Center, New York. The release says the network spans '45+ countries'; the FY2025 10-K says 'over 34 countries'. No source names a US client, a US deal or US delivery in a period, and no page says whether the product is offered outside the US. In anchor terms this is at most US availability by operator nexus, from a release, with no US production use. [5, 247, 236] |
| Pricing transparency | live | D | No charge is disclosed in any of the four series: no curation or platform fee, data fee, SSP take rate or DSP take rate, and no payer, collector, base or period; no Deal Sync API curfeetype was observed. The statements found are general. The release offers 'pre‑negotiated PMP deals and exclusive pricing' to the Media Machine's agents, whose connection is future tense. The April 2026 release promises 'More working media for clients enabled by material reduction in ad tech fees', and Digiday reports Adams's aim to 'reduce tech fees'. The FY2025 10-K describes revenue company-wide as 'agency arrangements in the form of fees for services performed, commissions, and from performance incentives or bonuses', which does not say how Curate is charged. In anchor terms: a general pricing assertion without a schedule. [5, 71, 72, 236] |
| Independent assurance | unconfirmed | D | No evidence found of a certification, accreditation or audit that names Stagwell Curate, its deals or Stagwell Inc. as the assured entity. Searched: both releases, Digiday, FreeWheel's Curation Hub page, the Q2 and Q3 2026 earnings materials and the FY2025 10-K sections read; TAG's and MRC's registries were not searched for Stagwell within budget. Sincera's models are a data input, not assurance. 'Stagwell Control', 'the network’s brand safety and media quality offering', is Stagwell's own product, so at most it would count as self-attestation, and no assurance scope is documented for it. [5, 71, 72, 234, 73, 236, 16] |
| Supply-path integrity | unconfirmed | D | No evidence found. No sample of Curate supply could be checked against ads.txt, sellers.json or SupplyChain, because no deal ID, seller list or domain list is public. The release says inventory is 'continuously evaluated using log‑level data and supply‑quality models powered by The Trade Desk’s Sincera'. Digiday reports 'scores for ad quality and supply-chain quality' and a plan to concentrate spend on a 'small set of preferred SSPs'. Neither names an IVT filter, an accredited measurer or a check for duplicated requests. These are self-descriptions of evaluation, not observed authorisation. [5, 71] |

Performance evidence: The marketplace is an 'always‑on, cross‑channel pool of pre‑vetted inventory', 'continuously evaluated using log‑level data and supply‑quality models powered by The Trade Desk’s Sincera', 'delivering enhanced transparency, measurement and accountability'. (vendor assertion, sponsor Stagwell Inc. (self-published release)) [5]; 'More working media for clients enabled by material reduction in ad tech fees' (Stagwell–FreeWheel, April 2026). (vendor assertion, sponsor Stagwell Inc. and FreeWheel (joint release, read in a reproduced copy)) [72]; Adams told Digiday that Curate should 'reduce tech fees', that clients 'can see our buy data', and that the process is the agency's own rather than a third-party 'black box'. (vendor assertion, sponsor Stagwell executive quoted in trade press (Digiday)) [71]; 'Launched Stagwell Curate, our new AI-powered marketplace and platform that curates ad inventory in-house.' (direct record, sponsor Stagwell Inc. (earnings presentation furnished as an 8-K exhibit)) [234]

### How to read a row

Read the band first, the counted cells second and the number last. A band is a distance from the cohort's field DQ, so a gap of a point or two inside a band means little. Index Marketplaces and Microsoft Curate sit 0.6 points apart in different bands, and the Depth-led and Standing-led weightings swap them.

Check the marks. The ± mark means the band changes when the row's n/e cells are filled at 1 or 4, and every banded row carries it. "Band moves" means a reweighting changes the band, which applies to Index Marketplaces, Microsoft Curate and TL Direct.

Read n/e as unknown. Shrinkage pulls a thin record hard toward the field mean, so Google Ad Manager Curation's 53, resting on four cells, says little either way.

Read the group means before the total. Microsoft Curate's Depth mean is the highest among banded rows, built on auction mechanics documented with their defaults [110, 111]. Its Reach rests on one route cell: deal creation for buyers on Google's Display & Video 360 (DV360), documented in a release note of 21 January 2026 [110, 111, 121].

### What the DQ cannot tell a buyer

The DQ cannot say whether a curated deal delivers its quality promise: no independent measurement found tests one [310, 101, 311]. It cannot say what a deal costs all-in. Where fees are documented, amounts appear in worked examples, with Index's own share negotiated per media owner [109], or not at all, as in Microsoft's fee stack [115, 216, 111] ([chapter 6](#fees)). It cannot say what a named deal contains ([chapter 3](#contents)). It cannot rank products by market weight, because on the readable vendor pages the sweep opened no product discloses a periodised curated-deal volume [15, 256, 223, 266]. It cannot say whether a deal's supply is authorised, because no row has a counted supply-path cell. Nor can it compare cohorts, or a DQ with a CQ or an NQ. The n/e cells in a row list what a buyer would still need to ask the vendor before paying for the deal.



## 12. What a buyer should ask before paying for curation


*What should a buyer get in writing before paying for a curated deal? Nearly everything: on the pages screened by the cutoff, no exchange product documents a named deal's full seller and domain list or any post-purchase protection, and no curation rate card is public [110, 248, 25, 11, 258] [115, 24, 25, 109] [212, 231, 281, 115, 24]. Contents, charges, auction settings, measurement, change rights and any agent's decision therefore have to come by contract.*

> **In short.**
> Every post-purchase power documented for the fifteen exchange products belongs to the curator, the publisher or the exchange [115, 24, 25, 109] [115, 110, 113]. Microsoft Curate lets a curator edit a live deal's margin at any time, and Index reserves the right to update targeting on deals that use LiveRamp's segments, to keep them to the United States [115, 113, 110] [109, 14, 240, 206]. No exchange product documents a full seller and domain list for a named deal on the pages opened, and the supply-chain files cannot show a curator that is not a paid seller in the chain [110, 248, 25, 11, 258] [143, 142, 141]. No public source gives the curation fee, data fee, exchange take and DSP (demand-side platform) fee, with payer, collector, base and period, for one named deal [115, 216, 290, 64]. IAB Tech Lab's Deal Sync API 1.0 lets a deal declare its curator and the type of curation fee, with no amount [96] [96]. For fourteen of fifteen exchange products the pages opened do not settle whether a curated deal gets priority or a floor advantage by default [111, 250, 24, 266] [14]. The two agent-run transactions that meet the production test say neither what the agent selected nor by which rule [172] [93]. On Q5's coding rule no term is documented in public for the thirteen scored products screened, the other eight code access unknown, and contract terms are unknown for all 21 [115, 24, 25, 109] [246, 20, 222] [59, 196, 190] [13, 125, 196, 208].

### What the record covers

The asks rest on the documentation of fifteen exchange products opened on 3 November 2026, the standards curated deals run on, the fee record and the agent record [115, 24, 25, 109] [96] [212, 231, 281, 115, 24] [93, 120, 172, 168, 171]. Each ask below names the evidence and chapter behind it, and the decision tool near the end gives the test that fits each buyer question.

The author has worked at Samba, a TV measurement and advertising data company that owns Semasio, since May 2025 [36]. He was executive vice-president at Verve Group from September 2022 to August 2024, and earlier worked at Lotame, whose Curated Marketplaces is coded for Q5, and at Hearts & Science. He co-leads the Signals & Measurement working group of AgenticAdvertising.org, which maintains AdCP (Ad Context Protocol), and maintains an open-source AdCP signals adaptor. No Fluff Advisory operates two agents listed in the AdCP registry. None of these earns any evidence weight ([chapter 13](#method)).

### Ask for the seller and domain list, and check it against sellers.json

No exchange product documents a full list of the sellers and domains in a named curated deal on the pages opened [110, 248, 25, 11, 258]. Equativ's help centre comes nearest, with a publisher-name table for Maestro deals of 50 publishers or fewer and no domains [3]. Microsoft Curate shows the buyer the deal name, does not say whether the buyer can see the seller list, and does not support ads.txt filtering on curated deal line items [110, 111]. Google's agency seat shows the deal's name, pricing and a "created by" curator label, and no page opened says the buyer sees the publishers or domains [59, 24, 248]. [Chapter 3](#contents) codes composition for every product scored.

Ask for the seller IDs and domains behind the named deal ID, for the flight dates, and check them against the supply-chain files. The sellers.json file labels each seller account as publisher, intermediary or both [142, 263]. SupplyChain 1.0, which carries the chain in the bid request, lists only entities in the direct flow of payment [143]. None of the files defines a curator role, so a curator appears only where it is a paid seller, and a data or selection provider that is not a paid seller cannot appear at all [143, 142, 141]. The Media Rating Council (MRC) auction transparency standard of January 2026 requires DSPs to validate ads.txt seller IDs against sellers.json and report undeclared chains [151].

Where the exchange supports the Deal Sync API, ask for three more fields. The pubcount field declares one publisher or several, auxdata flags paid data from a party other than the publisher, and dinventory says whether the site list may change during the flight [96]. The specification warns that deal criteria sent through it are not guaranteed to appear in bid requests carrying the deal ID, and tells buyers to apply matching targeting themselves [96]. A checked list settles who sold the impressions; it does not show which data was applied to each one [143, 142, 141].

### Ask for every charge in writing

No exchange or curator in the sources screened publishes a rate card for curation fees or for its own take on curated deals [212, 231, 281, 115, 24]. On the pages the exchange stream opened, four products document the curation fee's type, base, collector and payment route without amounts: Microsoft Curate, Google Ad Manager, Index Marketplaces and, through its knowledge base, Criteo's Commerce Grid [115, 24, 109, 117, 266]. On Google and Index the curator decides whether the buyer sees the fee; Microsoft and Commerce Grid do not say [115, 24, 109, 117, 266]. Google's two help pages disagree on whether the agency sees it [24, 59, 248].

Where the mechanics are documented, the fee comes out of the bid before it competes, on Microsoft by default unless seller and curator agree otherwise [110, 111] [24, 59, 248] [109] [266]. In Index's worked example a $10 bid pays the media owner $8.08 after a 5% partner fee and a 15% Index share, a share Index negotiates with each media owner [109]. AdExchanger, reporting Jounce Media's client-only research, describes curators and SSPs (supply-side platforms) building the fee into the floor passed to DSPs, with the SSP paying the curator [64].

The aggregate series cannot price a named deal. Jounce's April 2025 report, known through Playwire and AdExchanger, covered 1.5 billion bid requests from 44 exchanges worldwide, analysed in March 2025 [9, 64, 281]. Half its curated deal IDs carried no fee beyond the exchange's standard take [9, 64, 281]. A further 34% carried a margin-based fee (median 14%), 10% a fixed fee per thousand impressions (CPM, median $0.48) and 7% a static price [9, 64, 281]. Playwire also labels these as shares of bid requests [9, 64, 281]. The Association of National Advertisers (ANA) benchmark folds curation services into its SSP cost line and imputes SSP fee rates where log data is missing [287, 99, 26]. In a survey release from TripleLift, which sells TL Direct, 60% of 223 US and UK advertisers said they do not know what they pay for curation [291]. The release does not say who fielded the survey or when [291].

Ask for each charge in the path, the curation fee, the data fee, the exchange's take and the DSP's fee, with amount or rate, base, payer, collector and gross or net treatment. Ask whether each is deducted from the bid, built into the floor or added on top, and whether the curator's fee will show in the buyer's reports. The MRC auction standard requires tech fees and deal discounts to be made known and reported [151]. The implementation notes of OpenRTB, IAB Tech Lab's bid-request protocol, warn that a deal may fail to function correctly if deal bids are silently discounted for margins or fees [95]. Reconcile charges only within one transaction ([chapter 6](#fees)) [115, 216, 290, 64].

### Ask for the curator fields of the Deal Sync API

IAB Tech Lab finalised the Deal Sync API 1.0, which its standards page calls the Deals API, on 6 February 2026 [261, 249, 96]. Its Curation object names the curator by domain and carries the deal ID in the curator's namespace [96]. Its curfeetype field declares the fee type: 0 undisclosed, 1 percentage of spend, 2 flat fee, 3 CPM, 4 no fee paid for curation services [96]. The field carries no amount, base, payer or collector, and the guidance text calls it curationfee [96] [96]. A value of 4 speaks only for the curation service: the specification's own example pairs it with a fee-bearing data layer flagged in auxdata [96] [96]. No research stream that read the specification found an exchange, curator or DSP documented as populating these fields [261, 249, 96] [96] [96].

Ask whether the exchange supports the API for this deal. If it does, get the curator, cdealid, curfeetype and auxdata values in writing, then ask for the amount behind each type. Version 1.0 supports no deal revisions while allowing attributes to change during the flight, so later changes have to reach the buyer by contract [96].

### Ask for the auction settings of the named deal

For fourteen of fifteen exchange products, the pages opened do not settle whether a curated deal gets priority, a floor advantage or exclusive supply by default ([chapter 4](#access)) [111, 250, 24, 266] [14]. Microsoft Curate packages every curated deal as Open Auction and never overrides seller floors [110, 111]. Index has Marketplace bids compete net of fees unless otherwise specified, and gives a fixed price auction priority over all other auction bid types [198, 109] [14]. Google lets curators bundle Preferred Deals, which it defines as fixed-price and pre-auction [24, 250]. It applies a curator's floor only where it exceeds the publisher's [59, 24]. Microsoft samples curated deals that buyers stop bidding on, which can shrink the bid requests a buyer sees [113].

Ask, for the named deal, for the transaction type as the exchange labels it and the auction type: first price, second price or fixed. Ask whether deal bids rank above open-auction bids, how the deal's floor relates to the publisher's, which buyer seats may bid, and what sampling rule applies, with its trigger. OpenRTB carries an auction-type override, a floor and buyer-seat allow-lists for each deal, and its field defaults establish nothing about how an exchange configures one [139] [139]. The MRC auction standard requires auctioneers to disclose auction type, reserve prices and the priority rules that decide winners beyond price [151] [151] [151]. A written answer is still the exchange's statement: every entry in the access register is a platform's own, and none observes an auction [110, 25, 113].

### Ask for measurement against the same inventory before curation

The closest independent comparison in the record sets deals against different inventory [26]. The ANA's 2023 study covered 21 advertisers, predominantly US, and $123 million of spend from September 2022 to January 2023 [26, 63]. Private-marketplace (PMP) deals spanning more than 500 domains cost $5.52 CPM against $2.75 in the open marketplace, with 19.8% of impressions on made-for-advertising (MFA) sites against 26.8% [26]. Those are deal bands across advertisers, so they describe the market without testing any named deal's promise ([chapter 7](#quality)) [26].

The figures for PubMatic's December 2025 campaign with Butler/Till, among them a 98% video completion rate and 0% MFA inventory, come from the transacting parties, and none is independently verified [93]. OpenX's claim of 67% lower CPMs for Hyundai is set against an undefined "next-best alternative" [120].

MRC's guidelines define viewability and invalid traffic (IVT), the thresholds a promise can name [146] [148, 149]. Its auction standard encourages third-party, MRC-accredited IVT measurement and lists MFA among considerations it does not yet cover [151]. The four seals of the Trustworthy Accountability Group (TAG) each certify a company, and none certifies a deal [308, 152, 155, 154, 153].

Ask for the written promise: metric, threshold or uplift, definition and period. Ask for measurement by a party with no stake in the deal, with its funding stated, against comparable eligible inventory before curation. That means the same publishers and placements, bought outside the deal by the same buyer over the same dates, ideally split at random. Fix the margin first. This paper registered the smaller of 10 percentage points and the claimed uplift for a rate, 2 points for invalid traffic, and 10% relative for attention and carbon. An inconclusive result is not equivalence, and a result on one metric says nothing about another.

### Ask who can change the deal, and with what notice

Every power to change a live deal that the pages opened describe belongs to the curator, the publisher or the exchange, and none comes with a documented notice to the buyer [115, 24, 25, 109] [115, 110, 113] [109, 14, 240, 206]. On Microsoft Curate the curator can edit a deal's margin at any time and the buyer is fixed once the deal is saved [115, 113, 110]. Sellers can block curators, and Microsoft applies its sampling without curator action [115, 110, 113] [113]. Microsoft also switches on its cross-device enhancement on any curated deal that targets a segment, with opt-out only through an account representative [110].

On Google Ad Manager a curation partner can change the fee amount of a live data segment but not its fee type [24, 25]. Index lets the partner choose how much of its fee to disclose in reporting [109, 14, 240, 206]. It also reserves the right to update targeting on deals that use LiveRamp Data Marketplace segments, to keep them to the United States [109, 14, 240, 206]. An Index package built with a blank publisher list takes in new publishers automatically [14]. Commerce Grid lets the curator adjust the curation fee after the deal has synced to the DSP [115, 24, 25, 109].

The buyer is offered reporting. PubMatic describes deal-level usage reporting for buyers and log access "for verification" in its buyer console, and no PubMatic source opened covers notice, audit rights, export or remedies [246, 20, 222]. Index's product page claims "audit logs" [258], and a claimed log gives the buyer no right to audit. The MRC auction standard requires material changes to auction systems to be announced in advance under a formal change-notification policy [151]. Compliance is voluntary, and whether any auctioneer is accredited against it was not looked for in these streams [151].

Ask who can change the margin, the fee, the seller list, the targeting, the data and the sampling during the flight. Then write into the contract a notice period, a change log, an audit right and export of log-level delivery data. Add a remedy for misdescribed inventory or charges, in money or media, with a date.

### Where an agent is involved, ask what it decided and who approved it

Vendors report agent-run campaigns, and only one account read says what the agents selected, without the rule they used ([chapter 10](#agents)) [313] [172] [93]. AdExchanger reported on 14 October 2026 that the agency H/L's buyer agent and Magnite's seller agent had set up and activated a campaign on LG's home-screen inventory [172]. LG and H/L approved the agent-led plan before it went live [172]. Marketing Dive, reporting press materials, described PubMatic's December 2025 campaign with Butler/Till as planned and implemented by agents on PubMatic's platform, from a brief submitted through the agency's enterprise Claude interface [93].

AdCP's README lists a Curation surface as "Coming soon", so curation has no schema in AdCP at the cutoff [164]. Its Media Buy walkthrough has the seller curate, price or negotiate a plan from the brief, and its introduction gives orchestrators the job of making sure humans approve what matters [97, 176]. Its documentation leaves payment, legal agreements, disputes, credits and makegoods to account or platform integrations unless the relevant AdCP contract is present [97]. The AdCP registry lists test and sandbox endpoints beside production-named ones, so a listing establishes no deployment [165].

Ask which decision the agent made (inventory selection, price, plan or bid), on which inputs, under whose authority and with which permitted actions. Ask where it ran, what feedback it used, who approved the plan and when, and get the log of the decisions it executed on the buy.

### What only a contract can tell you

The public documentation leaves six things to the contract:

- the sellers and domains in the named deal, and whether they can change during the flight [110, 248, 25, 11, 258] [96];
- the amount of every charge, its base, who pays it and who collects it [115, 216, 290, 64];
- whether the curator's fee appears in the buyer's reports [115, 24, 109, 117, 266];
- the deal's priority, floor and sampling rules where the exchange publishes no default, and the threshold behind a published sampling rule [111, 250, 24, 266] [113];
- notice of changes, an audit right, export of log-level data and a remedy for misdescribed inventory or charges [115, 24, 25, 109];
- for an agent in the path, the decision it made and who approved it [172] [93].

The decision tool sets out each buyer question with the test that fits it, what the inference needs, where it fails and what it cannot tell you.

| Buyer question | Test that fits | What the inference needs | Where it fails | What it cannot tell you |
|---|---|---|---|---|
| What is in this deal? | Ask for the seller IDs and domains behind the named deal ID for the flight dates. Check each seller ID against the exchange's sellers.json and each domain's ads.txt, and read the SupplyChain object on delivered impressions. Where the exchange supports IAB Tech Lab's Deal Sync API 1.0, ask for the pubcount, auxdata and dinventory values as well. | The list belongs to this deal ID and period, not to a category or a set of example publishers; the supply-chain files are current; the exchange passes a complete SupplyChain object. | When the list can change during the flight (an Index package with a blank publisher list takes in new publishers automatically, and dinventory can declare changes allowed), when the exchange does not support ads.txt filtering on curated deals (Microsoft Curate), or when the curator or data provider is not a paid seller in the chain and so cannot appear in the files. | Show which data or selection rule was applied to each impression, or that the criteria sent through the Deal Sync API reached the bid requests; the specification says they are not guaranteed to. [110, 248, 142, 143, 96, 14] |
| What am I paying, and to whom? | Get every charge in the path in writing: curation fee, data fee, the exchange's take and the DSP's fee, each with amount or rate, base, payer, collector and gross or net treatment, and whether it is deducted from the bid, built into the floor or added on top. Reconcile them against log-level data for the same transactions. | All the charges sit inside one transaction perimeter and each party discloses its own; a fee deducted before the auction lowers the bid that competes. | When the curator hides its fee from buyer reporting (an option on Google Ad Manager and Index), when the exchange negotiates its share with each media owner, or when the fee is built into the floor passed to the DSP. Every zero-fee claim found covers one party's charge only. | Say whether the deal costs more than the same impressions bought in the open auction; that needs a matched comparison, and none is public in the sources screened. [115, 24, 109, 64, 151] |
| Does the deal declare its curator and fee type? | Where the exchange supports the Deal Sync API 1.0, read the Curation object: curator (a domain), cdealid and curfeetype (0 undisclosed, 1 percentage of spend, 2 flat fee, 3 CPM, 4 no fee paid for curation services), and the auxdata flag for paid data from a party other than the publisher. | The exchange populates the fields for this deal and keeps them current. | When curfeetype is 0, when no exchange or DSP in the path implements the API (none is documented as doing so in the sources screened), or when the deal changes after it is synced, since version 1.0 supports no deal revisions. | Give an amount, base, payer or collector. A curfeetype of 4 speaks for the curation service only, and the specification's own example sets it beside a fee-bearing data layer. [96, 249] |
| Does the deal get priority, a floor advantage or sampling? | Ask the exchange, for the named deal, for the transaction type, the auction type (first price, second price or fixed), priority against open-auction bids, the floor rule relative to the publisher's floor, the buyer seats allowed to bid, and any sampling or throttling rule with its trigger. Compare the answers with the exchange's documentation and with the disclosures MRC's auction transparency standard asks for. | The exchange's documented default applies to this deal and the seller has not configured it otherwise. | For twelve of fifteen exchange products the pages opened do not settle priority or floor treatment; Microsoft's sampling rule states no threshold; a Microsoft deal can draw on seller-negotiated supply on terms the pages do not describe; a Google package built on Preferred Deals is pre-auction by Google's definition of the type. | Show what happened in the deal's auctions. Every coded default is the platform's own statement, and only auction logs or an independent observation would show behaviour. [111, 110, 113, 24, 250, 151, 198] |
| Does the deal deliver the quality it was sold on? | Get the written promise (metric, threshold or uplift, definition, period). Then have a party with no stake in the deal measure it against comparable eligible inventory before curation: the same publishers and placements, bought outside the deal by the same buyer over the same dates, ideally split at random. Fix the margin before the test; this paper registered the smaller of 10 percentage points and the claimed uplift for a rate, 2 points for invalid traffic, and 10% relative for attention and carbon. | The comparator differs from the deal only by curation; selection, auction and measurement differences are documented; the measurer's funding is stated and, where accreditation exists, it covers the metric. | When the figures come from the seller, the curator or the agency, when the comparator is other inventory (the ANA's deal bands compare different deals across advertisers), or when the result is inconclusive, which is not equivalence. | Carry a result from one metric to another, or from one deal or period to the next. [26, 146, 148, 151, 93] |
| Am I paying more than the open auction for the same impressions? | Run or request a matched comparison: the same eligible opportunities, the same period and the same buyer, with all-in net cost per thousand impressions including every charge on both routes, and rebates, displaced charges and inventory mix reported separately. | Eligibility is the same on both routes, and every charge on each route is recorded. | When deal CPMs are set against open-market CPMs for different inventory, as the ANA's private-marketplace and open-marketplace figures are, or when a zero curator fee is read as an equal all-in cost. | Say whether a higher price buys anything; that needs the quality test as well. [26, 66, 9, 64] |
| Who can change the deal after I buy, and will I be told? | Ask who can change the margin, the fee, the seller and domain list, the targeting, the data applied and the sampling during the flight. Then write into the contract a notice period before any change, a change log, an audit right, export of log-level delivery data, and a remedy for misdescribed inventory or charges, in money or media, with a date. | The contract binds both the curator and the exchange, and the exchange records changes to live deals. | The public documentation screened gives these powers to the curator, the publisher or the exchange and documents no notice to the buyer: Microsoft margins editable at any time, Google live segment fees, Index's right to update targeting on deals using LiveRamp segments to keep them to the United States, Commerce Grid fees after sync. | Show whether a change was made unless change logs and delivery logs are handed over. MRC's change-notification requirement covers auction systems, and compliance with it is voluntary. [115, 24, 25, 240, 151, 96] |
| Is an agent making the decision? | Ask which decision the agent made (inventory selection, price, plan or bid), on which inputs, under whose authority and with which permitted actions, where it ran and what feedback it used; who approved the plan and when; and for the log of decisions it executed on the buy. | The parties record the agent's decisions and the human approvals, and will release them. | When the record is a release or a registry listing. The documented production cases come from the transacting parties and say neither what the agent selected nor by which rule, and AdCP has no curation schema at the cutoff. | Tell you whether the agent's choice beat a human's or the open auction's; no result in the sources screened was measured by a party without a stake. [172, 93, 120, 164, 176] |

### What would settle the open questions

Five public records would settle what this paper leaves open. An independent measurement of a named deal's quality against the open auction, on the same inventory and period, would code H3 for that promise ([chapter 7](#quality)). A published fee schedule or filing line for curation on a major exchange would move that exchange's codes under Q3 and H2b ([chapter 6](#fees)). A composition disclosure checkable against sellers.json would move that product's composition code under H6 ([chapter 3](#contents)). A documented default priority or floor rule on a major exchange would settle that exchange's H4 code, one of fourteen now unresolved ([chapter 4](#access)). A sell-side agent decision in production, recorded by a party without a stake, would remove H7's qualification ([chapter 10](#agents)).

### Coding Q5

Q5's registered rule codes a term documented when a public policy, terms page or help article states it, and access unknown where only a contract would state it; unknown contract terms are a limit of public evidence, not a finding. The population is the 21 products scored, each coded on four terms: change notification, audit rights, data exportability, and remedies for misdescribed inventory or charges.

| Products | Change notification | Audit rights | Data exportability | Remedies | Evidence |
|---|---|---|---|---|---|
| Microsoft Curate, Google Ad Manager Curation, Commerce Grid, Maestro by Equativ, TL Direct, FreeWheel Curation Hub, ClearLine, OpenXSelect, YMax | Not documented in public | Not documented in public | Not documented in public | Not documented in public | [115, 113, 110] [115, 110, 113] [24, 25] [115, 24, 25, 109] |
| SmartPMP, Lotame Curated Marketplaces (interest mark) | Not documented in public | Not documented in public | Not documented in public | Not documented in public | [59, 196, 190] [13, 125, 196, 208] |
| Index Marketplaces | Not documented in public | Not documented in public; "audit logs" claimed | Not documented in public | Not documented in public | [109, 14, 240, 206] [258] |
| PubMatic Connect | Not documented in public | Not documented in public | Not documented in public; log access asserted | Not documented in public | [246, 20, 222] |
| Seedtag NeuroX Curation, Adelaide High-Attention PMPs, Comscore-Certified Deal IDs, IAS Curation, Multilocal curation, dentsu Curate, Stagwell Curate, WPP Media Premium Marketplace | Access unknown | Access unknown | Access unknown | Access unknown | Outside the pages screened for these terms [115, 24, 25, 109] |

On the products' own pages, none of the thirteen products screened documents any of the four protections a buyer would want after paying, and the other eight were not screened for them [115, 24, 25, 109] [109, 14, 240, 206] [246, 20, 222] [59, 196, 190] [13, 125, 196, 208]. Documented controls run the other way: curators, publishers and exchanges can change a live deal's margins, fees, targeting, publishers and traffic [115, 24, 25, 109] [109, 14, 240, 206] [113]. A buyer who wants notice, audit, export or a remedy must write it into the contract.



## 13. Method, limits and corrections


*How the paper was built, what the checks caught and what could not be reached. It is published so a reader can check the paper rather than trust it.*

> **In short.**
> AI research agents worked from one brief and one set of rules. They produced 10 research workstreams, 1 vendor-universe sweep and 7 company profiles. A separate AI verifier, told to refute, checked each file. The research opened and logged 389 sources and put 454 claims in the ledger, 446 of them checked by a verifier. Verifiers made 699 checks: 491 supported as stated, 185 supported with a correction, 11 contradicted and 12 unresolved. Every scored cell was then re-checked at its source by another verifier. The brief had one outside review, by another AI model; the evidence and the draft had none before publication, and the published edition had one, which led to this corrected edition. One scoring pass set every score, and no independent second scorer exists. The corrections ledger holds 392 rows.

### What was fixed before the research

The brief was drafted on 2 November 2026 from the site's own published record. Seven checkers then opened the primary source behind 66 of its dated facts, figures and ownership claims and swept the product names of 46 companies. The brief's v1.1 recorded every verdict, and three facts drafted from the site's record were wrong: the attribution of 8 October 2024 attention tiers to Amazon DSP (they were The Trade Desk's), and two studies the brief expected to exist, a WFA programmatic study of 2024 and a 4A's study of early 2024.

Three registers were written on 2 and 3 November 2026, before any cell was scored. The hypothesis register records 13 entries: eight hypotheses (H1, H2a, H2b, H3, H5, H6, H7, and H4 recast as a descriptive access register) and five coded questions (Q1 to Q5). Each states its population, comparator and observable outcome, and the rules for supported, contradicted and unresolved. Absence of accessible evidence yields unresolved, never contradicted.

The product register was frozen in provisional form before the vendor-universe sweep and in final form after it, as the outside review asked. It holds 52 records, of which 22 are listed and 2 historical. The sweep moved YMax from the exchange cohort to curated deal products and Seedtag's NeuroX Curation the other way, and moved Adloox's attention segments and Samba's PMPs to listed. It also corrected entities and US availability as the vendors' own pages state them, and added two late candidates, Onetag's curation platform and Permutive Curation, as listed products for a later edition. Every change is a dated amendment in the scoring rules and a row in the corrections ledger.

The scoring rules were fixed on 3 November 2026, before any cell was scored: twelve dimensions with anchors for every value from 0 to 5, three cohorts defined by the purchased service with a frozen decision tree, an applicability mask per cohort, the evidence caps, the disclosure modes and the quotient. The scale anchors (dimension 8) were proposed before the sweep and kept after it, because the sweep found no curated-deal volume disclosed with a unit and a twelve-month window to calibrate against.

### Review history and the evidence freeze

The author ran one outside review of the brief with another AI model on 2 November 2026. It returned 46 findings: 27 on design and 19 on facts. Three checkers and the browser pane re-opened the primary source behind every factual finding before any change: 10 held as stated, 9 held with a correction, and none was wrong outright. The largest correction concerned PubMatic: the sentence the review quoted is not on the page it cited. The design findings reshaped the brief. Cohorts now follow the purchased service, a curated deal is coded on four separate fields, H2 split into cost and disclosure, H4 became an access register, and H7 defines what counts as an agent. Every disposition is published with the research record for the series.

The author then waived the two further outside reviews the pipeline planned, of the evidence and of the full draft, and directed that the paper be written at once. The evidence freeze therefore moved from the planned 4 December 2026 to 3 November 2026, the day the research ran. Status-sensitive facts (ownership, accreditation, dockets, protocol versions) were read on 2 and 3 November 2026 and are dated in the text. Anything found after the cutoff will be logged as a dated correction, not folded in silently.

After publication on 3 November 2026, the author had another AI model review the live edition. Its findings were checked at their sources on 4 November 2026 in a final live sweep, and the ones that held produced this corrected edition. Index's package page, dated 26 October 2026, defines a fixed price auction as having priority over all other auction bid types. Index therefore no longer codes as holding the essay's claim, and a Google package built on Preferred Deals no longer codes as failing it, because in both the curator chooses the priority ([chapter 4](#access)). PubMatic's own releases, which the agent stream had not screened, add a June 2026 account in which agents selected inventory, and the two earlier deployments no longer count as executed selections ([chapter 10](#agents)). H2b is now tested on charge types and bases alone, and the Deal Quotient chapter states what the bands rest on ([chapter 6](#fees), [chapter 11](#dq)). The matrix and score exports now carry the interest mark and a new relationship mark. One finding did not hold as cited: its link for the Havas campaign led to PubMatic's January launch release, and the campaign was found in a June release. The review was run by the author with an AI model; it is not an independent audit.

### How the evidence was gathered

Each workstream agent read the shared rules, the brief and the registers, and wrote one structured file. Each file lists its sources with publication, event and access dates, geography, period, funding and access limits. Each finding carries the supporting passage and locator, an evidence class, a confidence, limitations and counterevidence. In the data tables, every row names its source, period and claim status. The rules require every cited source to be opened, so a search snippet is a lead, not support.

Where a page refused an automated fetch, agents recorded an access error and moved on rather than trying another route. PubMatic's site refused the fetch tool throughout, so its pages were read in a browser by the profile's research agent and by the author's process, and a 2026 parliamentary answer that the fetch tool returned empty was read the same way. 68 registered sources were only partly readable (paywalled, gated, summary only or blocked), and the claims that rest on them say so.

Evidence classes follow the shared rules: direct record, independent measurement, commissioned measurement, vendor assertion, synthesis, inference, forecast and illustrative scenario. A vendor's own documentation is a vendor assertion, however technical. A third party paid by a vendor is not independent, and a study whose funding is unknown is not treated as independent. The author's own essays and earlier papers are registered as author's work (7 sources) and are cited only as the positions the paper tests, never as corroboration. Every absence claim is bounded to what was searched, and the searches are logged in each file.

### How it was checked

Every workstream, the vendor sweep and every profile went to a separate verifier agent that had not seen the research and was told to refute it. Verifiers opened the cited sources themselves and returned one of four verdicts per item: supported as stated, supported with a qualification (with the corrected wording), contradicted (with what the source says) or unresolved. A claim's final wording is its verifier's correction where one exists, and the ledger takes the most severe verdict for any claim checked twice. A finding no verifier checked stays unresolved in the ledger.

![Every stage was built to remove things, and the counts are computed, not typed.](figures/fig-pipeline.svg)

*Figure fig-pipeline. **Every stage was built to remove things, and the counts are computed, not typed.** The five stages of this project with the actual outputs of each. The verifier sample covers the material claims per file, not every claim; the rest are labelled unresolved rather than assumed. Corrections include withdrawn claims and score changes from the review passes. Units: counts. Scope: this research project, as of the cutoff. Caveat: Counts describe process volume, not evidence quality; a claim that survived verification is supported by its source, not proven true. Data: data/stats.json*

Scored cells had a second check. After the scoring pass, a separate verifier for each batch of rows opened the source behind every cell that no stream verifier had checked and every cell scored 4 or 5, and flagged any cell whose score exceeded its anchor or its evidence cap. Its verdicts, and the consistency questions it raised between rows, were decided against the anchors and recorded as scoring corrections, 43 in all. Two help-centre pages that a cell verifier opened, and that changed three of Equativ's cells, were added to the register as hand-added sources. No scored cell counts unless a verifier checked its facts; 0 unverified cells remain.

### How the products were scored

The scoring unit is a named purchased service in one of three cohorts. Exchange curation platforms are tools a curator or buyer uses to build deals on one exchange's inventory. Curated deal products are standardised deal IDs built by their operator. Managed client marketplaces are curated supply pools run for a buyer's own clients. Operator type and ownership are attributes, and a company can have rows in several cohorts. Each cohort has an applicability mask: auction mechanics applies only to exchange curation platforms, and n/a comes only from the mask, never product by product.

Each cell is scored from 0 to 5 against the dimension's anchors, then capped by the evidence: vendor documentation caps a cell at 4 and vendor marketing at 3. A 0 means a source documents absence or failure. Missing evidence is n/e, never 0, a partner's name alone is n/e, and a low disclosure score is never given after an access failure. Each cell records its disclosure mode: public, a documented authenticated interface, request-only or unknown. A product enters the matrix with at least four counted cells; 21 products entered and 7 did not.

Each row was scored once, by an AI scoring agent working from the fixed rules and the verified evidence, in batches of similar rows so that the same standard applied across a cohort. There was no blind second scoring and no inter-rater statistic exists. The Deal Quotient's construction, its tests and its limits are in [chapter 11](#dq). Its field mean is the mean of the cohort's Raw scores, which differs from the Contextual and Native Quotients, which pooled group scores.

### Limits

The paper scores what is documented in public by the cutoff, not what products do. Buyer-facing fee screens, deal consoles and contracts were not observed. Pricing, contract terms, realised prices, adoption and shares of trading are almost never public, and the paper lists them as unanswered rather than estimating them. A product with a thin public record scores lower than it may deserve, and 7 products fell below the admission rule for that reason. The tests in [chapter 11](#dq) show how much single cells move a quotient.

The research ran on 2 and 3 November 2026, against a freeze on the second day. Each file was verified once, deep profiles cover 7 products, and two of the three cohorts have too few band-eligible rows to carry bands. Several material documents could not be read in full: the ANA's Q2 2026 benchmark report (gated), Jounce Media's Supply Path Benchmarking reports (client-only, so its fee distribution rests on two secondary accounts), the full ISBA and PwC second study (summary only), a 2023 thesis on deal IDs and adverse selection (abstract only) and TAG's certification registry (did not render). Claims resting on those documents say so.

The United States is the primary market. Geography scores count documented US availability only; international breadth is recorded and earns nothing.

### Interests

The author has worked at Samba (TV measurement and advertising data; Samba TV until 25 June 2026) since May 2025. Samba owns Semasio, a contextual and audience data company. Semasio's curated-deal product and Samba's PMPs are in the register under the same rules; neither has enough counted cells to enter the matrix, and Samba's PMPs were moved to listed because Samba's own page names no PMP product. Samba also appears among ClearLine Activate's named data partners. The author was EVP at Verve Group from September 2022 to August 2024; Verve's Brand+ Marketplace was scored and fell below the admission rule. He worked earlier at Lotame, whose Curated Marketplaces row is admitted and carries the interest mark, and at Hearts & Science. Interest marks never change a score. No separate reviewer was available to check the conflicted admission and scoring decisions, so [chapter 11](#dq) publishes each cohort's field quotient with the interest-marked rows removed.

The author co-leads the Signals & Measurement working group of AgenticAdvertising.org, which maintains AdCP, one of the protocols assessed in [chapter 10](#agents). He maintains an open-source AdCP signals adaptor, a sell-side decisioning mechanism of the kind this paper studies, and No Fluff Advisory operates two agents listed in the AdCP registry. None of these earns any state or score; the registry entries and the author's materials are excluded from the evidence, and the author's essays are cited only as the positions under test.

No vendor funded, reviewed or saw the paper, and no third party was contacted for it. Every vendor is scored under the same published rules. The paper is AI-assisted research. It has not been peer reviewed, externally audited or independently validated.

### The claims

Every claim marked in the text is listed here with its evidence class, review outcome, confidence and supporting passages.

**Claim ledger**

| ID | Claim (final wording) | Ch. | Evidence class | Review | Confidence | Sources |
|---|---|---|---|---|---|---|
| CUR-A-F07 | ads.txt 1.1 is the current version: the specification's version table dates its final release to July 2022 (public comment April 2022) and the IAB Tech Lab standards page, last updated 14 September 2026, lists it as August 2022; 1.1 added the managerdomain and ownerdomain variables. | 1 | direct record | supported as stated | high | [141, 251] |
| CUR-A-F20 | On 16 December 2024 IAB Tech Lab announced a 2025 Curation Workstream under its Addressability Working Group and the rebranding of Seller Defined Audiences as Curated Audiences, with 'visibility of curation or enrichment partners' and data-leakage prevention as topics and 'plans for testing the framework in 2025'; by 3 November 2026 the Tech Lab's site search for 'curation' (20 results over two pages) returned no new specification, standard or guideline: the only standards page with 'Curated' in its title is the renamed Seller Defined Audiences page (last updated 25 August 2025, still serving the February 2022 specification), and the only specification-grade output naming a curator is the Deal Sync API 1.0, whose deals-related standards work the Tech Lab's own Curated Audiences page places in the 'Curation Sub Working Group' with 'a release expected in Q4 2025', as the September 2025 Onetag member perspective also states. | 1 | synthesis | supported with qualification | medium | [144, 252, 253] |
| CUR-E-F25 | Nexxen's Curated Marketplace page describes building 'custom deals and premium PMPs' by applying a partner's own, Nexxen's or third-party data 'through an AI-powered self-service UI or API' for data owners, retail media, curators, agencies and publishers, and states no auction type, priority, floor, exclusivity rule, fee or DSP name; the general-availability release could not be opened (HTTP 403). | 1 | vendor assertion | supported as stated | medium | [254] |
| CUR-E-F28 | Magnite announced on 1 October 2025 that curation inside ClearLine would let buyers and curators 'define deal terms, pricing, and targeting directly' and was 'Coming soon in beta', reaching '109 million US ad-supported households', and its blog of 9 September 2026 documents self-serve audience application in ClearLine Activate seats through line items and demand tags with Experian, LiveRamp, TransUnion, Throtle and Cadent as data partners; neither page states a deal type, auction type, priority, floor, exclusivity rule, fee, DSP name or general-availability date for curation. | 1 | vendor assertion | supported as stated | medium | [107, 108] |
| CUR-F-F01 | Audigent's SmartPMP was sold on its undated product page on 3 November 2026 as a single deal ID combining publisher inventory with Audigent first-party data; the page names no SSP, exchange or DSP, and the homepage says Audigent is integrated with 'over 20 of the industry's leading SSPs' without naming one. | 1 | vendor assertion | supported as stated | high | [193, 116] |
| CUR-F-F02 | Audigent documents the provenance of SmartPMP data at the level of source type and some providers only: '100% 1st party audience data, never 3rd party cookies', 'verified, opt-in data' and named publisher and data partners (Condé Nast, TransUnion, Warner Music Group, Penske, a360 Media, Fandom), with no per-segment collection method, intended use, permissions or legal basis on the pages read. | 1 | vendor assertion | supported as stated | high | [193, 116] |
| CUR-F-F05 | Experian Marketing Services sold Curated Deals ('a single Deal ID'), Deal Engine (self-service deal creation) and Deal Marketplaces (a library of pre-built deals) on a copyright-2026 page read on 3 November 2026 that names no SSP, exchange or DSP, no fee, no numeric quality claim and no Audigent, and describes its data only as 'Our proprietary audiences, partner audience ecosystem'. | 1 | vendor assertion | supported as stated | high | [255] |
| CUR-F-F10 | Lotame's copyright-2026 capability page sells 'Lotame's Curated Marketplaces' via PMPs from unnamed 'leading SSPs' into unnamed DSPs with no fee or deal ID stated; the exchange routes on record are Google Ad Manager (Google's 4 November 2024 partner list) and Onetag, whose 2 October 2025 release offers the Lotame Data Exchange taxonomy 'for both immediate Deal ID activation, as well as custom segment requests'. | 1 | vendor assertion | supported as stated | high | [13, 196, 207] |
| CUR-F-F12 | Multilocal describes itself as 'a curation services company powered by our curation platform' whose cureon engine is 'Embedded across 21+ SSP curation marketplaces' with none named in text; the exchange routes on official record are Google Ad Manager (partner list, 4 November 2024), Index Marketplaces (launch partner, 23 January 2024) and the Sovrn Ad Exchange (release of 29 September 2026), and Multilocal delivers PMPs 'directly to your DSP seat' with no DSP named. | 1 | vendor assertion | supported as stated | high | [129, 256, 187, 257] |
| CUR-F-F14 | Dstillery's only pages on deal IDs are dated 25 July and 11 August 2022: they say 'we are leveraging the Microsoft Curate supply exchange to create our ID-free Deal IDs' activatable 'on any leading DSP', with no fee or pricing; the 2022 bodies name Microsoft Curate while the launch post's boilerplate still names 'Xandr Curate', so the text was edited after the rename, and no later Dstillery page on deal IDs was found by the cutoff. | 1 | vendor assertion | supported as stated | medium | [61, 70] |
| CUR-F-F15 | Seedtag NeuroX Curation delivers 'standard Deal IDs into existing DSPs' from deals built 'directly on Seedtag's Exchange' with Liz's contextual signals; the undated page names no third-party SSP or DSP, states no fee ('Lower effective CPMs' is unquantified) and claims '3.5x higher neural engagement than non-contextual ads' with the evidence behind a link. | 1 | vendor assertion | supported as stated | medium | [128] |
| CUR-F-F23 | Google's 4 November 2024 Ad Manager curation launch names eight supported curation partners (Audigent, IAS, LiveRamp, Lotame, Multilocal, Permutive, PrimeAudience, Scope3), makes their deals 'available for Preferred Deals and Private Auctions via Display & Video 360 and Yahoo' at launch, and says 'Google Ad Manager will handle billing' without stating a fee, which is the only platform record naming DSP routes for five cohort B or candidate products. | 1 | vendor assertion | supported as stated | high | [196] |
| CUR-F-F24 | Index Exchange's 23 January 2024 launch release names Adelaide, IAS, LiveRamp and Multilocal among Index Marketplaces' global launch partners and Audigent as an early adopter, states 'there is no incremental expense for partners using Marketplaces', and says agencies 'continue to use their preferred DSPs', a second exchange record of routes for cohort B products that their own pages do not name. | 1 | vendor assertion | supported as stated | high | [187] |
| CUR-U-F05 | Index Marketplaces' product page (undated, copyright 2026) sells sell-side intelligence, optimisation and measurement to media owners, media buyers and solution providers, advertises streaming TV and online video, states 'all without any added platform fees', names no DSP and carries no status word, scale figure or geography. | 1 | vendor assertion | supported as stated | medium | [258] |
| CUR-U-F07 | FreeWheel's Curation Hub is a self-serve inventory-curation and deal-management platform for media buyers inside FreeWheel Advertiser Suite, advertised for CTV and activatable 'across any demand-side platform (DSP) including Buyer Cloud', FreeWheel's own DSP being the only one named. | 1 | vendor assertion | supported as stated | high | [16] |
| CUR-U-F32 | Onetag, discovered in the sweep and absent from the register, presents a 'Curation Platform Login' and an 'Exchange Login' without a product page, claims '2,000+ global premium publishers', '5,000+ buyers trading', '35+ global DSP partners', '50,000+ high-impact placements' and '200+ top brands' (all unperiodised), advertises 'custom AI optimization loops' 'with no additional fee', lists offices in Pisa, Milan and London with no US statement, and has acquired Aryel; it is a candidate cohort A product pending a product page. | 1 | vendor assertion | supported with qualification | low | [259] |
| CUR-A-F04 | OpenRTB 2.6-202606 carries no curator, packager or curation-fee attribute: a case-insensitive search of the full 2.6.md text for 'curat' on 3 November 2026 found the word only in the Channel object definition and in the phrase 'exchange curated string names', never on Pmp, Deal or Source. | 1, 3 | direct record | supported as stated | high | [139] |
| CUR-A-F17 | The Deal Sync API 1.0 separates three roles on a deal: origin (the system where the deal is first input, typically an SSP), curator (the packager of inventory, technology and/or data, 'most often, although not always' the party that sold the deal) and seller (the entity that sourced the demand and contracted with the buyer). | 1, 3 | direct record | supported as stated | high | [96] |
| CUR-F-F06 | Comscore launched Comscore Certified Deal IDs on 5 May 2025 'initially available through Magnite's' SSP with a documented selection rule: buyers 'target the Comscore Top 100, 500, or 2,000 site rankings' built from Comscore's MMX and Video Metrix rankings combined with Proximic contextual signals; no DSP and no fee are named. | 1, 3 | vendor assertion | supported as stated | high | [244] |
| CUR-F-F07 | On 10 June 2025 Comscore expanded Comscore-Certified Deal IDs to PubMatic's platform with Adelaide's AU added to its content-measurement and Proximic signals; the release names no DSP, no fee and no numeric quality result, and no current product page was found for the deal IDs. | 1, 3 | vendor assertion | supported as stated | high | [227] |
| CUR-F-F27 | No cohort B product documents the six provenance fields (source, provider, collection method, intended use, permissions, legal basis) per segment on the pages read on 3 November 2026: six name a source type and a provider or named signal (Audigent's named partners and 'opt-in' claim; Comscore's MMX and VMX rankings with Proximic; Adelaide's AU; IAS's and Seedtag's own contextual classifiers; Adloox's attention score), five name a source type only (Experian, Lotame, Multilocal, Samba, Semasio's homepage), one has a 2022 statement only (Dstillery), and none states collection method, intended use, permissions or legal basis. | 1, 3 | synthesis | supported with qualification | medium | [193, 244, 58, 128, 13] |
| CUR-E-F32 | Yieldmo's YMax page calls YMax 'a supply curation platform, not a DSP or SSP', activates packages 'through your existing DSP as standard deal IDs', asserts page-level packaging 'with full URL visibility' that is 'reviewable before you activate' with performance tracked by URL, and offers YMax+ as a managed service with 'exclusive supply access and unified bid decisioning without DSP constraints'; it states no auction type, priority, floor, fee or DSP name. | 1, 3, 4 | vendor assertion | supported as stated | medium | [11] |
| CUR-A-F27 | MRC's Digital Advertising Auction Transparency Standards (January 2026; steer team led by Omnicom with the ANA, WFA, 4As and IAB Tech Lab) apply to any 'Ad Auctioneer', defined as any company operating a system that evaluates bids, selects winners and sets prices 'even if it is not the final auction in a multi-stage auction process', so an SSP or exchange running a curated deal auction is in scope; compliance and MRC accreditation audit are voluntary, and the cover reads 'FINAL DRAFT' while the running header reads 'FINAL - JANUARY 2026'. | 1, 3, 4, 12 | direct record | supported as stated | high | [151] |
| CUR-G-F06 | GroupM introduced Premium Marketplace on 24 February 2022 as 'a unified programmatic marketplace that will increase media buying transparency and efficiency', available in EMEA and North America, 'Supported by global partnership agreements with Magnite and PubMatic', for connected TV, digital video and display, with clients able 'to select all-inclusive, commission-based or SaaS-based payment options'. | 1, 3, 6, 8 | vendor assertion | supported as stated | high | [7] |
| CUR-E-F21 | Index Exchange's release of 7 October 2025 states that 'Index does not charge data fees', that the Data Vendor Ecosystem carries no additional Index fee on data usage, that it was then in early access for Index Marketplaces with seventeen named launch vendors, and that it lets providers 'reach all DSPs' without naming any. | 1, 3, 6, 11 | vendor assertion | supported as stated | high | [188] |
| CUR-E-F23 | OpenX's OpenXSelect and Curated Packages pages describe a curation, supply-side targeting and deal creation platform for brands, agencies and data partners, transacted across unnamed DSPs, with more than 80 data partners, an identity graph available in the US only, named example publishers inside packages and custom packages by contact; neither page states a deal type, auction type, priority, floor, exclusivity rule or fee. | 1, 3, 6, 11 | vendor assertion | supported as stated | medium | [17, 10] |
| CUR-E-F27 | FreeWheel's Curation Hub pages describe a self-service inventory curation and deal management tool that builds data-enriched packages from FreeWheel publisher partners using standardised content metadata and audience signals and activates them 'as deals across any DSP', naming only Buyer Cloud, with reconciled reporting and 'Integration into unified ad decisioning among FreeWheel publishers'; the pages state no deal type, auction type, priority, floor, exclusivity rule or fee and do not name the SSP. | 1, 3, 6, 11 | vendor assertion | supported with qualification | medium | [16, 260] |
| CUR-A-F11 | IAB Tech Lab released the Deals API 1.0 for public comment on 4 December 2025 with comments open to 31 January 2026 and finalised it on 6 February 2026; the GitHub document is titled 'Deal Sync API Specification' while the Tech Lab page calls it the 'Deals API'. | 1, 3, 6, 12 | direct record | supported as stated | high | [261, 249, 96] |
| CUR-A-F15 | The Deal Sync API 1.0 Deal object can declare publisher multiplicity (pubcount: 0 undisclosed, 1 single publisher, 2 multi-publisher), the presence of paid non-publisher data (auxdata: 0 to 4) and whether the site or app list may change in flight (dinventory: 0 to 2); pubcount counts publishing companies only and refers intermediaries to the supply-chain files, and an inventory aggregator 'may also be considered a single publisher'. | 1, 3, 6, 12 | direct record | supported as stated | high | [96] |
| CUR-F-F08 | Adelaide's undated resource page (© 2019–2026) says its high-AU PMPs run 'across SSPs, including PubMatic, Microsoft Advertising (Xandr), TripleLift, Magnite, and OpenX', with placements 'grouped into high-, medium-, and low-AU sets' as separate deal IDs, no thresholds and no fee statement ('a fair price' only), and performance claims of up to 74% higher conversion rates, 111% higher checkout rates and 456% higher site visits (Teads), 77.3% higher unaided awareness and 49% higher consideration (a 'third-party Kantar study', funding unstated), and up to 69% higher conversions, 53% lower cost per conversion and a 60% cumulative conversion-rate increase 'versus standard bidding tactics' (Audi), none with a method on the page; Index Exchange appears on the page only in a logo list, not in the SSP sentence. | 1, 3, 7 | vendor assertion | supported with qualification | medium | [60] |
| CUR-F-F09 | IAS Curation was announced on 4 November 2024 as 'a deal-based enrichment pathway' on Google Ad Manager applying contextual avoidance, brand safety and suitability, MFA filtration and viewability benchmarks 'before it reaches a bidder'; the news page shows a second date of 6 July 2026, names no SSP, DSP or fee, gives no numeric result, and on 3 November 2026 the word 'curation' appears nowhere on IAS's homepage navigation. | 1, 3, 7 | vendor assertion | supported as stated | medium | [58, 262] |
| CUR-G-F09 | dentsu Curate was launched on 15 September 2020 by Dentsu Aegis Network as 'the new programmatic supply solution from DAN', which 'will leverage Xandr's technology platform' to 'curate supply direct from publishers' into 'a central marketplace', first in Hong Kong, South East Asia and Singapore and 'scaled to other APAC markets'; no 2025 or 2026 dentsu record of it was found. | 1, 3, 8, appendix | vendor assertion | supported as stated | high | [6] |
| CUR-A-F08 | sellers.json 1.0 (July 2019; finalised with the SupplyChain object on 31 July 2019 per the Tech Lab page last updated 7 October 2026) classifies every seller account as PUBLISHER, INTERMEDIARY or BOTH, defined by inventory ownership and whether the advertising system pays the entity directly; 1.0 is the only valid version value. | 1, 3, 12 | direct record | supported as stated | high | [142, 263] |
| CUR-A-F09 | SupplyChain 1.0 represents only entities in the direct flow of payment: every node's hp attribute 'should always be 1' in version 1.0, and the specification defers non-payment participants to a future version. | 1, 3, 12 | direct record | supported as stated | high | [143] |
| CUR-A-F16 | The Deal Sync API 1.0 tells buyers that deal criteria sent through the API are not guaranteed to be reflected in bid requests carrying the deal id and that buyers should always apply matching targeting themselves; it also says its purpose includes adding transparency into curated deals 'not currently available in the bid stream'. | 1, 3, 12 | direct record | supported as stated | high | [96] |
| CUR-E-F09 | Microsoft Curate enables 'Cross-device enhanced by Microsoft' automatically and at no cost on any curated deal that targets a segment, with opt-out only through an account representative, which is a documented in-flight enrichment applied by the exchange by default. | 1, 3, 12 | vendor assertion | supported as stated | high | [110] |
| CUR-E-F13 | Google Ad Manager shares contextual signals (URL, App ID, Pub ID, GPID, metro-level geography, TCF consent string), user identifiers and, separately, secure signals with curation partners, who build packages in a dedicated interface from targeting dimensions and up to ten own-data segments joined by AND logic; the agency subscribing to the resulting deal ID sees 'deal metadata including name and pricing' and a 'created by' curator label, and no page opened says the buyer sees the publishers or domains in the package. | 1, 3, 12 | vendor assertion | supported as stated | high | [59, 24, 248] |
| CUR-A-F01 | The current OpenRTB 2.x text is 2.6-202606, released 11 June 2026 and marked Latest; the two 2025 revisions were 2.6-202501 (27 January 2025) and 2.6-202505 (20 May 2025), and since 2.6-202211 the version number increments only on breaking changes. | 1, 4 | direct record | supported as stated | high | [264, 139] |
| CUR-A-F05 | OpenRTB 2.6's implementation notes (section 7.3) say an exchange 'may be configured by a publisher to prioritize all Deal ID bids above open auction bids, or to force a price auction between them with different floors by class of bid', call these 'multiple common practices', recommend that any prioritisation other than by price and any preferential floor for a seat carry a Deal ID, and describe use cases in which prioritisation is expected (a 'First Look Trading Agreement' locked to one buyer; a 'Direct Option Deal' on private inventory) alongside one in which Deal IDs are 'Package Tokens' with their own floors and 'Winner is decided by bid price'; the notes never use the word default. | 1, 4 | direct record | supported with qualification | high | [95] |
| CUR-A-F21 | Prebid's Deals documentation recommends that deal line items in the ad server carry a priority higher than the line items covering open-market header-bidding bids; the page concerns deals with header-bidding partners, mentions no curation, and docs.prebid.org has no page whose path contains 'curat' among about 1,400 sitemap URLs on 3 November 2026. | 1, 4 | direct record | supported as stated | high | [145, 265] |
| CUR-E-F05 | A Microsoft Curate curated deal is built for one selected buyer that cannot be changed after the line item is saved, the same seller inventory remains biddable in the open exchange, and the documentation (ms.date 2025-10-21) names Microsoft Invest and 'all major DSPs' as buying routes without naming any third-party DSP, although Microsoft said on 14 May 2025 that it would 'no longer support media buying through our DSP, Microsoft Invest, starting on February 28, 2026', so at the cutoff the only named route is retired by Microsoft's own statement and no documented route remains. | 1, 4 | vendor assertion | supported with qualification | high | [110, 111, 56] |
| CUR-E-F07 | Microsoft Curate onboards curators through a Microsoft Advertising representative (log-in at curate.xandr.com), lists 'preferred inventory access' and 'specially negotiated rates with sellers' among the assets curators bring, and lets a curated deal source supply either from the open exchange or from deals negotiated directly with sellers. | 1, 4 | vendor assertion | supported as stated | high | [56, 110] |
| CUR-E-F15 | Google Ad Manager's Marketplace Packages (Beta), publisher-curated packages in a limited closed beta for select CTV/OTT and audio publishers, are a distinct transaction type that performs at Open Auction priority by default, can be public or exclusive (exclusive packages invite bidders or DSPs only and may carry a floor), surface in the Authorized Buyers and Display & Video 360 marketplaces, and can be changed in any setting except package type or archived, after which subscribed buyers can no longer transact, with no buyer notification documented. | 1, 4 | vendor assertion | supported as stated | medium | [25] |
| CUR-E-F19 | Index's knowledge base defines a Marketplace deal ID as a private auction offering specific inventory to selected buyers, documents first-price ('the highest bid wins and the winning bidder owes the winning bid price') and fixed-price auction treatment, requires a CPM fee to be lower than the fixed price for a fixed-price deal to transact, and opts media owners in to all Standard Marketplaces by default. | 1, 4 | vendor assertion | supported as stated | high | [109] |
| CUR-E-F26 | TripleLift's TL Direct became generally available on 10 June 2026 to brands, buyers, publishers and curators after a beta; its page offers 1:1 PMPs and curated deals across a '5k+ publisher network', fixed-rate, floor-rate or hybrid deal structures, access by UI, API or MCP, a deal library for buyer discovery, container deals with agency partners, 'No TripleLift curation fees' and free first-party data onboarding, and says its TL Spark layer shapes traffic and adjusts floors; no DSP is named and no auction type, priority or exclusivity default is stated. | 1, 4 | vendor assertion | supported as stated | high | [130, 117] |
| CUR-E-F33 | Verve's Brand+ Marketplace page (formerly Smaato) advertises 'AI-powered deal curation', a Deals Library to 'Activate curated, audience-enriched deals', bid-stream and deal enrichment from SDK-powered deterministic and contextual signals, and delivery across NAM, APAC and EU in 150+ countries; it states no deal type, auction type, priority, floor, exclusivity rule, fee or named DSP. | 1, 4 | vendor assertion | unresolved | medium | [21] |
| CUR-E-F24 | Equativ's Maestro help centre (updated 23 April 2026) documents three deal pricing models (Market Floor, Custom Floor, Fixed Price), buyer seats or DSPs as a foundational deal setting, meta-deals that aggregate Auction packages, Preferred deals or Private auctions, curator margins or additional costs that 'influence the final media price paid to publishers', and a deal ID generated at finalisation for DSP activation with 'automatic DSP synchronization for select platforms', while the marketing page adds '500+' self-serve clients and auto-generated PMP configurations; neither page states auction type, priority, or whether the buyer sees the margin. | 1, 4, 6 | vendor assertion | supported as stated | medium | [114, 15] |
| CUR-E-F11 | On Google Ad Manager the curation partner enters a package or segment fee as a percentage of media or a CPM, may mark it visible or hidden from the buyer, and Google pays the partner out of the bid price, which the publisher page says is 'taken out of the bid price before it enters the auction'; the Ad Manager publisher page calls the fee 'fully transparent to the buyer' while the Authorized Buyers agency page says the agency 'may or may not see the fee value depending on the curation partner's visibility settings'. | 1, 4, 6, 12 | vendor assertion | supported with qualification | high | [24, 59, 248] |
| CUR-E-F18 | Index Exchange's knowledge base (last updated 9 September 2026) states that bids in Index Marketplaces compete net of fees removed before the auction, that a marketplace partner sets its fee as a percentage of media spend or a fixed CPM (USD only), that Index's revenue share is negotiated with each media owner, and illustrates a $10 gross bid less a 5% partner fee and a 15% Index share paying the media owner $8.08 (the partner $0.50, Index $1.42), or $7.65 with a $1.00 CPM partner fee (partner $1.00, Index $1.35). | 1, 4, 6, 12 | vendor assertion | supported as stated | high | [109] |
| CUR-E-F29 | Criteo's Commerce Grid product page calls the product 'the world's first commerce SSP', trades 'via open exchange and via deal IDs' with 'seamless PMP activation', asserts that it 'Prioritizes high-value deals in auctions, reducing unnecessary traffic', that 'Fees only apply if curated deals win impressions', activation 'through any DSP' with '60+ third party DSPs plugged directly', and consented Commerce Audiences data, and names no DSP, auction type, floor, exclusivity rule or fee rate; the five-step deal guide the stream tried returned 404, but Commerce Grid's knowledge base is live and documents a Criteo tech fee charged to the curator and a curator margin, curated bids submitted to the final auction net of those fees and charged only on a win, deprioritisation of low-value curated deals in auction processing, one DSP seat per demand deal, and Deal Sync to DV360 and The Trade Desk. | 1, 4, 6, 12 | vendor assertion | supported with qualification | medium | [266] |
| CUR-E-F03 | Curators on Microsoft Curate take no margin by default; a margin set as a percentage of the buyer's bid or a fixed CPM is deducted, together with Microsoft's curator tech fee, before the net bid enters the seller's auction, and on seller deals with a fixed floor the configured margin is a minimum because the curator takes the difference between the curator and seller floors; fees apply only when the curated bid wins; the platform enforces a maximum margin rate whose value the pages do not state; and no page opened says whether the buyer sees the margin. | 1, 4, 6, appendix | vendor assertion | supported with qualification | high | [115, 111] |
| CUR-A-F02 | OpenRTB 2.6-202606 defines the Pmp object with three attributes, of which private_auction (integer, default 0) restricts bidding to the listed deals when set to 1; the default is an open auction in which all bids are accepted. | 1, 4, 12 | direct record | supported as stated | high | [139] |
| CUR-A-F03 | OpenRTB 2.6-202606's Deal object carries id, bidfloor (float, default 0, CPM), bidfloorcur (default USD; does not inherit from Imp.bidfloorcur), at (1 first price, 2 second price plus, 3 the bidfloor is the agreed deal price), wseat (buyer-seat allow-list), wadomain (advertiser-domain allow-list), guar (default 0), mincpmpersec, durfloors and ext. | 1, 4, 12 | direct record | supported as stated | high | [139] |
| CUR-A-F29 | On floors, the MRC auction standard requires the use of reserve prices to be clearly disclosed and the reserve price to be reported to participants before and after an auction, states that 'reserve prices should not be set on a per-buyer basis – the same floor should apply simultaneously to all buyers', and says that in OpenRTB auctions employing reserve prices 'should be managed through a Deal ID structure'. | 1, 4, 12 | direct record | supported as stated | high | [151] |
| CUR-A-F30 | On winner determination, the MRC auction standard requires the variables and relative weighting that affect the winner beyond price, including 'demand source priority' and 'seller defined business rules', to be made known and reported, and requires that 'direct deals or certain managed service buys that supersede an auction and may be the reason other bidders lose despite price' be generally disclosed. | 1, 4, 12 | direct record | supported as stated | high | [151] |
| CUR-A-F31 | On fees, the MRC auction standard requires the variables affecting nominal-to-effective bid conversion, including 'deal discounts or tech fees (gross or net)', to be made known and reported, requires how discounts and tech fees are applied to be consistent and transparent, and encourages fees to be applied pre-auction ('net bidding'). | 1, 4, 12 | direct record | supported as stated | high | [151] |
| CUR-U-F03 | Microsoft's documentation of 21 October 2025 defines Microsoft Curate's curators as retailers, data companies, independent trading desks and other media companies, lets buyers act as curators, and states that curated deals 'can be bought via Microsoft Invest and all major DSPs', a route statement that predates and has not been updated for the announced end of Invest's media buying on 28 February 2026. | 1, 5 | vendor assertion | supported as stated | high | [56] |
| CUR-U-F27 | Adloox's learning-center page (17 July 2026) documents that a buyer can 'select Adloox Attention Segments while building curated deals on PubMatic, thanks to our partnership with Qwarry' in the PubMatic Media Console via 'Create Auction Package Deal', priced at '$0.15/CPM', for desktop and mobile web with the instruction to avoid CTV, and states 'Adloox is now part of Scope3'; the deal is built by the buyer in PubMatic's tool, so Adloox supplies a segment rather than a standardised deal product. | 1, 5 | vendor assertion | supported as stated | high | [12] |
| CUR-U-F33 | Permutive's documentation describes a curation product in which publishers opt in first-party signals, 'Permutive handles all demand generation, audience packaging, deal creation, campaign optimization, and buyer reporting', advertisers access 'Permutive-branded curated deals' 'through Monetize, Index Exchange Marketplaces, Pubmatic, Google Curated Deals, and other SSP curation seats' (OpenX listed as an activation bullet), 'Permutive maintains transparent fee structures capped at 40%' with no base or payer stated, and the product's environment table marks only Web as supported (iOS, Android, CTV and API Direct 'No'); it is a candidate cohort B product absent from the register. | 1, 5 | vendor assertion | supported with qualification | medium | [267] |
| CUR-U-F08 | Google Ad Manager's curation settings remain in Beta on the help page read on 3 November 2026 (with the caveat that Beta features 'might not be available in your network'); curation partners define the fee of a segment or package, Google describes it as 'a single fee that is fully transparent to the buyer' and pays it to the partner 'taken out of the bid price before it enters the auction'; the page lists twelve named curation partners: Adelaide, Audigent/Experian (Audience), Audigent/Experian (Contextual), DoubleVerify Inc., Integral Ad Science, LiveRamp, Inc., Multilocal Media, Permutive, PrimeAudience, Proximic by Comscore, SWYM.ai and System1. | 1, 5, 11 | vendor assertion | supported with qualification | high | [59] |
| CUR-U-F17 | Experian's Consumer View curation page sells three named services: Curated Deals ('Bespoke private marketplaces (PMPs) that combine premium data and inventory into a single Deal ID'), Deal Engine (self-service deal creation for buyers) and Deal Marketplaces (a library of pre-built curated deals), the last described as ready to activate 'on platforms like demand-side platforms (DSPs) and supply-side platforms (SSPs)', under the copyright of Experian Information Solutions, Inc., with no channel, named platform, fee or figure, no mention of Audigent, and only site-level US cues (a 'United States' country-selector default and a US toll-free number) rather than a product geography statement. | 1, 5, 11 | vendor assertion | supported with qualification | high | [255] |
| CUR-U-F26 | Samba's site documents PMPs in one homepage sentence ('exclusive segments and PMPs built from Samba's deterministic cross-screen' signals, page metadata 2 November 2026) under its Audiences product, naming no platform, deal ID, channel or fee, and the Audiences page itself (per U) describes segments only, so the register's 'Samba PMPs' row rests on a single marketing sentence and is not documented as a purchased deal service. | 1, 5, 11 | vendor assertion | supported with qualification | medium | [268] |
| CUR-E-F20 | Index Marketplaces partners choose the level of fee transparency they disclose in reporting: under the disclosed model the Marketplace Owner Fee and Vendor Fee are 'optionally disclosed', under the undisclosed model media spend is 'Always disclosed for buyers, Undisclosed for Media Owners' and the fee lines are undisclosed. | 1, 6 | vendor assertion | supported as stated | high | [109] |
| CUR-F-F18 | Adloox sells no curated-deal product; its learning-center page (published 17 July 2026) documents four attention segments ('Top 20%' and 'Top 40% of URLs with the best Adloox attention score', desktop and mobile, 'updated daily') that a buyer adds inside PubMatic's 'Create Auction Package Deal' flow via Qwarry at 'Pricing: $0.15/CPM', the only charge amount with a base found on any cohort B page; payer, collector and period are not stated. | 1, 6 | vendor assertion | supported as stated | high | [12] |
| CUR-A-F13 | curfeetype declares the type of curation fee, never its amount, and the specification's guidance text calls the same attribute 'curationfee', an internal inconsistency in the final 1.0 text. | 1, 6, 12 | direct record | supported as stated | high | [96] |
| CUR-F-F16 | Semasio's HypertailPMP service page returned HTTP 404 on 3 November 2026 (as on 2 November); its homepage (copyright 2025, banner 'Semasio is now part of Samba') lists the service as 'Reach your ideal consumers in long-tail media spaces', names PubMatic only in a testimonial and states no fee, so whether a curated-deal product is sold today is unresolved. | 1, 7 | vendor assertion | supported as stated | high | [269, 270] |
| CUR-E-F06 | Microsoft Curate curators compose a deal from universal categories, seller-defined categories, individual sellers and the publishers those sellers expose, allowlists and blocklists (the AppNexus Blocklist always applies), with viewability or completion thresholds as hard limits; the buyer is shown the deal name but not the line-item name; ads.txt filtering is unsupported for curated deal line items; the pages do not state whether the buyer can see the seller list; and sellers can block named curators and report by curator member ID. | 1, 7, 12 | vendor assertion | supported as stated | high | [110, 111] |
| CUR-G-F02 | On Stagwell's own record the rule owner of Stagwell Curate is Stagwell (it applies 'its own intelligence, quality standards and performance signals upstream'); the release says partners 'including Freewheel and Magnite will provide access to deep, direct publisher relationships' (future tense, like the agent connection), and the supply-quality models are 'powered by The Trade Desk's Sincera'; no third party is named as setting the rules. | 1, 8 | vendor assertion | supported with qualification | high | [5, 71] |
| CUR-G-F13 | Walmart Connect introduced Connect Select on 27 April 2026 as 'a curated marketplace' inside Walmart DSP (Magnite's headline: 'Walmart DSP Marketplace Connect Select') that 'At launch' 'includes inventory from VIZIO, premium CTV publishers, including Paramount, and Warner Bros. Discovery' and SSPs the article lists as FreeWheel, Index Exchange, Magnite and PubMatic, so the supply is multi-publisher and multi-exchange on the operator's list; deals 'are available through Inventory Controls and Deal Desk', and the article attributes 'post-auction optimizations' to the SSP partnerships. | 1, 8 | vendor assertion | supported with qualification | high | [271, 272] |
| CUR-G-F17 | Roku Curate (page dated 27 April 2026) 'combines Roku's proprietary insights and inventory' with 'in-market targeting data from partners like Best Buy Ads, Criteo, Fandango, Fetch, Instacart and Kroger'; partners supply data, not inventory; Roku contrasts it with a standard deal ID or PMP, where 'the publisher sends inventory, while the targeting and decisioning happen at the DSP level', saying that with Roku Curate 'Roku packages the data and media supply'; 'Roku can then activate directly or send the package to the advertiser's preferred demand-side platform', so it is single-publisher and falls outside the multi-publisher subset. | 1, 8 | vendor assertion | supported with qualification | high | [8] |
| CUR-G-F20 | Google's Display & Video 360 Marketplace help page describes 'a centralized location for discovering and activating inventory' with a Featured tab of 'curated bundles' ('Lists of curated inventory that advertisers can assign directly across line items') and a Discover tab listing 'publishers, packages, central partner deals, and private deals'; it does not say who assembles packages or bundles and mentions 'The exchange provided deal floor price'. | 1, 8 | vendor assertion | supported as stated | high | [273] |
| CUR-G-F23 | Yahoo announced Backstage on 19 June 2023 as 'a new offering that will provide Yahoo DSP advertisers with a direct path to curated, premium publisher inventory' that 'enables publishers to integrate directly with the Yahoo DSP', with 'over 100 publishers' signed and availability 'to all Yahoo DSP advertisers in Q3 2023'; the current page calls it 'curated MFA-free inventory' without naming the curator or a measurer, and neither document states a fee. | 1, 8 | vendor assertion | supported as stated | high | [274, 275] |
| CUR-G-F24 | Viant's Direct Access page describes 'Viant's innovative supply path optimization program' that brings 'buyers and sellers closer through direct inventory and first-party data integration', shows 19 CTV and audio publisher logos and names Disney and NBCUniversal in its FAQ, and sells IRIS-enabled inventory as 'Independence from SSP-curated deal ID limitations'; it promises to 'Reduce ad tech fees and non-working media spend' but states no fee amount, take rate or selection rule. | 1, 8 | vendor assertion | supported with qualification | high | [276] |
| CUR-U-F31 | Roku's page of 27 April 2026 describes Roku Curate as 'a premium advertising package' that 'combines Roku's proprietary insights and inventory with high-value partner data' from Best Buy Ads, Criteo, Fandango, Fetch, Instacart and Kroger, activated by Roku directly or sent to the advertiser's preferred DSP (none named), with 'lower fees' asserted without amounts; the inventory is Roku's own, so the product sits outside the multi-publisher subset. | 1, 8 | vendor assertion | supported as stated | high | [8] |
| CUR-U-F15 | PubMatic's buyer page describes inventory 'across CTV, video, mobile app, and display—curated through our SSP and enriched by Connect's data signals', says 'Curated deals can be activated through PubMatic's Activate solution or your DSP of choice', advertises a 'Gen AI-curated marketplace' built from natural-language prompts, and gives company-wide figures of 1,900 premium publishers and 'over 821 billion daily ad impressions'; the Connect for Publishers page gives data owners 'access to Deal ID transactions'; the seed sentence 'transparent, curated packages transacted via a single deal ID' is not on the buyer page read on 3 November 2026. | 1, 8, 10 | vendor assertion | unresolved | medium | [223, 20] |
| CUR-U-F10 | InMobi's Buyer Hub page describes 'An AI Agent for Media Buyers' that lets users 'Discover inventory, build deals, and optimize campaigns through natural language, with traders in control of every decision', records 'Agency Access \| August 31st, 2026', names no exchange, DSP or deal-ID mechanism, labels a feature 'Self Serve Premium Supply', and mentions the US only in the alt text of a sample-query image; InMobi's Business Wire release of 13 August 2026 could not be opened (HTTP 403). | 1, 11 | vendor assertion | supported with qualification | low | [174] |
| CUR-U-F12 | Sovrn's buyer page offers 'private marketplace deals, attention deals, curated packages, and Commerce Audiences' under an offer heading rather than a named platform, advertises connected TV and streaming, web display and outstream video, and mobile app inventory, states sales coverage 'US / EU / CA / APAC', and names the operating company as 'Sovrn, Inc.' in its footer (the register's 'Sovrn Holdings, Inc.' appears only in a LinkedIn URL). | 1, 11 | vendor assertion | supported as stated | high | [277] |
| CUR-U-F13 | Yieldmo's YMax page calls 'YMax.ai' 'the only supply curation platform' and says 'YMax deal packages activate through your existing DSP as standard deal IDs', advertises video and CTV, offers 'YMax+' as 'a fully managed service', and does not describe a tool by which a third party builds deals on Yieldmo's exchange, so the purchased service as documented is operator-built deal packages (cohort B by the decision tree) rather than a cohort A build tool. | 1, 11 | inference | supported as stated | medium | [11] |
| CUR-U-F14 | Verve's Brand+ Marketplace page (interest-marked) carries the former name in its title ('Brand+ Marketplace (formerly Smaato)'), advertises CTV, video streams, mobile apps and the open web with 'AI-powered deal curation' and a 'Deals Library' as features, states reach in '150+ countries' and delivery 'across NAM, APAC and EU', names no DSP, and identifies the operating entity as 'Verve Group, Inc.'. | 1, 11 | vendor assertion | unresolved | high | [21] |
| CUR-U-F24 | Seedtag's NeuroX Curation is 'The Self-Serve Interface to the Neuro-Contextual Exchange', giving activation teams 'pre-bid control of Seedtag's Neuro-Contextual Exchange' over 'One owned supply path' to 'Build Web and CTV deals in a single flow' and push 'standard Deal IDs into existing DSPs'; the page's body names only web and CTV deals (OLV and display are navigation labels), and the self-serve build tool on the operator's own exchange reads as cohort A by the decision tree's first step, subject to the author's decision. | 1, 11 | inference | supported with qualification | medium | [128] |
| CUR-A-F14 | Version 1.0 of the Deal Sync API supports no deal revisions, proposals or negotiations, while allowing that attributes may change during a deal's flight; differential overrides are promised for 1.1. | 1, 12 | direct record | supported as stated | high | [96] |
| CUR-A-F33 | The MRC auction standard requires material changes to auction systems or methods to be communicated to participants in advance under a formal change notification policy, with version histories and impact assessments, and requires live tests of auction changes to follow MRC Minimum Standards disclosure; for unified seller platforms it acknowledges contractual limits on sharing 'deal terms' and requires publisher-approved minimum disclosure. | 1, 12 | direct record | supported as stated | high | [151] |
| CUR-E-F08 | On Microsoft Curate the curator can edit a deal's margin at any time, cannot change the buyer after saving, can delete a deal and still report on it, and the documentation states no buyer notification, audit right, export right or remedy for a changed or misdescribed deal. | 1, 12 | vendor assertion | supported as stated | medium | [115, 113, 110] |
| CUR-E-F17 | On Google Ad Manager a curation partner can change the fee amount of a live data segment but not its fee type, and no opened page documents notification to buyers of fee or package changes, audit rights, export rights or remedies. | 1, 12 | vendor assertion | supported as stated | medium | [24, 25] |
| CUR-E-F39 | On every page opened for the fifteen products, the only documented post-purchase controls belong to the curator or publisher (Microsoft: margins editable at any time, buyer fixed; Google: fee amount changeable on live segments, packages editable except type, archiving cuts off subscribed buyers; Index: partner-chosen fee disclosure; Commerce Grid: curation fee amount adjustable after a deal is synced to the DSP), and no page documents change notification to the buyer, an audit right, data exportability for the buyer or a remedy for misdescribed inventory or charges. | 1, 12, appendix | synthesis | supported with qualification | medium | [115, 24, 25, 109] |
| CUR-B-F02 | In the 2023 study, 59% of matched programmatic ad spend went to the open marketplace and 41% to private-marketplace deals (spend-weighted, no CTV), while the Q1 2025 report restates the same 2023 data as a 54.2% median PMP share across advertisers and a 40.9% aggregate. | 2 | independent measurement | supported as stated | high | [26, 63] |
| CUR-B-F09 | For July to September 2025 (21 contributing of 39, $142 million, 17.5 billion impressions) the aggregate PMP share fell one point to 81.6% while the median rose to 88.8%; CTV was 99.2% PMP by spend and web plus mobile 65.2%; spend-weighted PMP CPM rose to $9.88 against $3.55 for OMP. | 2 | independent measurement | supported as stated | high | [98, 100] |
| CUR-B-F13 | The Q2 2026 benchmark (26 August 2026, 'Working media improves, but pricing pressure returns') is readable only as a title and three-sentence summary without login, and no Q3 2026 edition was listed on 3 November 2026, so the ANA series ends at Q1 2026 for this paper. | 2 | direct record | supported as stated | high | [278, 279] |
| CUR-B-F19 | Jounce describes two curation mechanisms: an exchange that packages 'supply from multiple publishers through a curated deal' and 'occupies the strategic control point', and third-party curators that operate 'on top of an ad exchange, evaluating each ad opportunity before bid requests are issued to DSPs and instructing the exchange to decorate each outbound bid request with one or more curated deal IDs'. | 2 | independent measurement | supported as stated | high | [101] |
| CUR-B-F21 | AdExchanger reports from the same Jounce analysis that 'the average bid request now includes an open auction, as well as 3.5 different deal IDs' and that 'large DSPs see more than 1,000 eligible deal IDs for a single impression', which is why an auction-share figure cannot be converted to a spend or deal-ID share. | 2 | independent measurement | supported as stated | medium | [64] |
| CUR-B-F23 | Jounce's free 2023 edition projects the open internet at '$85B' for 2023 and open programmatic at $81.1 billion, whereas the ANA attributes an '$88 billion' global open-web estimate to 'State of the Open Internet Report, Jounce Media, 2023'; the 2023 edition's extractable text carries no auction-mix chart with deal-ID shares, the 2024 edition is subscriber-only and no 2026 edition was listed on 3 November 2026, so the March 2023 point of the Jounce series could not be established from public documents. | 2 | direct record | supported as stated | medium | [104, 26, 280, 281, 101] |
| CUR-B-F24 | ISBA's 2020 study (ISBA with the AOP, carried out by PwC, published 6 May 2020) covered 15 advertisers, eight agencies, five DSPs, six SSPs and 12 publishers representing about £0.1 billion of UK programmatic spend, identified over a thousand distinct supply chains, matched 290 of them end to end (31 million impressions, a 12% match rate), and found publishers received 51% of advertiser spend with 15% an 'unknown delta'. | 2 | independent measurement | supported as stated | high | [105, 106] |
| CUR-D-F21 | Criteo's FY2025 10-K (Criteo S.A., incorporated in France, Nasdaq Global Select Market, CRTO) describes Commerce Grid as 'a commerce-focused Supply Side Platform ("SSP") for media owner data and inventory monetization', certified by TAG for transparency, and states that Criteo typically charges retailers a negotiated SSP fee and sometimes a technology fee while brands pay a negotiated DSP fee; no fee rate and no 'curation' or 'curated' wording appears in the portion read, and the Iponweb acquisition appears only in XBRL tags there. | 2 | direct record | unresolved | high | [282] |
| CUR-B-F25 | ISBA/PwC's second study (fieldwork 1 September to 31 October 2022, published 18 January 2023; 11 advertisers, seven agencies, six DSPs, six SSPs and 10 publishers) analysed 1.3 billion impressions, of which 104 million were matchable and 61 million (58%) matched buy side to sell side, reduced the unknown delta to 3% (under 1% for private marketplaces) and lifted publisher revenue to 65% of spend; PMP deals were about one-fifth of matchable impressions and matched above 70%, 'in part due to Deal IDs facilitating impression matching from DSP to SSP'. | 2, 3, 7 | independent measurement | supported as stated | high | [106, 283] |
| CUR-B-F01 | The ANA's 2023 transparency study covered 21 marketers, $123 million of ad spend and 35.5 billion impressions between September 2022 and January 2023, reconciled by TAG TrustNet from impression log-level data, with 35 qualitative interviews by Kroll. | 2, 4, 7 | independent measurement | supported as stated | high | [26] |
| CUR-B-F03 | The 2023 Complete Report sorted PMP deals into three profiles by domain count (1 to 10 publisher-specific; 11 to 500 seller-specific 'auction packages'; over 500 'SSP direct'), and stated that publisher-sold deal IDs 'often include prioritized access to premium supply' whereas intermediary-sold deal IDs 'bundle together sites and apps from many publishers and rarely carry auction priority', their value proposition being 'inventory curation'. | 2, 4, 7 | independent measurement | supported as stated | high | [26] |
| CUR-D-F26 | None of the seven FY2025 annual filings read (Magnite, PubMatic, Criteo, Comscore, The Trade Desk and Viant 10-Ks; Nexxen's 20-F) discloses, in the Business and risk-factor sections read (the first 100,000 characters of each 10-K and 200,000 of the 20-F, which end before the MD&A and financial statements), a numeric take rate or a deal, PMP or curated share of its own activity; each describes its compensation as a fee that is generally a percentage of spend (with flat fees or fixed CPMs at Nexxen and Viant, and negotiated SSP, technology and DSP fees at Criteo), so those sections supply no SSP-level series for H1 and no product-level charge for H2b. | 2, 5, 6 | synthesis | supported with qualification | medium | [212, 231, 284, 30, 285, 282] |
| CUR-B-F06 | For April to September 2024 (21 contributing advertisers of 37 participating, $235 million, 38.5 billion impressions), the ANA reported PMP at 59% of spend excluding CTV and 66% including CTV ('averages across marketers'), average PMP CPM $7.46 against $3.26 for OMP, 97% of CTV spend transacted through a deal ID, CTV PMP CPM $15.00 against $5.54, and TrueAdSpend of 43.9 cents per dollar; the Q1 2025 report restates the period as a 56.2% median and a 65.9% aggregate PMP share. | 2, 6 | independent measurement | supported as stated | high | [102, 286, 63] |
| CUR-B-F08 | For April to June 2025 (21 contributing of 39, $146 million, 18.6 billion impressions) the median PMP share rose to 87.8% (a median per the Q3 2025 report's median table) and the aggregate to 82.6%; PMP CPMs were $7.15 against $4.41 for OMP (the ANA's release calls these 'averaging' and the Q1 2026 table carries them without a basis label; the gated Q2 2025 report is the only place a median label could be confirmed) and spend-weighted CPMs were $7.43 against $3.64; CTV reached 44.2% of spend and the ANA put its efficiency opportunity at $26.8 billion (21.8% of a $123 billion Madison and Wall base that includes DOOH and digital audio). | 2, 6 | independent measurement | supported with qualification | high | [66, 98, 99] |
| CUR-B-F15 | The benchmark methodology states that where SSP log-level data or identifiers are unavailable 'estimated SSP fee rates are applied based on periodic SSP disclosures', prioritises medians over averages with one data point per marketer, and excludes an advertiser from a segment's distribution below $1,000 of spend; it publishes no match-rate or coverage figure. | 2, 6 | direct record | supported as stated | high | [287, 102] |
| CUR-B-F20 | Jounce's April 2025 Supply Path Benchmarking report, as reported by Playwire (10 April 2025) and AdExchanger (1 May 2025), found that about half of curated deal IDs (Playwire's headings say 'of bid requests') carried no fee beyond the exchange's standard take, 34% a margin-based fee (median 14%), 10% a fixed CPM fee (median $0.48) and 7% a static price, with a median effective margin of 30% and a median effective CPM of $0.51 per Playwire; neither account states the sample month, AdExchanger's March 2025 dating applies to its 1.5 billion-request auction-mix analysis, and the report itself is client-only. | 2, 6 | independent measurement | supported with qualification | medium | [9, 64] |
| CUR-H-F06 | In the ANA's Q1 2026 report (66 contributing of 86 participating marketers, $160 million, 20.9 billion impressions, January to March 2026), the higher-performing half by TrueAdSpend carried transaction costs of 27.1% against 29.5% for the lower half (DSP platform 6.6% against 8.2%, DSP data 3.4% against 5.0%, DSP other 4.3% against 4.5%, SSP platform 12.8% against 11.8%), seller revenue 72.9% against 70.5%, PMP share 61.1% against 62.2%, average CPM $5.45 against $7.40 and TrueCPM $7.46 against $19.04; the report attributes the gap to media productivity, not transaction costs. | 2, 6 | independent measurement | supported as stated | medium | [99] |
| CUR-H-F14 | PubMatic's FY2025 Form 10-K states that its fees to publishers are 'generally a percentage of the value of the advertising impressions', that revenue is reported net of amounts paid to publishers, that SPO 'represented approximately 55% of total activity' in 2025, that it has 'experienced fee pressure' and faces DSP requests for 'volume discounts, fee revisions, and rebates', and reports revenue of $282,926 thousand (2025) and $291,256 thousand (2024); Connect is 'our solution that provides additional data and insights to buyers', sold separately, and no numeric take rate, gross spend or Connect fee appears in the portions read. | 2, 6 | direct record | supported as stated | high | [231] |
| CUR-B-F07 | For November 2024 to March 2025 (23 contributing of 39 participating, $242 million, 41.9 billion impressions), the median PMP share was 64.5% and the aggregate 67.2%; excluding CTV the split was 53% PMP to 47% OMP, and 99.5% of CTV spend was executed through deal IDs; spend-weighted CPMs were $6.09 overall, $10.43 PMP and $3.03 OMP, with medians of $5.62, $9.37 and $3.28. | 2, 6, 7 | independent measurement | supported as stated | high | [63] |
| CUR-H-F07 | The ANA benchmark's published methodology applies 'estimated SSP fee rates' from 'periodic SSP disclosures' where SSP log-level data or identifiers are unavailable, defines SSP costs as 'investment in supply-side technology, inventory access, and curation services', and excludes agency, ad-serving and managed-service fees from the cost waterfall; so an ANA SSP fee figure can be partly imputed, and any curation fee collected through an SSP sits inside the SSP bucket with no separate curation-fee line. | 2, 6, 12 | direct record | supported as stated | high | [287, 99, 26] |
| CUR-H-F24 | IAB Tech Lab's Deal Sync API 1.0 Curation object carries 'curfeetype' with values 0 undisclosed, 1 percentage of spend, 2 flat fee, 3 CPM and 4 'no fee is paid for curation services', and its guidance says the attribute 'provides information about the type of fee being applied in the bidstream, but not what that fee is'; the specification therefore standardises disclosure of a fee type, never an amount, base, payer or collector, and 1.0 'does not support differential overrides'. | 2, 6, 12 | direct record | supported as stated | high | [96] |
| CUR-B-F12 | For January to March 2026 participation reached 86 marketers with 66 contributing data ($160 million, 20.9 billion impressions), median CPMs fell to $4.41 from $5.55, the market-level TrueAdSpend index was 43.3% and median MFA exposure rose to 1.1%; the Q1 2026 report does not restate the PMP share series, saying only that 'Marketplace spend percentages are comparable' between its higher- and lower-performing cohorts, whose tables show PMP ad spend at 61.1% and 62.2%. | 2, 7 | independent measurement | supported as stated | medium | [103, 99] |
| CUR-B-F26 | PwC's 2023 summary recommends that advertisers, agencies, ad-tech vendors and publishers 'consider investing more in well-curated PMPs, given their higher impression match rates and publisher revenues', while noting that lower fraud, viewability, brand-safety and data-leakage risks were 'outside this study'. | 2, 7 | independent measurement | supported as stated | high | [106] |
| CUR-U-F18 | Comscore announced on 10 June 2025 (Reston, VA) the expansion of its Comscore-Certified Deal IDs to include Adelaide's AU metric, enabled in PubMatic's SSP and covering 'many of the top CTV publishers', with no DSP named and no fee or volume disclosed; the release identifies Comscore, Inc. as 'Comscore (NASDAQ: SCOR)' and Proximic as the source of 'its Proximic programmatic targeting capabilities'. | 3 | vendor assertion | supported as stated | high | [227] |
| CUR-AUDIGENT-F04 | Audigent names the selection logic of its deal products (audience match to 'premium curated publisher inventory'; contextual isolation of 'hundreds, or even thousands of contextually relevant inventory sources'; AI and machine learning that 'package and optimize consumer-safe data with premium inventory supply'; 'hand-vetted inventory' in DV360), but no page read on 3 November 2026 states a threshold, an inclusion or exclusion rule, a quality filter or an update cadence. | 3 | vendor assertion | supported as stated | high | [193, 194, 192, 190] |
| CUR-AUDIGENT-F05 | No source opened for this profile lists the sellers, domains or a seller count of any identified SmartPMP, ContextualPMP, CognitivePMP or Audigent package. The composition wording is product-wide marketing, made up of adjectives ('premium curated publisher inventory', 'hand-vetted inventory'), publisher verticals ('some of the biggest names in entertainment, sports, and lifestyle'; DV360 deals 'covering various sectors' and sports content) and an unquantified 'hundreds, or even thousands' of inventory sources. The homepage's 476,080 web/mobile sites is a company-wide, unperiodised figure. | 3 | synthesis | supported with qualification | medium | [193, 116, 190] |
| CUR-AUDIGENT-F06 | Audigent documents the provenance of its deal data per segment only at the level of source type and named providers ('100% 1st party audience data, never 3rd party cookies'; 'verified, opt-in'; Condé Nast, TransUnion, Warner Music Group, Penske, a360 Media, Fandom; Experian deterministic signals and The Weather Company in the DV360 packages). Its Platform Privacy Policy (last updated 2 September 2026) adds platform-wide collection methods (cookies, beacons, tags, mobile SDKs, ID syncing), purposes and legal bases (consent for targeted advertising; legitimate interests), with no per-segment or product-specific documentation. | 3 | vendor assertion | supported with qualification | high | [193, 116, 190, 239, 194] |
| CUR-AUDIGENT-F14 | In the sources screened by 3 November 2026, this review found no certification, accreditation or audit covering Audigent or its deals, no supply-chain sample (ads.txt, sellers.json, SupplyChain) for any Audigent deal, and no statement of IVT filtering method or measurer. | 3 | synthesis | supported with qualification | low | [116, 195] |
| CUR-INDEXMARKETPLACES-F03 | The package builder (updated 26 October 2026) documents the selection rule a partner sets: publisher targeting (blank for all eligible publishers) or deal targeting (beta), domain and app-bundle include and exclude lists of up to 10,000 entries with exclude taking priority, up to four include segment rule sets and one exclude, Web or In-App environments, device types including three CTV types, Banner, Native and Video formats, geography (zip and postal codes in beta), dayparting, a minimum viewability probability, VCR and CTR (both beta), video placement settings, streaming-TV content (networks and channels in beta), IAB Tier 1 and Tier 2 categories and custom key values (beta), with at least one targeting dimension other than dayparting required. | 3 | vendor assertion | supported with qualification | high | [14] |
| CUR-INDEXMARKETPLACES-F12 | The knowledge base listed 49 active Data Vendors on 21 October 2026 (up from seventeen named at the Data Vendor Ecosystem's early access on 7 October 2025), each with data types, channels and markets, and a data-usage-conditions page restricts LiveRamp Data Marketplace segments to US users, with Index reserving the right to update targeting on deals that use them 'to include the United States and only the United States'; a partner must have the vendor's permission to target its segments, and no page opened documents collection method, intended use, consent or legal basis per segment. | 3 | vendor assertion | supported with qualification | medium | [94, 240, 14, 188] |
| CUR-LOTAME-F07 | Lotame's PHD EMEA case study (page dated 25 February 2025) reports that curated PMPs for an unnamed global CPG drinks brand in the UK, Spain and Singapore, 'customized using PHD EMEA's carefully vetted inventory "allow" lists' across unnamed SSPs, produced 'projected cost savings of up to 33%' and 'a View-Through Rate consistently above 70%, beating industry benchmarks of 60%', with no period, sample, measurer or benchmark source. | 3 | vendor assertion | supported as stated | high | [125, 210] |
| CUR-LOTAME-F08 | The PHD case is the only record that documents provenance for a named curated-deal segment: it gives the luxury-traveller and whiskey-drinker segment definition (declared frequent spirits buyers with luxury-travel behaviour or intent), its collection route ('loyalty / credit cards and online actions across publisher sites') and 'No modeling was done', and quotes the agency that the brand's 'CDAP (Consumer Data Assessment Process)' scrutinised Lotame's 'data sourcing, consent, and activation practices'. | 3 | vendor assertion | supported as stated | medium | [125] |
| CUR-LOTAME-F09 | Lotame's Services Privacy Notice (last modified 19 November 2025) documents company-wide, not per curated deal, that Lotame collects data through pixels, tags, cookies (Lotame PID, Panorama ID), SDKs, APIs and file transfers, with identifiers sourced from Lotame itself (cookie IDs), from 'Business customers, data partners' (IP address, third-party ID, email) and, for device IDs, also from 'graphing partners'; licenses television viewership data from smart-TV and streaming-device companies for CTV targeting; states in its GDPR supplement that its legal basis is consent given to its business customers and data partners, with TCF vendor ID 95; lists 'Racial and ethnic data (U.S. only)' and 'Political, religious, and philosophical beliefs (U.S. only)' among behaviour categories; and says Lotame 'is unable to capture any UOOM signal'. | 3 | vendor assertion | supported with qualification | high | [208] |
| CUR-MICROSOFTCURATE-F12 | Microsoft Curate curators select inventory with Microsoft-defined universal categories (applied from Microsoft's review of inventory content), seller categories, seller and publisher include or exclude lists, inventory type (App & Web, App Only, Web Only), allowlists and blocklists (the AppNexus Blocklist always applied), buy-side targeting including segments and geography, and viewability or completion-rate thresholds that act as hard limits on predicted rates, while ads.txt targeting is not supported for curated deal line items. | 3 | vendor assertion | supported as stated | high | [110] |
| CUR-MICROSOFTCURATE-F14 | Microsoft Curate automatically enables 'Cross-device enhanced by Microsoft', at no cost, on any curated deal that targets a segment, with opt-out only through an account representative, and the page does not describe the graph's data source or consent route. | 3 | vendor assertion | supported as stated | high | [110] |
| CUR-MICROSOFTCURATE-F16 | Adalytics reported in 2024 that Microsoft Xandr was one of four exchanges that appeared to relay www3.forbes.com inventory, which DeepSee.io classed as made-for-advertising, to buyers as the 'www.' subdomain, and that some of the mis-declared inventory transacted via apparent PMP deal IDs; the report identifies no Microsoft Curate deal and does not say whether the mis-declaration was intentional. | 3 | independent measurement | supported as stated | medium | [67] |
| CUR-PUBMATICCONNECT-F10 | PubMatic names the selection logics of its own packages without thresholds or update cadence (for example an 'OMSDK ‘Measurable’ Package', an 'Inventory Quality Score Marketplace', a 'High Video Completion Rate Package', vertical and category customisation and live-sports content packages), while the only documented thresholds and cadence come from a partner (Adloox's 'Top 20%' and 'Top 40%' attention tiers, 'updated daily'). | 3 | vendor assertion | supported as stated | medium | [228, 224, 12] |
| CUR-PUBMATICCONNECT-F12 | Connect's data sources are named by type and partly by provider (buyers' first-party data, publisher data, 'specialized partner' data, 'more than 190 premium data partners' including commerce media networks Instacart, Intuit and Western Union, Adloox via Qwarry, and Comscore-Certified Deal IDs with Adelaide's AU), but no opened record documents per-segment collection method, intended use, permissions or legal basis. | 3 | vendor assertion | supported as stated | medium | [246, 230, 228, 222, 227] |
| CUR-STAGWELLCURATE-F10 | No seller list, domain list, seller count or seller category is public for Stagwell Curate or any of its packages. The release offers 'deep, direct publisher relationships' through partners and 'enhanced transparency, measurement and accountability', and Adams told Digiday that clients 'can see our buy data; they can see how their inventory is being put together', with no documented interface or example deal. | 3 | synthesis | supported as stated | medium | [5, 71, 16] |
| CUR-PUBMATICCONNECT-F11 | PubMatic asserts composition-visibility capabilities for its authenticated Media Buyer Console (an 'Inventory Discovery Tool' covering 'the past 30 days', 'Log-Level Transaction Verification' and 'Two-Way Fee Transparency' that lets buyers 'know exactly where the inventory is coming from'), and its Deal Management agent accepts 'publisher lists', but no identified Connect deal with a seller or domain list was found in the sources opened. | 3, 6 | vendor assertion | supported as stated | medium | [222, 233] |
| CUR-B-F27 | Adalytics' Forbes report (2024) documented that ads on the 'www3.forbes.com' subdomain, which DeepSee.io classed as made-for-advertising and which ran 201 or more impressions in one slideshow session, were relayed by Microsoft Xandr, TripleLift, PubMatic and Magnite as the 'www' domain and that 'some of this mis-declared inventory was observed transacting via what appear to be Private Marketplace (PMP) deal IDs', while Forbes did not deploy IAS, DoubleVerify or Oracle Moat publisher tags on that subdomain. | 3, 7 | independent measurement | supported as stated | medium | [67] |
| CUR-B-F29 | Adalytics' pre-bid bot report documents ads served to declared data-centre bots through named PMP deal IDs on Index Exchange, Magnite, TripleLift, Yieldmo, TrustX, GumGum and Kargo (examples dated March 2022 to November 2024), on inventory where the ad code referenced DoubleVerify or IAS pre-bid avoidance segments and where the exchanges state they filter bots pre-bid with HUMAN Security; it cites HUMAN's 2019 claim that Kargo's 'curated marketplace boasts less than 0.3% of invalid traffic'. | 3, 7, 11 | independent measurement | supported as stated | medium | [69] |
| CUR-G-F07 | By 9 February 2023 GroupM said Premium Marketplace hosted 'roughly 30 active advertisers', had 'integrations with 80% of top 20 connected TV publishers' in North America (naming Fox, AMC, Univision, LG, Philo and Xumo), average win rates of 'roughly 35%' against about 15% outside it, and CPMs 'nearly one-third lower' for one client, all operator figures without method or period. | 3, 8 | vendor assertion | supported as stated | medium | [18] |
| CUR-G-F08 | The most recent WPP Media page on Premium Marketplace found is the GroupM-branded article of 9 February 2023 hosted on wppmedia.com; whether the product is still sold under WPP Media in 2026 is not stated on either page opened, so its status is coded unknown. | 3, 8, 11 | inference | supported as stated | medium | [18] |
| CUR-U-F11 | Magnite's blog of 9 September 2026 (byline 'Magnite Team') describes ClearLine Activate (CLA) as the self-serve audience layer of ClearLine, naming its data onboarding partners (Experian, LiveRamp, TransUnion, Throtle, Cadent), direct third-party data partners (Acxiom, Comscore, Captify, Lotame, Nielsen, OpenAP, Samba, TruthSet, Genius Sports, TargetSmart) and clean rooms, quotes Mike Laband, group SVP, US revenue, in the body, and carries no beta or general-availability word, named DSP, fee, scale figure or channel statement. | 3, 11 | vendor assertion | supported with qualification | medium | [108] |
| CUR-CLEARLINE-F10 | Adalytics' pre-bid bot report documents ads served to declared data-centre bots through Magnite-transacted PMP deal IDs, including 'MGNI-MD-9-5198' (Unilever Dove), 'TRD-9262-00da4' (Unilever Dove, 19 November 2024) and 'MGNI-AP-5083' (T-Mobile, 24 November 2024), and notes that Magnite is among the platforms that publicly claim pre-bid bot filtering with HUMAN Security; the examples predate ClearLine curation and none is identified as a ClearLine deal. | 3, 11 | independent measurement | supported with qualification | medium | [69] |
| CUR-INDEXMARKETPLACES-F14 | No page opened documents a buyer-facing list of the media owners, sellers or domains in a Marketplace Package: the knowledge base offers reporting to publishers, buyers, DSPs and partners and 'unified reporting and consolidated billing' without stating a seller or domain breakdown, the product page asserts 'Real-time dashboards and audit logs' and 'Total transparency', and the launch release promises media owners visibility of who buys their inventory, so composition disclosure to buyers rests on general assertions. | 3, 11 | vendor assertion | supported as stated | medium | [198, 200, 187] |
| CUR-MICROSOFTCURATE-F13 | The only buyer-facing element of a Microsoft Curate deal documented on the pages opened is the deal name (and an optional code for external buyers' deal mapping); no page documents a buyer-facing list of sellers, publishers or domains, while sellers can block curators and report on curator activity. | 3, 11 | vendor assertion | supported as stated | medium | [110, 111] |
| CUR-PUBMATICCONNECT-F18 | PubMatic's record on supply-path integrity for Connect deals is self-description only (the console 'ensures brand safety and quality control by adhering to industry standards such as sellers.json and ads.txt'), with no observed sample of a Connect deal's ads.txt, sellers.json or SupplyChain records; PubMatic's own blog says curators 'don’t sit in the media payment chain', while the console page says PubMatic 'manages the revenue share with third-party partners', so curator and data-partner payments run through PubMatic's billing rather than as SupplyChain nodes. | 3, 11 | inference | supported as stated | medium | [222, 225] |
| CUR-A-F32 | For OpenRTB auctioneers the MRC auction standard requires complete SupplyChain and SupplyChainNode objects, requires DSPs to validate ads.txt seller ids against sellers.json, to ensure the chain ends at the inventory owner and to report hop counts and mis- or undeclared chains, and requires DSP support for ads.txt 1.1 and correct MANAGERDOMAIN use. | 3, 12 | direct record | supported as stated | high | [151] |
| CUR-E-F22 | Index's Marketplaces product page offers sell-side users deal packaging 'without any added platform fees' and media owners incremental demand 'without added fees or extra work', claims 'Real-time dashboards and audit logs' and intelligence applied 'in the opening milliseconds of the auction', and states no deal type, priority, floor, exclusivity rule, fee amount or DSP name. | 3, 12 | vendor assertion | supported with qualification | high | [258] |
| CUR-INDEXMARKETPLACES-F19 | Post-purchase control on Index Marketplaces sits with the partner, the media owner and Index, not the buyer, on the pages opened: the partner chooses the fee disclosure model and a custom deal fee cannot later revert to the default, media owners can opt out of Marketplaces demand, and Index 'reserves the right to update targeting' on deals that use LiveRamp Data Marketplace segments to restrict them to the United States (a right scoped to enforcing that vendor's US-only condition); no page documents notice to the buyer, an audit right, data export or a remedy for misdescribed inventory or charges. | 3, 12 | vendor assertion | supported with qualification | medium | [109, 14, 240, 206] |
| CUR-A-F18 | The Deal Sync API 1.0 says curated deals may carry request-by-request price points for which suppliers offer bid guidance rather than floors, and its Terms object makes dealfloor guidance unless pricetype is Fixed (pricetype default 2 = second price plus; 0 dynamic, 1 first price, 3 fixed). | 4 | direct record | supported as stated | high | [96] |
| CUR-E-F02 | Microsoft Curate offers three curated-deal pricing types (Fixed: CPM with a Fixed Price; Standard: Dynamic CPM with a Floor Price; Market: Dynamic CPM using the seller's yield-management floors), lets the curator set a floor 'above, at, or even below a seller floor', and runs the Dynamic CPM deal in the seller's first- or second-price auction at the buyer's bid. | 4 | vendor assertion | supported as stated | high | [112, 110] |
| CUR-E-F16 | Every Google Ad Manager and Authorized Buyers curation page opened carries the Beta notice that the feature 'might not be available in your network', and the publisher page defers general availability to future release notes, so US availability and production use cannot be coded from these pages. | 4 | vendor assertion | supported as stated | high | [59, 24, 248] |
| CUR-E-F30 | InMobi's Buyer Hub page dates agency access to 31 August 2026 with sales-assisted onboarding, describes natural-language inventory discovery and deal building 'with traders in control of every decision', shows top publishers, eCPM and ad requests, and quotes InMobi's Kunal Nagpal that Buyer Hub was earlier opened to curators; it states no deal type, auction type, priority, floor, exclusivity rule, fee, DSP name or exchange name. | 4 | vendor assertion | unresolved | medium | [174] |
| CUR-E-F31 | Sovrn's Advertising Demand page offers 'private marketplace deals, attention deals, curated packages, and Commerce Audiences' under a 'Curated Deals' heading, reached by contact, with 'Transparent reporting' and coverage 'US / EU / CA / APAC'; buyer platforms appear only as logo image file names (Amazon Ads, BidSwitch, Basis, Google DV360, TTD, Simplifi, Beeswax, RTB House), and the page states no auction type, priority, floor, exclusivity rule or fee. | 4 | vendor assertion | unresolved | medium | [277] |
| CUR-INDEXMARKETPLACES-F04 | Index documents two auction types for Marketplace Packages, both chosen by the partner: First (the highest bid wins and pays its bid, with a floor CPM in USD of at least $0.10) and Fixed (the winner pays the negotiated fixed price); DSPs are sent the higher of the package floor and the highest publisher floor grossed up for fees (a $5.00 publisher floor with a 10% fee becomes $5.56 against a $4.00 package floor), the floor 'must be higher than a Publisher's floor setting for the deal ID to transact', and a fixed-price deal whose price is below the publisher floor plus fees is ineligible. | 4 | vendor assertion | supported as stated | high | [14, 199] |
| CUR-PUBMATICCONNECT-F09 | On floor treatment and access, PubMatic's CEO said on 7 March 2025 that PubMatic applies no different floor rules to curated data sets or inventory, that a curator wins only 'if the net CPM to the publisher is higher than the alternative', and that publishers are opted into curation 'by default' and can opt out; no source opened documents auction type, priority, exclusivity or request sampling for Connect deals, and the only priority setting found ('priority P6') is in a publisher PG example for the Deal Management agent. | 4 | vendor assertion | supported as stated | medium | [226, 233] |
| CUR-I-F02 | The liability opinion found that First Look required DFP publishers to give AdX a first right of refusal on every impression, that AdX kept that first look even when a publisher wanted to rank other exchanges first, and that AdX advertisers could win even when rival exchanges' advertisers offered more; the court held First Look anticompetitive conduct within the DFP auction. | 4, 9 | direct record | supported as stated | high | [78] |
| CUR-I-F03 | The liability opinion found that Last Look let AdX see the winning header-bidding bid, which publishers had to pass to DFP as a price floor, and outbid it in an otherwise sealed auction; by 2019 Google agreed to remove Last Look and introduced Unified Pricing Rules at the same time. | 4, 9 | direct record | supported as stated | high | [78] |
| CUR-I-F04 | The liability opinion found that sell-side dynamic revenue share let AdX vary its take rate impression by impression, keeping an overall take rate of around 20% while cutting below 20% on competitive impressions with high header-bidding offers, to exploit the Last Look advantage. | 4, 9 | direct record | supported as stated | high | [78] |
| CUR-I-F05 | The liability opinion found that Unified Pricing Rules prohibited DFP publishers from setting higher price floors for AdX than for other exchanges and for AdWords demand than for other buying tools, while still permitting higher floors on third-party exchanges than on AdX, which reduced publishers' ability to control pricing, quality and demand diversification. | 4, 9 | direct record | supported as stated | high | [78, 156] |
| CUR-I-F07 | The remedies opinion decided an injunction against tying DFP to AdX, a bar on reimplementing First Look and Last Look, and deprecation of Unified Pricing Rules for all indirect transaction types, with Google allowing publishers to set different floor prices across buying tools and exchanges and not discounting DFP fees for uniform floors. | 4, 9 | direct record | supported as stated | high | [156] |
| CUR-I-F08 | The remedies opinion grouped the accepted behavioural remedies into an injunction, interoperability (AdX bids to rival ad servers and to Prebid; DFP-Prebid integration, including publishers' choice to route programmatic guaranteed and preferred deals through Prebid), data sharing, anti-discriminatory bidding rules and buy-side rules for AdWords and DV360; where Google's implementation timeframe was shorter the court deferred to it (12 to 15 months for DFP-Prebid interoperability). | 4, 9 | direct record | supported as stated | high | [156] |
| CUR-I-F09 | The remedies opinion decided a six-year term, expected implementation within 15 months, and a judgment taking effect 60 days after entry except that work to appoint the Monitor and form the Technical Committee starts on entry; the parties were ordered to file one jointly proposed final judgment within 30 days, labelling any unresolved provision by proponent. | 4, 9 | direct record | supported as stated | high | [156, 80] |
| CUR-I-F10 | The jointly proposed final judgment of 2 October 2026 (Dkt 1861-1, 41 pages) applies solely to Indirect Demand, defined as demand through open auctions or private auctions, for open-web display inventory only, and excludes AdSense, AdMob, DV360 and any Google tool other than Google Ad Manager or Google Ads; the parties dispute whether 'AdX' means all Ad Exchange functionality (Plaintiffs) or the open-auction and private-auction functionality (Google). | 4, 9 | direct record | supported as stated | high | [79] |
| CUR-I-F11 | Under the proposed judgment the Effective Date is 60 days after entry, tolled day for day during any stay; the judgment would expire on the sixth anniversary of the Effective Date; and the bracketed disputes include 6 (Plaintiffs) versus 12 (Google) months for AdX-Prebid and AdX-rival-ad-server interoperability, Google's proviso 'Subject to the outcome of any motion to stay this Final Judgment pending appeal', who proposes the Monitor slate, and what DFP must pass to Prebid auctions (Plaintiffs: all pricing information and Google audience data; Google: DFP pricing rules, publisher audience data and deal information). | 4, 9 | direct record | supported as stated | high | [79] |
| CUR-I-F12 | The proposed judgment would bar Google from reimplementing First Look or Last Look in DFP or AdX for Indirect Demand, require Header Bidding Trafficking for all DFP publishers within 6 months, deprecate Unified Pricing Rules with no uniform-floor requirement or fee discount for uniform floors, and within 15 months provide a server-to-server DFP-Prebid connection for programmatic guaranteed and preferred deals that publishers may, but need not, use. | 4, 9 | direct record | supported as stated | high | [79] |
| CUR-I-F16 | On 27 October 2025 Judge Castel granted partial summary judgment to publisher plaintiffs in the MDL, precluding Google from relitigating the separate ad server and ad exchange markets, their worldwide scope, the unlawful DFP-AdX tie and the anticompetitive conduct of Unlawful Tying, First Look, Last Look, Dynamic Revenue Share and Unified Pricing Rules, and otherwise denied the motions; on 25 June 2026 Google and the competitor plaintiffs stipulated, and the court ordered, that the same preclusion applies in the competitor cases. | 4, 9 | direct record | supported as stated | high | [131, 132] |
| CUR-A-F28 | The MRC auction standard requires ad auctioneers to disclose the auction type and the method converting winning bids to clearing prices, and where a buying tool mixes O&O and third-party supply with different auction types or rules, to disclose that and let participants opt out of specific supply sources. | 4, 12 | direct record | supported as stated | high | [151] |
| CUR-E-F12 | Google Ad Manager lets curation partners set a floor price on a curated package, and that floor applies only where it is higher than the publisher's own floor for the impression. | 4, 12 | vendor assertion | supported as stated | high | [59, 24] |
| CUR-E-F34 | Across the 17 product-by-transaction-type rows in the access register (15 products; Google Ad Manager coded for Private Auction, Preferred Deal and Marketplace Packages), on the pages the stream opened auction type is documented for 5 rows (4 default, 1 configurable), priority for 4 (all default), floor treatment for 6 (all configurable), buyer exclusivity for 7 (3 default, 4 configurable) and request sampling for 1 (default), and 9 of the 15 products have every attribute unknown; Commerce Grid's knowledge base, which the stream did not reach, documents deprioritisation of low-value curated deals in auction processing and one DSP seat per demand deal, so with it priority is documented for 3 rows, request sampling for 2 and buyer exclusivity for 8 (4 default, 4 configurable), and the all-unknown count stays at 9. | 4, 12 | synthesis | supported with qualification | medium | [110, 25, 113] |
| CUR-D-F01 | Publicis Groupe's acquisition of LiveRamp Holdings, Inc. was announced on 17 May 2026 at $38.50 a share in cash (enterprise value $2.167 billion, equity value $2.546 billion) and completed on 2 November 2026, when MMS USA Holdings, Inc. closed the merger, LiveRamp's shares were removed from NYSE trading and LiveRamp became a wholly owned direct subsidiary of Parent. | 5 | direct record | supported as stated | high | [49, 48] |
| CUR-D-F02 | Publicis Groupe announced an agreement to acquire Lotame on 6 March 2025 (closing expected early in Q2 2025, no price disclosed) and recorded in note 3.1 of its 2025 half-year financial report the acquisition of 100% of Lotame Solutions, Inc. in April 2025 for a cash consideration of EUR 127 million, identifiable net assets of EUR 31 million and provisional goodwill of EUR 96 million (table stated in millions of euros). | 5 | direct record | supported with qualification | medium | [40, 41] |
| CUR-D-F03 | Experian announced on 4 December 2024 that it had acquired Audigent, described as a data activation, curation and identity platform with SmartPMP, ContextualPMP and CognitivePMP products, without disclosing terms; Experian's FY25 results (14 May 2025) report US$1,244m of acquisitions in the year without naming Audigent or breaking out its consideration. | 5 | direct record | supported as stated | high | [38, 39] |
| CUR-D-F04 | Samba TV announced on 31 October 2024 its acquisition of Semasio, a contextual targeting and audience data provider, on undisclosed terms, with Semasio to continue as an independently run business under a general manager; the release does not mention PMPs, deals or curation. | 5 | direct record | supported as stated | high | [36] |
| CUR-D-F05 | Scope3 announced on 23 September 2026 that it is becoming Apostra, a platform 'built on AdCP' for advertisers and their agents; the post's footer names Apostra, Inc. and links to 'Scope3 by Apostra' for sustainability, but the post itself says nothing about the sustainability business, legal structure, funding or ownership, so the event is coded as a rebrand with unknown legal form. | 5 | vendor assertion | supported with qualification | medium | [54] |
| CUR-D-F06 | Adloox changed hands twice: Scope3 announced its acquisition of Adloox on 4 November 2024 (per Adloox's own post of 5 November 2024) and Peer39 announced on 3 June 2026 that it had acquired Adloox, with the Adloox team joining Peer39; neither post discloses terms, and Peer39's post does not name the seller, so at the cutoff Adloox is coded as controlled by Peer39 rather than by Apostra (Scope3). | 5 | vendor assertion | supported with qualification | medium | [37, 50] |
| CUR-D-F07 | Equativ announced its combination with Sharethrough as a 'merger' on 12 June 2024 (combined net recurring revenue above $200M, more than 720 employees, 18 countries, no terms or closing date), called it 'Equativ's acquisition of Sharethrough in June 2024' on 9 June 2025 when the Sharethrough brand was folded into Equativ, and named Bridgepoint as its primary investor. | 5 | vendor assertion | supported as stated | medium | [32, 33] |
| CUR-D-F08 | Equativ (founded as Smart AdServer, renamed Equativ in 2022) confirmed Bridgepoint as its new principal investor on 27 February 2023 (headline: 'new majority shareholder'), stating that the United States already represented 40% of its revenues; no sellers, terms or valuation are disclosed. | 5 | vendor assertion | supported as stated | medium | [29] |
| CUR-D-F09 | Microsoft acquired Xandr, Inc. on 6 June 2022, per its 2022 annual report, which describes Xandr as a technology platform for delivering its digital advertising solutions and says Microsoft now powers 'one of the largest marketplaces for premium advertising'; the purchase consideration was not reached in the portion read. | 5 | direct record | unresolved | high | [31] |
| CUR-D-F10 | Outbrain Inc., which entered into a Share Purchase Agreement with Altice Teads S.A. on 1 August 2024, closed its acquisition of Teads on 3 February 2025 for a cash payment of $625 million (subject to adjustments) and 43.75 million shares of common stock, with two Altice Teads designees to join the board, and renamed itself Teads Holding Co. effective 6 June 2025 (Nasdaq ticker OB to TEAD on or about 10 June 2025); Altice Teads' post-closing percentage is not stated. | 5 | direct record | supported with qualification | high | [35, 42] |
| CUR-D-F11 | DoubleVerify announced on 6 August 2026 an agreement and plan of merger under which Neptune BidCo US Inc., the parent company of The Nielsen Company (US), LLC, will acquire DV for $13.60 per share in cash (enterprise value approximately $2.15 billion); stockholders adopted the agreement at a special meeting on 29 October 2026 (approximately 94% of votes cast, over 75% of outstanding shares), DV now expects closing by the first quarter of 2027 (the August release said by the end of Q4 2026), has suspended earnings calls and withdrawn guidance, and reported Q3 2026 revenue of $205.6 million (+9%); the transaction was pending at the cutoff. | 5 | direct record | supported with qualification | high | [51, 52] |
| CUR-D-F12 | Integral Ad Science entered a merger agreement on 24 September 2025 with Igloo Group Parent, Inc. and a merger sub, both affiliates of investment funds managed by Novacap Management Inc., and announced completion on 23 December 2025 in an all-cash transaction valuing IAS at approximately $1.9 billion ($10.30 a share), after which IAS common stock ceased trading on Nasdaq. | 5 | direct record | supported as stated | high | [43, 44] |
| CUR-D-F13 | Omnicom completed its acquisition of The Interpublic Group of Companies, Inc. on 26 November 2025; Interpublic shareholders received 0.344 Omnicom shares per share, legacy Omnicom holders own about 60.6% and legacy Interpublic holders about 39.4% of the combined company, whose pro forma combined revenue exceeds $25 billion, so KINESSO's client marketplaces now sit inside Omnicom. | 5 | direct record | supported with qualification | high | [47] |
| CUR-D-F14 | TripleLift announced on 29 March 2021 that a majority stake would be acquired by Vista Equity Partners, with closing anticipated in Q2 2021 after regulatory review; the announcement gives no terms, and completion is not confirmed in the source opened. | 5 | vendor assertion | supported as stated | medium | [27] |
| CUR-D-F15 | Magnite closed its acquisition of SpotX from RTL Group on 30 April 2021 for $640 million in cash and 12,374,315 Magnite shares, a total purchase price of approximately $1.14 billion. | 5 | direct record | supported as stated | high | [28] |
| CUR-D-F17 | Multilocal's 1 October 2026 OAREX arrangement is a $4 million receivables-collateralised 'ADVANCE' facility that the company says lets it grow 'without giving up equity and control'; it is financing, not a change of ownership, and the article describes no US operations. | 5 | vendor assertion | supported as stated | high | [55] |
| CUR-D-F22 | Comscore's FY2025 10-K (NASDAQ Global Select Market, SCOR) records a recapitalisation agreed on 26 September 2025 and closed on 29 December 2025 that exchanged and retired all Series B preferred stock held by Charter Communications Holding Company, LLC, Liberty Broadband Corporation and Pine Investor, LLC (a Cerberus-advised vehicle), each receiving 4,223,621 Series C preferred shares, 3,286,825 common shares and a fixed $2.0 million cash payment due in June 2028, and reduced the holders' board rights to one designee each on a seven-member board with a 16.66% voting threshold; the 10-K portion read names no 'controlled company' status and refers to Proximic by Comscore without any 'curated' or 'Comscore-Certified' wording. | 5 | direct record | supported with qualification | medium | [45] |
| CUR-U-F16 | Audigent's SmartPMP page describes 'a first-to-market, breakthrough data + inventory product' for digital media buyers, names no channel, SSP, DSP, fee or Experian ownership, gives a New York office address and a marketplace count of '1000+ Comscore CognitivePMPs', and carries no date or copyright. | 5 | vendor assertion | unresolved | medium | [193] |
| CUR-U-F20 | IAS Curation was announced on 4 November 2024 as 'the launch of IAS Curation with Google Ad Manager', a 'deal-based enrichment pathway designed to curate inventory at the source' for 'global advertisers', and Integral Ad Science is still listed among Google Ad Manager's curation partners on the help page read on 3 November 2026, so the product is coded live on Google's route; the IAS page names no channel, fee or other platform and still carries the pre-take-private listing 'Nasdaq: IAS'. | 5 | synthesis | supported as stated | medium | [58, 59] |
| CUR-U-F36 | Vendor pages opened by this verifier confirm the entity names for Experian Information Solutions, Inc., Comscore, Inc. (NASDAQ: SCOR), Adelaide Metrics Inc., Integral Ad Science, Inc., Stagwell (NASDAQ:STGW) and WPP Media Limited, and Sovrn's footer reads 'Sovrn, Inc.' rather than the register's 'Sovrn Holdings, Inc.'; the PubMatic, Magnite, Nexxen, Verve, Lotame and Roku copyright lines rest on U's read alone; no page states a headquarters beyond an office address (TripleLift, Audigent, Onetag) or a dateline. | 5 | synthesis | supported with qualification | medium | [277, 130, 5, 13] |
| CUR-B-F16 | ANA fee components in the series: buy-side transaction costs were 16.8% of ad spending in the April to September 2024 data (DSP platform 8.6%, data and additional costs 8.2%); private-marketplace DSP platform costs were '1.1 percent lower' than open-marketplace costs (the report's wording, read here as 1.1 points); the December 2024 report put average CTV DSP fees at '8.78 percent, with little difference between marketplace type', a figure the Q1 2025 report restates as a median of $8.78 falling to $5.31; and in Q4 2025 median SSP platform costs fell to 12.7%. | 6 | independent measurement | supported with qualification | medium | [102, 100] |
| CUR-H-F04 | The ANA benchmark's Q1 2025 report tabulates its transaction-cost series as medians across contributing advertisers: DSP platform costs 8.0% (2023 column), 8.6% (April to September 2024) and 7.9% (November 2024 to March 2025); DSP data costs 6.0%, 4.6% and 1.0%; DSP additional costs 2.0%, 3.6% and 4.5%; SSP platform costs 13.0%, 13.0% and 13.7%, with the top of the SSP quartile range at 16.9% in Q1 2025; the 2023 column's values coincide exactly with the 2023 Complete Report's sequential waterfall averages ($80, $60, $20 and $130 per $1,000), and that report describes its cost figures as averages and publishes no medians, so the 2023 'medians' are to be labelled as the Q1 2025 report's re-tabulation of the 2023 study rather than as medians the 2023 study itself reported. | 6 | independent measurement | supported with qualification | high | [63] |
| CUR-H-F05 | On the ANA benchmark's sequential cost waterfall (averages across advertisers), total transaction costs were 29% in the 2023 study, 29.8% in the report covering April to September 2024, 26.1% for November 2024 to March 2025 (seller revenue 73.9%), 23.7% for Q4 2025 and 25.9% for Q1 2026, while TrueAdSpend moved 36 cents (2023), 43.9 (2024 report), 41.0 (Q1 2025), 36.3 (Q4 2025) and 43.3 (Q1 2026). | 6 | independent measurement | supported as stated | high | [63, 99] |
| CUR-H-F08 | The ANA's Q2 2025 release (14 August 2025) reports 'PMP CPMs averaging $7.15 vs. $4.41 for OMP', a TrueCPM Index of 36.5% (from 37.8%), PMP share of spend 'surged from 64.5% to 87.8%' and a median SSP count down from 19 to 17, and carries no DSP, SSP or curation fee figure; that $7.15 and $4.41 are medians across the 21 contributing advertisers (aggregate $7.43 against $3.64) rests on the brief's fact-check of the Q2 2025 report, which this stream did not open and which should be cited directly. | 6 | independent measurement | supported with qualification | medium | [66] |
| CUR-H-F09 | The ISBA/PwC Programmatic Supply Chain Transparency Study (6 May 2020; 15 advertisers, 8 agencies, 5 DSPs, 6 SSPs and 12 publishers; about GBP 0.1 billion of UK programmatic spend; 290 supply chains and 31 million impressions matched end to end) found publishers received 51% of advertiser spend on average and that 15% of advertiser spend was an 'unknown delta', about one-third of supply-chain costs, which the CMA restates as 49% absorbed by intermediaries; ISBA and PwC's second study (18 January 2023) restated those 2020 figures to 57% publisher net revenue and a 17% unknown delta so as to exclude agency fees, verification tools and ad serving, the second study's scope, so the 51% and the CMA's 65% are not on one basis. | 6 | independent measurement | supported with qualification | high | [105, 288] |
| CUR-H-F10 | ISBA and PwC's second study (summary published 18 January 2023; 11 advertisers, 7 agencies, 6 DSPs, 6 SSPs and 10 publishers; 1.3 billion impressions analysed for 1 September to 31 October 2022; a nine-month study) reported an impression match rate of 58% (12% in 2020), an unknown delta of 3% (15% in 2020, restated to 17% on the second study's basis; under 1% for private marketplace deals) and a publisher share of advertiser spend of 65% against a restated 57% for 2020, so the knowledge page's 'risen by 8%' is 8 percentage points on a basis that excludes agency fees, verification tools and ad serving; the summary says the results are not representative of the broader programmatic ecosystem. | 6 | independent measurement | supported with qualification | medium | [283] |
| CUR-H-F13 | Magnite's FY2025 Form 10-K defines its take rate as 'our fee as a percentage of advertising spend', states that it 'varies by publisher and transaction type', 'tends to be lower for reserve auctions compared to auctions with multiple bidders, and tends to be lower on CTV transactions compared to other channels', and is higher for its managed service business, that Magnite 'may also negotiate lower take rates with large sellers', and that fee concessions 'may be more prevalent among CTV publishers where inventory is scarce'; it reports revenue of $713,953 thousand (2025) and $668,170 thousand (2024), with 10%, 14% and 18% of revenue reported on a gross basis in 2025, 2024 and 2023, and discloses no numeric take rate, no total ad spend and no fee for the curation inside ClearLine in the portions read. | 6 | direct record | supported with qualification | high | [212] |
| CUR-H-F15 | The Trade Desk's FY2025 Form 10-K describes its platform fee as 'generally based on a percentage of our clients' total spend on our platform' and reports gross spend of $13,394,683 thousand (2025) and $12,040,872 thousand (2024) against revenue of $2,896,284 thousand and $2,444,831 thousand, so revenue was 21.6% of gross spend in 2025 and 20.3% in 2024 (a derived estimate of the blended platform revenue share; the filing does not call it a take rate and revenue includes value-added services and data). | 6 | direct record | supported as stated | high | [284] |
| CUR-H-F17 | The CMA's July 2020 market study (Appendix R, UK intermediary data for 2019) estimated that publishers receive around 65% of initial advertiser spend (an 'adtech take' of about 35p per pound), a weighted average DSP direct take rate of 14% (range 5% to 42%; self-service platforms 4% to 16%, service-heavy platforms 20% to 42%) plus 2% in third-party fees via DSPs, a weighted average SSP direct take rate of 16% (range 8% to 25%), Google DSP take rates of [5-15%] for DV360 and [10-20%] for Google Ads, an AdX SSP take rate of [10-20%], about 30% when Google Ads and AdX are both used, and that most SSPs retained a higher revenue share on open auctions than on private marketplaces. | 6 | direct record | supported as stated | high | [288] |
| CUR-H-F20 | Microsoft Curate's documentation (ms.date 21 October 2025) states that curators take no margin by default and must set a non-zero margin as a percentage of the buyer's bid or a fixed CPM; that the buyer's bid has buyer fees removed (reported as Curator Revenue), then curator margin and Curator Tech Fees are deducted, and the remainder (Curator Net Media Cost) goes to the seller auction where seller fees are deducted; that the platform enforces an unstated maximum margin rate; and that Microsoft's own charges to curators are the Broker Auction Service Charge (a percentage of the buyer's media cost after buy-side charges, as a fee or a deduction) or the Margin Auction Service Charge (a percentage of the curator's gross margin; example: margin $0.18, MASC 30%, fee $0.06), with the seller's charge (SASC) quoted as a percentage of the buyer's media cost (example 10%). | 6 | vendor assertion | supported as stated | high | [115, 216] |
| CUR-H-F27 | The Ozone Project's blog post of 10 July 2025 asserts that '80% of curation revenue comes from just 10% of curators' and that '95% of publishers are unclear on curation fees, with average curator margins at 44%', 'Based on our extensive research', with no sample, period, geography, method or definition of margin. | 6 | vendor assertion | supported as stated | low | [289] |
| CUR-H-F29 | The 'tech tax' register for curation in the sources screened comprises: AdExchanger's framing of curation fees as 'the same old ad tech tax' (1 May 2025) resting on Jounce's finding that half of curated deal IDs carry no curator fee; the CMA's 2020 note that intermediary fees 'are sometimes referred to as the "adtech tax" or "take rate"' and Arete Research's submission that DSPs 'have 15-20% take rates (depending on "data fees"), with another ~20% taken by SSPs/ad networks'; and vendor rebuttals without schedules (Index: 'no hidden tolls or platform taxes'; TripleLift: 'No TripleLift curation fees'; Google: 'a single fee that is fully transparent to the buyer'); no SSP in the sources screened published a curated-deal fee schedule in reply. | 6 | synthesis | supported as stated | medium | [64, 288, 188, 117, 290] |
| CUR-AUDIGENT-F10 | Audigent discloses charge types for its deals but no amount, rate, base, payer, collector or period. The type sentence is 'PMP pricing is determined by bid floors and curation fees only'; the general assertions are 'Removing detrimental bid shading practices', 'more fairly priced PMP alternatives', a 'single price' with 'no additional data management platform (DMP) fees' in DV360 and 'reduces overall buyer costs' on Amazon DSP. | 6 | vendor assertion | supported as stated | high | [192, 193, 190, 191] |
| CUR-INDEXMARKETPLACES-F08 | Data-vendor charges on Index Marketplaces are a separate fee line: each Data Vendor 'may collect a fee called a Marketplace Vendor Fee', set as a fixed CPM or a percentage of media with an optional up-to CPM cap, only the highest fee per vendor applies when several of its segments activate a deal, fees from different vendors add (in Index's example $1.00 and $0.40 on a $10.00 spend), vendor fees are included in the total Marketplace fees, a partner sees the fee when selecting segments and is never billed for its own segments. | 6 | vendor assertion | supported as stated | high | [197] |
| CUR-INDEXMARKETPLACES-F09 | A Marketplace operates under a disclosed or undisclosed fee model chosen by the partner: under the disclosed model the Marketplace Owner Fee and Vendor Fee are 'Optionally disclosed', under the undisclosed model media spend is 'Always disclosed for buyers, Undisclosed for Media Owners' and the fee lines are undisclosed, so buyer visibility of the curation and data fees is partner-controlled, not a default. | 6 | vendor assertion | supported as stated | high | [109] |
| CUR-LOTAME-F12 | No record opened discloses any charge in the deal path of a Lotame curated marketplace with an amount or rate, base, payer, collector and period: Lotame offers only the 2024 general assertion that the offering 'Cuts down operational workload and data/tech fees', the one documented exchange route lets a curator hide its package fee from the buyer, and the SSP and DSP take rates for Lotame's deals are not stated anywhere opened. | 6 | synthesis | supported as stated | medium | [209, 24, 188] |
| CUR-MICROSOFTCURATE-F10 | Microsoft's MASC worked example does not reproduce: the Charges for Sellers page states that a 30% MASC on a $0.18 margin is $0.06, whereas 30% of $0.18 is $0.054 (about $0.05); $0.06 would correspond to a rate of about 33%. | 6 | inference | supported as stated | high | [216] |
| CUR-PUBMATICCONNECT-F07 | PubMatic's CEO told AdExchanger on 7 March 2025 that a curation transaction carries PubMatic's 'core SSP fee' plus 'a data fee in the curation transaction based on the curator's fee', 'negotiated partner by partner', that the curator sets its own fee, and that the buyer's CPM 'is declared in the SSP when the curator sets up the deal'; no rate, schedule or payer statement for either PubMatic charge is published in the sources opened, and the 10-K gives only the corporate basis ('fees charged to our publishers, which are generally a percentage of the value'). | 6 | vendor assertion | supported as stated | medium | [226, 222, 231] |
| CUR-PUBMATICCONNECT-F08 | The only priced input documented for a curated deal on PubMatic is a third-party data fee: Adloox's learning-center page of 17 July 2026 lists four Qwarry-hosted Adloox attention segments ('Desktop Top 20%', 'Desktop Top 40%', 'Mobile Top 20%', 'Mobile Top 40%') selectable 'while building curated deals on PubMatic' at '$0.15/CPM', without naming the currency or stating who pays it, who collects it or whether it is included in PubMatic's fee. | 6 | vendor assertion | supported as stated | high | [12] |
| CUR-STAGWELLCURATE-F09 | No source opened discloses any charge in Stagwell Curate's deal path (curation or platform fee, data fee, SSP take rate, DSP take rate). The operator's statements are general: 'pre‑negotiated PMP deals and exclusive pricing' (July 2026), 'material reduction in ad tech fees' (April 2026) and, in Digiday's paraphrase, Adams's aim to 'reduce tech fees'. Company-wide, Stagwell's FY2025 10-K says its primary source of revenue is 'agency arrangements in the form of fees for services performed, commissions, and from performance incentives or bonuses'. | 6 | synthesis | supported with qualification | high | [5, 72, 71, 236] |
| CUR-H-F16 | The liability opinion of 17 April 2025 in United States v. Google records that industry participants call an ad-tech tool's percentage fees its 'take rate', that 'between $20 to $40 of every $100 spent on digital advertising goes to the ad tech companies' (citing an exhibit and Google testimony), that AdX maintained 'an overall take rate of around 20%' while varying it between 15% or lower and 25% or higher per impression under sell-side dynamic revenue share, that rival exchanges 'often charged closer to 10%' and exchanges generally take 'between 10% and 20% of revenue from the winning bid', that Index Exchange found 'nominal, at best' win-rate effects from experiments including setting its take rate to zero, and, in a citation parenthetical summarising defence-expert testimony and exhibits, that DFP fees equate to a take rate of '1.7 percent to 1.3 percent' against around 20% 'for an auction run on AdX'. | 6, 9 | direct record | supported with qualification | high | [78] |
| CUR-H-F19 | Google's curation documentation (Beta) states that curators enter a package or data-segment fee 'as percentage of media or CPM', that package fees can be marked visible or hidden from buyers (Authorized Buyers Help), that 'Curation partners define the fee of the data segment or curated package. It is a single fee that is fully transparent to the buyer', that 'Google then handles the fee payment to the curation partner' and the fee is 'taken out of the bid price before it enters the auction' (Ad Manager Help); neither page states Google's own revenue share on a curated package, who pays, or how the fee appears in buyer reporting, and curated packages can bundle Private Auctions and Preferred Deals. | 6, 9 | vendor assertion | supported as stated | high | [24, 290] |
| CUR-CLEARLINE-F07 | No source opened discloses any charge for ClearLine or its curation (curation fee, data fee or Magnite's take on curated deals); at company level the FY2025 10-K defines the take rate as 'our fee as a percentage of advertising spend', says it 'varies by publisher and transaction type', 'tends to be lower for reserve auctions compared to auctions with multiple bidders' and 'on CTV transactions compared to other channels' and is higher for managed service, and reports revenue generally as a percentage of ad spend, 10% of it gross in 2025, without a numeric rate. | 6, 11 | direct record | supported as stated | high | [212, 62] |
| CUR-INDEXMARKETPLACES-F07 | Index's knowledge base (updated 9 September 2026) documents that a Marketplace Partner sets its fee as a percentage of media spend or a fixed CPM (USD only), that Index's revenue share is negotiated with each media owner, that fees are removed from the bid before the auction, and illustrates a $10 gross bid less a 5% partner fee and a 15% Index share paying the media owner $8.08 (partner $0.50, Index $1.42), or $7.65 with a $1.00 CPM partner fee (partner $1.00, Index $1.35). | 6, 11 | vendor assertion | supported as stated | high | [109] |
| CUR-MICROSOFTCURATE-F09 | Microsoft documents the Curate fee stack by type, base, payer, collector and order of deduction but publishes no amounts: buyer fees are removed first (leaving Curator Revenue); the curator margin (none by default; a percentage of Curator Revenue or a fixed CPM, capped by an unstated maximum rate) and Microsoft's Curator Tech Fees (BrASC on the buyer's media cost after buy-side charges, as a fee or deduction per contract, or MASC on the curator's gross margin) are then deducted; and the remainder goes to the seller auction, where the seller auction service charge (SASC) is deducted; all curation charges apply only on a win. | 6, 11 | vendor assertion | supported as stated | high | [115, 216, 111] |
| CUR-E-F36 | For the curation-fee series, four of fifteen cohort A products (Microsoft Curate, Google Ad Manager Curation, Index Marketplaces and, through its knowledge base rather than its product page, Commerce Grid) document the fee type, base, collector and payment route without amounts; two (TripleLift, and Index's marketplaces page addressed to the sell side) make a general assertion of no fees without a schedule; nine disclose nothing on the pages opened or could not be opened; and buyer visibility of the fee is curator-controlled on Google and Index and unstated on Microsoft and Commerce Grid. | 6, 12 | synthesis | supported with qualification | medium | [115, 24, 109, 117, 266] |
| CUR-H-F11 | Jounce Media's April 2025 Supply Path Benchmarking report (client-only; 1.5 billion bid requests from 44 exchanges, analysed in March 2025 per AdExchanger) found, as reported by Playwire (10 April 2025) and AdExchanger (1 May 2025), that 50% of curated deal IDs (Playwire also labels the shares 'of bid requests') carried no fee beyond the exchange's standard open-auction take, 34% a margin-based fee with a median of 14% (Playwire: about 36% of those between 5% and 10% and 9% below 5%; AdExchanger: 'nearly 40%' between 5% and 10%), 10% a fixed CPM fee with a median of $0.48 and 7% a static price, with a median effective margin of 30% and a median effective CPM of $0.51 per Playwire; Jounce's public State of the Open Internet 2025 (pp. 37-38) treats exchange take rates only as mark-ups embedded in declared floors (illustrative 10%, 15% and 20% on a $0.60 floor) and refers to a client-portal floor-price report of February 2024, and neither it nor Jounce's home page carries a curation-fee statement. | 6, 12 | independent measurement | supported with qualification | medium | [9, 64, 281] |
| CUR-H-F12 | AdExchanger's 1 May 2025 account of the Jounce report describes the payment mechanics of sell-side curation fees: curators and SSPs 'factor their curation fees into the floor price passed to DSPs', 'an SSP accrues and pays the curator's fee, rather than the DSP', publishers 'still end up with the same cut they'd get if the deal was curated via a DSP', and for a static-price deal Jounce's Chris Kane says the split of the gap between the declared price and the clearing price is unknown: 'We don't know how much of it is retained by the exchange, versus data providers, versus curators, versus passed on to the publisher'. | 6, 12 | synthesis | supported as stated | medium | [64] |
| CUR-H-F26 | TripleLift's survey release of 17 June 2026 ('223 advertisers across the U.S and U.K') reports that '60% of buyers said they do not know what they are paying for curation' and '45% of buyers described themselves as neutral regarding fee transparency from current providers', with its product officer calling the 60% 'a fundamental accountability failure'; the release does not say who fielded the survey or when. | 6, 12 | commissioned measurement | supported as stated | low | [291] |
| CUR-H-F23 | Adloox's learning-center page (published 17 July 2026) documents four Adloox attention segments hosted by Qwarry in PubMatic's Media Console ('Create Auction Package Deal') at '$0.15/CPM', the only disclosed data-fee rate found for a curated-deal input in this stream; the page does not say who pays the $0.15 or who collects it. | 6, appendix | vendor assertion | supported as stated | medium | [12] |
| CUR-B-F04 | Made-for-advertising sites took 21% of the 2023 study's impressions and 15% of spend on a web-domain subset, carried media CPMs 25% lower than non-MFA sites and passed standard measurability, viewability, IVT and brand-safety screens; the Complete Report re-based the annual MFA estimate to about $10.1 billion (15% of $67.3 billion, the $88 billion global open-web estimate less $20.3 billion of CTV). | 7 | independent measurement | supported as stated | high | [26] |
| CUR-B-F17 | The ANA benchmark's MFA series by marketplace shows higher MFA exposure in the open marketplace than in PMPs (Q4 2025: OMP 0.98%, PMP 0.23%; median 0.45% overall), but the Q4 2025 report states that 'no buying path is immune without continuous auditing' and that actively managed OMP 'can deliver comparable TrueCPM efficiency to PMPs'. | 7 | independent measurement | supported as stated | medium | [100] |
| CUR-B-F22 | Jounce's December 2024 SPO report, as cited in the 2025 edition, found that 18 of the 20 largest MFA publishers contracted by more than 95% in the 18 months after the ANA's 2023 study, while one MFA operator sustains itself by launching a new site every three to four weeks to evade block lists. | 7 | independent measurement | supported as stated | medium | [101] |
| CUR-B-F28 | Adalytics' January 2024 observations on a sample of 22 MFA-classified websites found ads from 16 of the 17 publicly disclosed participants in the ANA's 2023 study, and stated that 'some ad exchanges which claim to work with vendors such as Jounce Media to eliminate Made for Advertising inventory from Private Marketplace (PMP) deals appear to continue to serve brands' ads on MFA sites specifically via PMP deals', naming Yieldmo, PubMatic, TripleLift, OpenX and ShareThrough in section headings; the report says it 'is not meant to be construed as a comprehensive or quantitative audit'. | 7 | independent measurement | supported as stated | medium | [68] |
| CUR-B-F30 | Balocco's 2023 Erasmus thesis (abstract) finds that premium contracts (PMPs) 'lead to adverse selection in the open market (RTB)' 'while not increasing advertisers' utility' and are 'not the effective signaling device they are often assumed to be', and argues that publishers' replacement of the waterfall with header bidding 'eliminates all incentives for ad exchanges to keep sell-side fees low', which the abstract links to the rise in sell-side fees 'commonly referred to as the Ad Tech-Tax'; the full text (a PDF listed on the record) returned HTTP 403 behind a Cloudflare challenge. | 7 | independent measurement | supported with qualification | low | [118] |
| CUR-C-F01 | PubMatic's release of 11 June 2026 (Paris) reports that a programmatic video campaign for an unnamed French mass-market retailer, bought by Agence 79 (Havas Media Network) through PubMatic's curation, achieved a click-through rate five times that of 'the same campaign on the open market', a video completion rate 16.7% above target and viewability 12.5% above an 80% target, with every figure sourced to 'Agence 79 internal data' and no period, impression volume, measurement vendor or method published. | 7 | vendor assertion | supported as stated | high | [119, 292] |
| CUR-C-F02 | OpenX and Chalice AI's release of 16 June 2026 claims that custom AI models running as 'always-on bidding agents within OpenX's infrastructure' delivered, for a Hyundai video campaign, a 20% reduction in cost per high-value action and a 67% decrease in CPMs compared with 'the next-best alternative' (headline: 'second-best-on-plan'), described as 'Early results' with no period, geography, sample, measurement party or method. | 7 | vendor assertion | supported as stated | high | [120] |
| CUR-C-F03 | Microsoft's Curate product page presents three figures under 'Curate by the numbers', '74% Lower CPA', '25% Lower CPM' and '+12% Viewability rate', each footnoted only as '2023 Microsoft Curate first-party data', with no comparator, base, geography, sample or method, while the Microsoft Learn documentation page for Curate (ms.date 21 October 2025) makes no quality or performance claim. | 7 | vendor assertion | supported as stated | high | [121, 56] |
| CUR-C-F04 | Audigent's case page states that, 'Through rigorous testing', SmartPMPs drove 12% to 67% lower CPC, 1.6x to 2.5x higher CTR and 12.5% to 42% higher VCR than 'traditional DSP-level data to open exchange inventory', with no date, advertiser, geography, sample, period, measurer or method; its Alcon case reports a 27% beat of Alcon's target KPI and an 87% fall in cost per online eye exam 'from the start of the optimizations', with conversions tracked by Alcon's floodlight tag. | 7 | vendor assertion | supported as stated | high | [122, 123] |
| CUR-C-F06 | Lotame's launch release of 14 August 2024 for Lotame Curation (available in North America, EMEA and APAC) reports that 'During early trials of Lotame Curation, results showed 20% better campaign efficiencies, improved video completion rates by 15%, and 21% higher viewability', with no comparator, sample, period, geography or measurer (interest mark: the author's former employer). | 7 | vendor assertion | supported as stated | high | [124] |
| CUR-C-F07 | Lotame's undated PHD EMEA case study (hosted under '2025_Case_Studies') reports that curated marketplaces for an unnamed global CPG drinks brand in the UK, Spain and Singapore, customised with PHD EMEA's vetted inventory allow-lists and run across unnamed SSPs, produced 'projected cost savings of up to 33%' and 'a View-Through Rate consistently above 70%, beating industry benchmarks of 60%', with no period, sample, measurer or source for the benchmark (interest mark: the author's former employer). | 7 | vendor assertion | supported as stated | high | [125] |
| CUR-C-F08 | Adelaide's release of 7 March 2023 launching High Attention Markets with Google Display & Video 360, Xandr, Magnite, PubMatic, Teads, Sovrn, TRUSTX and Sharethrough asserts that 'high-AU media delivers 20-80% better performance than standard viewable media', that a meta-analysis of nearly 40 case studies found 'an average of 31% upper-funnel and 56% lower-funnel lift compared to traditional metrics', and that Colgate's first campaigns on PubMatic high-attention auction packages 'drove a 40% increase in media quality versus standard media', none with a measurement party, period or method. | 7 | vendor assertion | supported as stated | high | [126] |
| CUR-C-F09 | Teads' release of 17 August 2023 (hosted by Adelaide) says its high-attention omnichannel PMPs, bought through Google Display & Video 360, The Trade Desk and Yahoo, include placements with 'an average attention score 31% higher than Adelaide's Teads AU benchmarks', and lists outcome lifts for five unnamed advertisers measured by Adelaide and unnamed 'third-party brand measurement partners' (12.5% higher unaided awareness and 32.7% higher familiarity; 13.15% higher favorability; 41.6% higher store visitation; 455.6% higher site visits and 110.8% higher checkout actions; 8x higher in-store visits and a 74% higher conversion rate) alongside three projections ('would improve overall campaign performance by 31%'; '54.5% higher Unaided Awareness' at a 30 AU threshold; 'could improve overall campaign performance by 67%'), with no periods, samples or baselines. | 7 | vendor assertion | supported with qualification | high | [127] |
| CUR-C-F10 | Yieldmo's YMax page claims '+81% CTR and -65% CPC' against IAB contextual targeting 'in head-to-head testing on identical inventory' and '+33% CTR and +91% incremental reach' for Companion Audiences against ID-based targeting 'In testing', and says YMax 'curates inventory in real-time using AI and optimization to enhance viewability, completion, and engagement' without any viewability figure; no period, sample, measurer or published method accompanies either test. | 7 | vendor assertion | supported as stated | high | [11] |
| CUR-C-F11 | Seedtag's NeuroX Curation page carries one figure, '3.5x higher neural engagement' than non-contextual ads, a vendor-defined neuroscience metric with no period, sample or measurer, plus an unquantified claim of 'Lower effective CPMs'; it publishes no attention, viewability or MFA figure, and its 'Success Stories' link is marked Coming Soon. | 7 | vendor assertion | supported as stated | high | [128] |
| CUR-C-F12 | Multilocal's homepage asserts ranges and single figures with no comparator (except carbon), period, measurer or method: '20-40% increase in performance', 'Expect a 30–50% boost', '24–28% boost in yield', '40%+ Lift in performance', '75% Avg. reduction in cost-effeciency' (sic), '75% Higher CTR than projected', '38% Increase in Video Completion Rate (VCR)' and 'Cut your campaign's carbon footprint by 40–60% compared to traditional programmatic ads', while its hosted case studies (Maglr) returned 'Account expired' on 3 November 2026 so none of the case figures surfaced in search could be opened. | 7 | vendor assertion | supported with qualification | medium | [129, 293] |
| CUR-C-F13 | Equativ's Maestro by Equativ curation page publishes operational and scale figures ('Cuts planning time by up to 40%'; '500+' self-serve clients, '30K+' deals created and '50 Billion' impressions delivered, unperiodised), a Forrester SSP Wave (Q4 2024) rating of 5/5 on the Curation criterion, and an Omnicom Media Group testimonial of 'better CTR, increased views and broader reach' without figures, but no viewability, MFA, IVT, suitability or attention figure for a curated deal. | 7 | vendor assertion | supported with qualification | high | [15] |
| CUR-C-F14 | TripleLift's TL Direct general-availability release of 10 June 2026 quantifies the platform's economics from a Forrester Total Economic Impact study ('save an average of 40 hours per campaign'; '$357,000 in avoided labor costs over three years'; 'campaigns deliver a $3.70 return on ad spend (ROAS)'; '$1.3 million in incremental profit from increased advertising value over three years') and characterises beta results without numbers ('The Beta results are impressive'), publishing no viewability, MFA, IVT or attention figure for TL Direct curated deals. | 7 | commissioned measurement | supported as stated | medium | [130] |
| CUR-C-F15 | Index Exchange's release of 7 January 2026 adding Adelaide AU to Index Marketplaces ('without added fees or custom integrations') and Comscore's release of 10 June 2025 on Comscore-Certified Deal IDs with AU on PubMatic contain no quantified deal-quality claim: Index's only figure ('testing indicating improvements in accuracy by 25%') concerns GPID support, AU's predictive claim carries no figure or method, and Comscore's release is wholly qualitative. | 7 | vendor assertion | supported as stated | high | [204, 227] |
| CUR-C-F16 | Six register products publish no quantified quality or performance claim in the vendor sources opened on 3 November 2026: Stagwell Curate (release of 16 July 2026: 'pre-vetted inventory', Sincera-powered models, no figure), IAS Curation (release of 4 November 2024: 'MFA filtration available before it reaches a bidder', no MFA share), Magnite ClearLine (release of 1 October 2025: no result; curation 'Coming soon in beta'), Nexxen Curated Marketplace (product page and GA release of 1 October 2025: none), FreeWheel Curation Hub (page: none) and Experian Curated Deals (page: 'more predictable and measurable performance' against 'standard PMPs', no figure); Google Ad Manager Curation's launch post of 4 November 2024 and Microsoft's Learn documentation likewise publish none. | 7 | synthesis | supported as stated | high | [5, 58, 107, 254, 189, 260, 255, 196] |
| CUR-C-F18 | Of the quality dimensions this paper registers, viewability is the only one with a quantified vendor claim attached to a curated deal product in the sources opened (PubMatic: 12.5% above an 80% target; Microsoft Curate: '+12% Viewability rate'; Lotame: '21% higher viewability'), none with a stated baseline rate; no source opened reports a made-for-advertising share, an invalid-traffic rate, a suitability rate or a carbon methodology for a named curated deal, with MFA appearing only as a filter description (IAS) and carbon only as an unmethoded range (Multilocal). | 7 | synthesis | supported as stated | high | [119, 121, 124, 58, 129] |
| CUR-F-F11 | Lotame's launch release of 14 August 2024 claims from 'early trials' '20% better campaign efficiencies, improved video completion rates by 15%, and 21% higher viewability' with no method, sample, period or advertiser, describes the data as 'cookieless publisher data', and asserts that the offering 'Cuts down operational workload and data/tech fees' without a schedule; the current page adds 'No MFAs. No waste.' and 'Higher-than-your-DSP match rates' without figures. | 7 | vendor assertion | supported as stated | high | [209, 13] |
| CUR-F-F13 | Multilocal's undated homepage claims 'a 20-40% increase in performance', a '30–50% boost' for advertisers and agencies, a '24–28% boost in yield' for media owners, a '40–60%' carbon-footprint cut 'compared to traditional programmatic ads' and a stat block of 75% (higher CTR than projected), 75% (average reduction in cost-efficiency), 40%+ (lift in performance) and 38% (VCR increase), with no source, sample or method, and describes its data as a Media Quality Index 'covering 35,000+ domains and apps' and '62,000 audience data points' without stating where the audience data comes from. | 7 | vendor assertion | supported with qualification | high | [129] |
| CUR-F-F17 | Samba's homepage (page metadata 2 November 2026) describes its Audiences product as 'exclusive segments and PMPs built from Samba's deterministic cross-screen signals', while the Audiences page names no PMP, deal ID, exchange, DSP, fee or collection method, so no documented curated-deal product exists on the site beyond that line. | 7 | vendor assertion | supported as stated | high | [57, 268] |
| CUR-F-F28 | Every quality or performance figure found on the cohort B and candidate pages (at least 25 numeric claims, not 17) is a vendor or partner assertion without a published method on the page; the only claim with a stated comparator is Audi's '60% increase in cumulative conversion rate ... versus standard bidding tactics' on Adelaide's page, which names no method; and this review found no independent measurement with a published method of any of the twelve products' deals against a pre-curation comparator on the vendor pages and searches screened by 3 November 2026. | 7 | synthesis | supported with qualification | medium | [60, 209, 129, 70] |
| CUR-AUDIGENT-F07 | Every quantified performance claim for Audigent's deals in the sources opened is a vendor assertion on an outcome metric (CPC, CTR, VCR, advertiser KPI, share, revenue) without a stated sample, period, geography or independent measurer. These are '12% - 67% lower CPC', '1.6x - 2.5x higher CTR' and '12.5% - 42% higher VCR' against 'traditional DSP-level data to open exchange inventory', and Alcon's '27%' KPI beat and '87%' fall in cost per online eye exam. None addresses a registered quality dimension (MFA, viewability, IVT, suitability, attention, carbon). | 7 | vendor assertion | supported as stated | high | [122, 123, 187] |
| CUR-INDEXMARKETPLACES-F15 | Every performance figure found for Index Marketplaces is Index's own case-study or release claim without a named measurer or method: conversion rate up 56% and cost per conversion down 26% for a US financial-services brand (August to November 2024), cost per booking down 41% and bookings up 14% for a hotel brand in Europe over two months, 9x brand preference on streaming TV for an outdoor lifestyle brand (a case whose body names Index Marketplaces), 84% of impressions in live College Football Playoff games, CPMs 7% below an undefined benchmark and 0% of impressions outside college-football content for dentsu X and Dos Equis (27 April 2026), and a 25% GPID accuracy improvement; none measures MFA share, viewability, invalid traffic, suitability or attention for a Marketplace deal with a stated method. | 7 | vendor assertion | supported with qualification | medium | [201, 202, 203, 204] |
| CUR-MICROSOFTCURATE-F15 | Microsoft's quantified quality and performance claims for Curate are unmethoded vendor assertions: '74%' 'Lower CPA', '25%' 'Lower CPM' and '+12%' 'Viewability rate' footnoted '2023 Microsoft Curate first-party data', and an expected average view or completion rate '10-15% higher than the threshold you set'. | 7 | vendor assertion | supported as stated | high | [121, 110] |
| CUR-PUBMATICCONNECT-F13 | Every quantified quality or performance claim found for PubMatic curation is a vendor or agency assertion without published method: the Agence 79 campaign (CTR five times 'the same campaign on the open market', VCR +16.7% and viewability +12.5% against targets, 'Source: Agence 79 internal data'), the March 2025 release's 25%+ eCPM and 37% CPM claims, a 2022 Goodway Group case for the Consumer Experience Auction Package built with Jounce Media (180%+ win rates, 24% viewability over an undefined benchmark, 7-14% more cost-efficient CPMs than other deals, audibility 'as reported by Moat', Goodway Group internal data, 1 February to 31 May 2022), and a 2022 Lotame identity case on the Connect for Buyers page (interest mark); none measures a registered quality dimension against a pre-curation comparator with a published design, so H3 stays unresolved on this product's record. | 7 | vendor assertion | supported with qualification | high | [119, 230, 246] |
| CUR-A-F22 | MRC's Viewable Ad Impression Measurement Guidelines v2.0 (desktop, 18 August 2015) count a display impression as viewable at 50% of pixels on an in-focus tab for one continuous second post-render, and a video impression at 50% of the ad's pixels for two continuous seconds of play. | 7, 12 | direct record | supported as stated | high | [146] |
| CUR-A-F24 | MRC's Invalid Traffic Detection and Filtration Standards Addendum (IVT 2.0), issued as a June 2020 update of the 27 October 2015 version 1.0, divides invalid traffic into General Invalid Traffic (list- and rule-based categories) and Sophisticated Invalid Traffic, 'more difficult to detect situations that require advanced analytics, multi-point corroboration/coordination, significant human intervention, etc.'. | 7, 12 | direct record | supported as stated | high | [148, 149] |
| CUR-G-F05 | The connection of Stagwell's Media Machine buying agents to Stagwell Curate is stated in the future tense ('its buying agents will connect directly to Stagwell Curate'), so no agent-executed selection on Curate is documented as production use by the cutoff; Stagwell's Q3 2026 results release (8-K exhibit, 29 October 2026, on SEC EDGAR) names neither Stagwell Curate nor The Media Machine, FreeWheel, Magnite, curation or a marketplace, referring only to 'agentic Stagwell Machines' software products, and the 10-Q's financial notes name none of them either. | 8 | synthesis | supported with qualification | medium | [5, 71] |
| CUR-G-F10 | KINESSO's 21 September 2023 page describes 'creating curated marketplaces specific to each client’s function and needs' without naming a product, exchange, SSP, DSP or deal mechanism, and states IPG ownership ('IPG Mediabrands is the media and marketing solutions division of Interpublic Group (NYSE: IPG)'). | 8 | vendor assertion | supported as stated | high | [294] |
| CUR-G-F11 | Omnicom's Form 8-K exhibit of 26 November 2025 announced 'the successful completion of its acquisition of The Interpublic Group of Companies, Inc.' at '0.344 Omnicom shares for each share of Interpublic common stock', with legacy Omnicom holders at 'approximately 60.6%' and legacy Interpublic holders at 'approximately 39.4%'; the release mentions no marketplace, programmatic product or curation. | 8 | direct record | supported as stated | high | [47] |
| CUR-G-F12 | Publicis's 6 March 2025 announcement said Lotame 'will be positioned as part of Epsilon, accessible to all Publicis clients and teams', with closing 'expected to close early in the second quarter of 2025' and price undisclosed; the release describes Lotame's 'end-to-end data marketplaces' and 'global data marketplace' (a data marketplace, not a curated-supply or media marketplace) and does not mention curation, deals, supply or media buying, so Publicis's use of Lotame as a curation asset is not documented on this record. | 8 | vendor assertion | supported with qualification | high | [40] |
| CUR-G-F15 | Kroger Precision Marketing's Private Marketplace, reported on 20 October 2021, lets agencies and brands apply 'Kroger audience data to programmatic campaigns within their preferred ad-buying platform' through 'a deal ID that allows for appropriate programmatic use', with 'quality controls including verification and brand safety tagging'; the inventory, publishers and exchanges are not described, so publisher multiplicity is unknown. | 8 | vendor assertion | supported as stated | medium | [75] |
| CUR-G-F16 | Albertsons Media Collective's offering page describes 'Offsite Display through Google’s DV360 and Collective TV' and a partnership with The Trade Desk to 'bring our verified-buyer audience and measurement solutions', with no curated inventory, private marketplace, deal ID or publisher list on the page; its 2025 blog describes matched 'curated packages' of owned enterprise media offered 'without incurring any additional costs'. | 8 | vendor assertion | supported as stated | medium | [295, 296] |
| CUR-G-F18 | The Trade Desk describes OpenPath (page of 24 February 2026) as 'providing access to high-quality inventory via direct connections between advertisers and a publisher's ad server' and says it 'does not function as an intermediary', with 'hundreds of publishers' (naming New York Post, FUNKE, Valnet, The Arena Group and Vizio); the page states no fee and attributes inventory inclusion to publishers 'making their inventory available', with a '3x higher' win-rate claim footnoted to The Trade Desk platform data for October 2024 across 'dozens of advertisers in various verticals', without a stated method or controls. | 8 | vendor assertion | supported with qualification | high | [297] |
| CUR-G-F19 | The Trade Desk announced OpenSincera on 13 May 2025 as a free application providing 'Sincera's rich advertising metadata', with 'metrics on the ad experience such as ads-to-content ratio, page weight, average ads-in-view, and ad refresh rate', 'available at open.sincera.io, starting on June 6, 2025', after acquiring Sincera 'earlier this year'; the release assigns it no curation or deal-building role. | 8 | vendor assertion | supported as stated | high | [237] |
| CUR-G-F21 | Amazon Ads' Inventory Tier brand suitability control (14 November 2025) offers three advertiser-selected tiers (Expanded, Standard, Restrictive; served to the fetch tool as 'Esteso, Standard e Restrittivo') that 'balance reach and brand safety', and is a brand-suitability control, not a supply-quality or floor mechanism. | 8 | vendor assertion | unresolved | high | [298] |
| CUR-G-F22 | Amazon Ads' undated supply-quality article (served in Italian) says it 'recently' launched the Certified Supply Exchange programme to strengthen relationships with major ad exchanges (none named), offers Amazon Publisher Direct access to 'thousands' of invitation-only publishers, includes its quality measures 'at no additional cost', and charges 'a single 1% technology fee' when advertisers set up guaranteed deals with specific publishers outside Amazon; the page states neither the base of that percentage nor the certified exchanges. | 8 | vendor assertion | supported with qualification | medium | [299] |
| CUR-G-F25 | Apostra's blog of 23 September 2026 states 'Scope3 is becoming Apostra', 'built on AdCP', giving advertisers and media companies 'one place to buy and sell every kind of advertising, anywhere in the world', with sellers told 'You stay in control of your pricing, packaging, and relationships'; it describes no curated-deal product, deal ID or fee. | 8 | vendor assertion | supported as stated | high | [54] |
| CUR-G-F26 | OpenX describes OpenXBuild as a software suite whose Real-Time Bidstream API lets a marketer 'deploy your data and decisioning in a secure container within OpenX cloud infrastructure' to 'enrich and shape bid requests before they reach the DSP', and its 16 June 2026 release says Chalice AI's models 'operate as always-on bidding agents within OpenX's infrastructure' for a Hyundai campaign, with vendor-claimed '20% lower cost per high-value action and 67% lower CPMs' against an undefined 'second-best-on-plan' comparator. | 8 | vendor assertion | supported as stated | medium | [300, 120] |
| CUR-J-F13 | Stagwell launched The Media Machine on 19 June 2026 ('over 20 intelligent agents', connected to 'Google's GMP products, Meta, Microsoft & LinkedIn, TikTok and The Trade Desk', 'keeping humans firmly in control of every critical decision', no client named) and Stagwell Curate on 16 July 2026 ('available immediately to clients across the Stagwell network'), stating that 'its buying agents will connect directly to Stagwell Curate' in the future tense, so no agent connection to the curated marketplace is documented as live. | 8, 10 | vendor assertion | supported as stated | high | [183, 5] |
| CUR-J-F17 | Apostra's post of 23 September 2026 states 'Scope3 is becoming Apostra', that the platform is 'built on AdCP, the open protocol we helped create that lets agents talk to each other', and that 'Today, campaigns are live on Apostra across markets around the world', naming markets (the US, UK, South Africa, Brazil, the Netherlands and France) but no customer, publisher, volume or transaction, and describing a sell-side agent only in product terms. | 8, 10 | vendor assertion | supported as stated | medium | [54] |
| CUR-J-F11 | OpenX and Chalice AI announced on 16 June 2026 that, for advertiser Hyundai, Chalice's models 'operate as always-on bidding agents within OpenX's infrastructure' under OpenX's Custom Decisioning (containerisation) and 'continuously evaluate and value video impressions in real time based on their likelihood to drive high-value actions', claiming a 20% reduction in cost per high-value action and a 67% decrease in CPMs against 'the next-best alternative' (sub-head: 'second-best-on-plan') with no period, agency, sample or method. | 8, 10, 12 | vendor assertion | supported as stated | medium | [120] |
| CUR-U-F30 | Dentsu Aegis Network's release of 15 September 2020 launched 'Dentsu Curate' as 'a new programmatic supply solution for clients' on Xandr's technology platform, deployed in Hong Kong, South East Asia and Singapore and 'to be scaled to other APAC markets', described as a 'Supply Path Optimisation & Curated Marketplace offering'; the product text names no US availability (the page's only US references are a site-footer legal disclaimer and a USA site link), and no later page documents its status or channels. | 8, 11 | vendor assertion | supported with qualification | high | [6] |
| CUR-I-F14 | Six sell-side companies named in the brief sued Google after the liability opinion: OpenX (E.D. Va. No. 1:25-cv-01282, filed 4 August 2025), PubMatic (E.D. Va. No. 1:25-cv-01482, 8 September 2025), Magnite (E.D. Va. No. 1:25-cv-01541, 16 September 2025), Equativ and Sharethrough (E.D. Va. No. 1:25-cv-01755, 13 October 2025), Index Exchange (first in E.D. Va. No. 1:25-cv-02003 on 10 November 2025, voluntarily dismissed by notice of 17 December 2025, then S.D.N.Y. No. 1:25-cv-10477, filed 17 December 2025) and Teads (S.D.N.Y. No. 1:26-cv-06591, 3 August 2026); the OpenX complaint and the PubMatic and Magnite releases cite the April 2025 opinion by date, and the Teads release cites the E.D. Va. ruling without dating it; the opinion is a liability finding, not a judgment. | 9 | direct record | supported with qualification | high | [301, 302, 303, 132, 304, 305] |
| CUR-I-F15 | The competitor suits are coordinated in the S.D.N.Y. MDL (In re Google Digital Advertising Antitrust Litigation, 1:21-md-03010-PKC): Equativ's case was transferred under JPML conditional transfer order CTO-16 (stay lifted 18 December 2025, received 8 January 2026 as 1:26-cv-00140); Pre-Trial Order No. 20 of 13 January 2026 ordered consolidated amended pleadings; a consolidated amended complaint was filed on 23 February 2026 by Equativ, PubMatic, OpenX and Magnite; Google moved to dismiss the consolidated competitor complaints on 29 April 2026 (opposition 27 May, reply 12 June); and the case management plan of 17 June 2026 covers eight competitor actions (OpenX 1:25-cv-10817, Magnite 1:25-cv-10818, PubMatic 1:25-cv-10819, Equativ 1:26-cv-00140, Index Exchange 1:25-cv-10477, Kargo 1:25-cv-10630, Sovrn 1:26-cv-01587, TRUSTX 1:26-cv-03135) with fact discovery due 10 November 2026 and expert discovery 30 April 2027. | 9 | direct record | supported as stated | high | [132] |
| CUR-I-F17 | On 5 September 2025 the European Commission fined Google EUR 2.95 billion in AT.40670, finding dominance in EEA-wide markets for publisher ad servers (DFP) and programmatic ad buying tools for the open web (Google Ads, DV360) and abuse since at least 2014 by favouring AdX in DFP's ad selection and in Google Ads/DV360 bidding; it ordered Google to end the self-preferencing and to cease its conflicts of interest, gave 60 days for a proposal, and stated a preliminary view that only divestment of part of Google's services would do; the provisional non-confidential decision was published on 15 January 2026 and the register shows no later step. | 9 | direct record | supported as stated | high | [134, 133] |
| CUR-I-F18 | On 14 November 2025 Google published its response to the Commission: immediate product changes including letting publishers set different minimum prices for different bidders in Google Ad Manager, proposals on conflicts of interest and interoperability, no break-up, and an appeal; its annulment action T-794/25 (Google and Alphabet v Commission) was lodged at the General Court on 20 November 2025, published in the Official Journal on 12 January 2026, and is pending. | 9 | vendor assertion | supported as stated | medium | [135, 136] |
| CUR-I-F19 | On 12 March 2026 Executive Vice-President Ribera answered a parliamentary question that the AdTech decision imposes two obligations (cease the practices; eliminate conflicts of interest), that the Commission's preliminary view in its statement of objections was that a structural remedy is likely the only way to meet the second, that it 'has not reached a final conclusion on that matter', and that it was seeking market participants' input on Google's compliance plan, including from Proxistore. | 9 | direct record | supported as stated | high | [137] |
| CUR-I-F20 | The CMA's Competition Act investigation into Google's ad-tech conduct (case 51145 per the brief; opened 25 May 2022; statement of objections 6 September 2024) remained 'Open' with no decision, its latest timetable phase being 'Considering next steps, whilst monitoring international developments' estimated for December 2025 to July 2026, on a page last updated 19 May 2026; the CMA's strategic market status designations of Google in general search and search advertising (10 October 2025) and in its mobile platform (22 October 2025) concern search or mobile platforms, and the two case pages' update logs through 23 September 2026, which list the conduct requirements and commitments made under them, mention no ad-tech, ad exchange, display or deal mechanics (the conduct-requirement and commitment documents themselves were not opened). | 9 | direct record | supported with qualification | high | [138, 157, 158] |
| CUR-I-F21 | By the cutoff, twenty-three US states had enacted comprehensive consumer privacy laws: the nineteen on IAPP's chart of 24 November 2025, all in force by 1 January 2026 (the latest effective dates being Indiana, Kentucky and Rhode Island), plus four enacted in 2026 (Oklahoma, Alabama, Louisiana and Vermont, per IAPP's report of 17 June 2026) that take effect in 2027 and 2028; so nineteen laws were in force at the cutoff. These state laws, not a federal statute, are the US rules that govern the sale and processing of personal data used in in-flight enrichment. | 9 | synthesis | contradicted | medium | [85] |
| CUR-I-F22 | California's privacy regulator lists its 'CCPA Updates, Insurance, Cybersecurity Audits, Risk Assessments, and Automated Decisionmaking Technology (ADMT) Regulations' as a completed package of September 2025 and states that the CCPA regulations and the data broker registration and accessible deletion mechanism (DROP) regulations are 'Effective January 1, 2026'. | 9 | direct record | supported as stated | medium | [159] |
| CUR-I-F23 | On 14 January 2025 the FTC finalised a consent order banning data broker Mobilewalla from collecting consumer data from online real-time bidding advertising exchanges for purposes other than participating in those auctions, the first time the agency alleged that practice to be unfair, and from selling sensitive location data; the Commission voted 4-1. | 9 | direct record | supported as stated | high | [160] |
| CUR-I-F24 | The Court of Justice ruled on 7 March 2024 (C-604/22) that the TCF's TC String is personal data where it can by reasonable means be associated with an identifier such as an IP address, that IAB Europe is a joint controller for the TC String's creation and use, and that this joint controllership does not extend automatically to participants' subsequent processing for targeted advertising. | 9 | direct record | supported as stated | high | [161] |
| CUR-I-F25 | By IAB Europe's own account of 14 May 2025, the Belgian Market Court annulled the APD's February 2022 decision except the part finding IAB Europe a joint controller, which it upheld solely for the creation and use of TC Strings and not for participants' own processing (IAB Europe's words: 'upheld only part of the APD decision'), with execution of the 2023 action plan remaining suspended; a law-firm summary reports the EUR 250,000 fine as upheld; IAB Europe launched TCF v2.3 in April 2025 with an adoption deadline of 28 February 2026. | 9 | vendor assertion | supported with qualification | low | [162, 163] |
| CUR-J-F01 | On 3 November 2026 the AdCP repository's release tags were v3.2.8 (published 28 October 2026, marked Latest, not a pre-release), v3.1.31 (30 October 2026) and v3.3.0-beta.3 (31 October 2026, pre-release), with v3.1.30 (28 October), v3.3.0-beta.2 (30 October) and v3.3.0-beta.1 (29 October) between them. | 10 | direct record | supported as stated | high | [175] |
| CUR-J-F04 | AdCP's publisher states that 'AdCP 3.2 is now generally available' on its homepage without a date, and its docs intro states 'AdCP 3.2 is generally available (release 3.2.1, wire pin "3.2")' while the repository's latest 3.2 tag is v3.2.8 and the intro notes that 'One architecture does not mean every channel or commercial function has equal implementation depth today.' | 10 | direct record | supported as stated | high | [167, 97] |
| CUR-J-F05 | AgenticAdvertising.org's own pages claim that 'On October 16, 2025, the first-ever agent-to-agent media buy was executed' with 'real money, real inventory from LG Ads', with agents handling 'creative, targeting, and approval with human oversight', and that '157+ companies have joined' since; AdExchanger's launch report of 15 October 2025 describes only a hypothetical press demo by Swivel and PubMatic's statement that 'aspects of an overall campaign are being tested and executed', and no record of the 16 October buy from a party other than the protocol's publisher was opened in this stream. | 10 | vendor assertion | supported as stated | medium | [167, 177, 166] |
| CUR-J-F07 | The Prebid Sales Agent (github.com/prebid/salesagent, 'A media sales agent that implements the AdCP Media Buy protocol') is labelled 'Status: alpha. A pre-1.0 implementation of a pre-release protocol (AdCP 3.1.1), under active development', targets 'AdCP spec version 3.1.1 via the adcp==6.6.0' Python SDK as package adcp-sales-agent 2.0.0, had its latest commit on main on 28 October 2026 (1c202df2) and 39 stars; docs.prebid.org gives it no version or status label and names no deployment; AdCP's docs call it 'a community example, not a maintained reference implementation'. | 10 | direct record | supported as stated | high | [171, 178, 97] |
| CUR-J-F08 | IAB Tech Lab's AAMP 3.0 was announced on 22 September 2026 with OpenProposal, 'a new specification that defines how sellers can represent advertising products', in public comment originally 'until October 22, 2026' and extended 'to Tuesday, November 10, 2026'; the OpenProposal repository's README states 'This is a draft. Nothing here is a ratified IAB Tech Lab standard. Do not build production systems against it expecting stability', with the comment draft tagged v3.0-draft-1 on 24 September 2026; AAMP 2.3 was released on 30 July 2026, and the standard page describes AAMP 2.0 ('added 5 new transaction types') without a date in its text. | 10 | direct record | supported as stated | high | [170, 168, 179, 92] |
| CUR-J-F09 | AAMP's reference implementations are the Seller Agent SDK (v2.4.2, 18 August 2026; last commit 28 October 2026), 'An AI-powered inventory management system for publishers and SSPs to automate programmatic deal negotiation, booking, and distribution using IAB Tech Lab standards (OpenDirect 2.1, Deals API v1.0, sellers.json)' whose README states 'Curators package and curate inventory on behalf of buyers. Agent Range is pre-registered to curate deals.', the Buyer Agent SDK (v2.4.2, 18 August 2026; last commit 28 October 2026) and the Core Orchestration repository agentic-direct, whose MCP layer 'Returns mock responses' and which was last pushed on 28 January 2026. | 10 | direct record | supported as stated | high | [169, 180, 181, 306] |
| CUR-J-F10 | No named AAMP production deployment with a period and an executed decision appears in the sources opened: the standard page says AAMP 2.3 'takes agentic media buying beyond demos and first live implementations to capabilities integrated into' production media buying stacks without naming any; the AAMP 3.0 release's only pilot is IAB Australia's seller-side pilot in a sandbox 'before any live spend is involved', WPP Media's quote anticipates movement 'from proof-of-concept tests towards the wide-spread use of AI Agents', and the press index carries 'PMG Integrates IAB Tech Lab AAMP Standards Into Alli Operating System' (7 April 2026, not opened). | 10 | synthesis | supported as stated | medium | [92, 170, 179] |
| CUR-J-F12 | InMobi's Buyer Hub page, dated 'Agency Access \| August 31st, 2026', describes 'An AI Agent for Media Buyers' that lets agencies 'Discover inventory, build deals, and optimize campaigns through natural language, with traders in control of every decision', says it was earlier opened to curators, names no customer, exchange or DSP and makes no US-availability statement; separately, the AdCP registry lists an 'InMobi Sales Agent' at a production-named host (adcp-prod.exchange.inmobi.com) and a non-production twin. | 10 | vendor assertion | supported as stated | medium | [174, 165] |
| CUR-J-F14 | Magnite's release of 16 September 2026 names ITN, DIRECTV Advertising, Samsung Ads, iHeartMedia, Newton Research, Databricks, Butler/Till and TF1 PUB around Magnite Orchestration and says 'Key partners across the industry are using Magnite Orchestration to bolster their agentic strategies', but states no executed transaction, names neither AdCP nor AAMP, and does not mention a product called Magnite Showcase; TF1 PUB's quote is 'we're interested in exploring how Magnite Orchestration can build on that foundation'. | 10 | vendor assertion | supported as stated | medium | [184] |
| CUR-J-F18 | Kargo's Project KERA, 'its new agentic media buying and creative engine' that 'turns a single campaign brief into a fully executed, performance-ready media plan', was announced as a closed beta in a release reproduced by ExchangeWire PressBox on 7 April 2026 naming 'Hershey's' as 'one of Project KERA's Beta test partners, along with other CPG, travel, and emerging brands', with workflows 'from full end-to-end execution to collaborative, human-in-the-loop workflows', no protocol named and no quantified results. | 10 | vendor assertion | supported as stated | medium | [173] |
| CUR-J-F19 | Microsoft Advertising's statements are positioning, not production records: the 14 May 2025 post says 'the future of digital engagement is conversational, personalized, and agentic', that 'we will no longer support media buying through our DSP, Microsoft Invest, starting on February 28, 2026', and that 'through Microsoft Curate, we're empowering publishers to monetize proprietary data and access unique demand', without describing agents executing transactions; a 10 September 2026 post on Microsoft Curate ('The Power of Curation') is listed with the summary 'Learn how Microsoft Curate helps publishers and data owners turn proprietary assets into scalable revenue through programmatic curation', and no Microsoft agent or AdCP/AAMP connector was found in the sources opened. | 10 | vendor assertion | unresolved | medium | [185, 186] |
| CUR-J-F20 | Index Exchange has no entry in the AdCP registry's 34 agents and no agent announcement was found in four searches (two site-restricted), and PubMatic, Magnite, Yahoo and Kargo, all named on the AdCP homepage as members, also have no registry entries, so the registry under-represents the implementations that vendors document elsewhere. | 10 | synthesis | supported as stated | medium | [165, 167] |
| CUR-FINAL-F03 | PubMatic's release of 27 April 2026 claims '30 live end-to-end agentic campaigns now running globally at time of publication', in the United States, France, the Netherlands, Australia and India, and names Amnet's campaign for Interbev in France and Abovo Maxlead's CTV campaign in the Netherlands, without campaign dates, spend or a description of what the agents selected; the counts and outcomes are the vendor's own. | 10 | vendor assertion | supported as stated | medium | [307] |
| CUR-J-F02 | AdCP's README on main (3 November 2026) lists Media Buy (get_products, create_media_buy, get_media_buy_delivery), Signals (get_signals, activate_signal) and Trusted Match among its protocol surfaces and lists a 'Curation' surface for 'Media inventory curation' as 'Coming soon', so curation has no schema representation in AdCP at the cutoff. | 10, 12 | direct record | supported as stated | high | [164] |
| CUR-J-F03 | AdCP's documentation defines roles by function ('Media buying agents discover inventory and execute campaigns'; 'Signals agents find and activate audiences'; 'Orchestrators coordinate the workflow and make sure humans approve what matters'), gives no definition of a 'sales agent' beyond 'sales-agent endpoints' in brand.json, and in the Media Buy walkthrough describes a seller-planned path in which 'The seller should curate, price, or negotiate a plan from the brief' with draft proposals as seller-authored media plans. | 10, 12 | direct record | supported as stated | high | [97, 176] |
| CUR-J-F06 | The AdCP agent registry API listed 34 agents on 3 November 2026: 27 typed 'sales', 4 'signals', 1 'buying' (Swivel), 1 'creative' (Celtra) and 1 'unknown'; the list includes explicitly non-production endpoints (AdCP Test Agent, 'InMobi Exchange ADCP NonProd', 'goTom Sandbox', 'idm by briefable(demo)'), six publisher storefronts hosted on Optimera's Prebid Seller Gateway (WebMD, TheWrap, Bustle Digital Group, Carpenter Media Group, Trusted Media Brands, Mediatonik), production-named endpoints for InMobi ('InMobi Sales Agent', adcp-prod.exchange.inmobi.com), Equativ, LoopMe, BidMachine, Dstillery, Rediads, Pubx and others, and two entries belonging to the author (excluded from evidence). | 10, 12 | direct record | supported as stated | high | [165] |
| CUR-J-F16 | Marketing Dive reported on 17 March 2026 that a Clubtails campaign for Geloso Beverage Group, run by agency Butler/Till, was 'planned and implemented completely by artificial intelligence agents' on PubMatic's platform (named AgenticOS by March 2026; the 15 December 2025 launch release, read in a syndicated copy, says the campaign 'is running through PubMatic's Activate platform' and calls the workflow 'AdCP-enabled', citing AdCP as 'an open protocol from AgenticAdvertising.org'), that it 'ran in December 2025' across CTV, online and mobile, geotargeted to adults 21 and over, with a brief submitted through Butler/Till's enterprise Claude interface; the figures, buy-side supply chain costs 'reduced ... by 5.5x' ('according to press materials shared with Marketing Dive'), '40% more impressions than planned for', 'a 98% video completion rate' and 'Less than 1% of inventory failed DoubleVerify standards' (per press details) and '0% of the inventory comprised made-for-advertising sites' (attributed to PubMatic), come from the transacting parties and none is independently verified; the December release gave no figures and promised full results for Q1 2026. | 10, 12 | vendor assertion | supported with qualification | medium | [93] |
| CUR-E-F38 | On the vendor pages the stream opened no cohort A product names more than one DSP in text: Microsoft names Microsoft Invest (whose media buying Microsoft said on 14 May 2025 it would no longer support from 28 February 2026) plus 'all major DSPs', FreeWheel names Buyer Cloud plus 'any DSP', Commerce Grid's product page counts '60+ third party DSPs', Google refers to 'all DSPs integrated with agency deals', Index to 'all DSPs', Sovrn shows DSP logos only as image file names, and the other eight name none; Commerce Grid's knowledge base, however, documents Deal Sync to DV360 and The Trade Desk with deal surfacing one and three hours after creation, which is two named third-party activation routes with documented functionality. | 11 | synthesis | supported with qualification | medium | [56, 16, 266, 188] |
| CUR-U-F01 | TL Direct, TripleLift's self-serve curation platform, became generally available on 10 June 2026 to brands, buyers, publishers and curators after a beta, with an advertised scope of web, mobile and connected TV, UI, API and MCP access, a '5K+' direct publisher network and no DSP named as a route. | 11 | vendor assertion | supported as stated | high | [130] |
| CUR-U-F04 | Equativ's curation platform, Maestro by Equativ, advertises display, video, native and CTV formats in a 'global omnichannel marketplace' and publishes unperiodised platform figures of 500+ self-serve clients, 30K+ deals created and 50 billion impressions delivered, with no DSP, fee amount or US statement on the page. | 11 | vendor assertion | supported as stated | medium | [15] |
| CUR-U-F06 | Nexxen Curated Marketplace is offered by Nexxen International Ltd. to data owners, retail media companies, curators, agencies and publishers to 'build custom deals and premium PMPs' 'across device types', with no channel, DSP, fee, scale figure, status word or geography stated on the page. | 11 | vendor assertion | unresolved | high | [254] |
| CUR-U-F09 | Criteo's Commerce Grid is described as 'the only SSP built to seamlessly connect media owners with commerce advertising budgets at scale', serving media owners, agencies and retailers across 'display, in-app, web, mobile, native, video, and CTV', connected to '60+ third party DSPs', with curation fees that 'only apply if curated deals win impressions' and company-level demand figures ('over $2.2B in annual ad spend from 17,000+ global brands and agencies') that are not curated-deal figures. | 11 | vendor assertion | supported as stated | medium | [266] |
| CUR-U-F23 | Dstillery's 'FAQ: Deal IDs' post, dated 25 July 2022 but naming 'the Microsoft Curate supply exchange' as where its ID-free Deal IDs are created, says deals 'can be created for Display, Video, and Native' and 'can be activated on any leading DSP', with Custom AI audiences as 8 deals and pre-built audiences as 1; the post's current status cannot be settled from the page because the Microsoft Curate name post-dates the post date. | 11 | vendor assertion | supported as stated | low | [61] |
| CUR-U-F25 | Semasio's homepage (copyright 2025, interest-marked) states 'Semasio is now part of Samba' and lists HypertailPMP as a service ('Reach your ideal consumers in long-tail media spaces'), names PubMatic only in a testimonial and Denmark only in client names, and gives no deal-ID mechanism, channel, fee, geography or legal entity, so the register's curated-deal row is coded 'unclear' pending the service page. | 11 | vendor assertion | unresolved | medium | [270] |
| CUR-U-F28 | Stagwell announced on 16 July 2026 (New York City) that Stagwell Curate, 'a new centralized deal marketplace and bespoke inventory curation platform', is 'available immediately to clients across the Stagwell network through their existing agency teams' across 'CTV, online video, display and audio', with FreeWheel and Magnite as partners and 'supply-quality models powered by The Trade Desk's Sincera', and that the Media Machine's buying agents 'will connect directly to Stagwell Curate'. | 11 | vendor assertion | supported as stated | high | [5] |
| CUR-U-F35 | None of the 34 readable pages discloses a periodised curated-deal spend or impression figure for the product; the only product-level volume figures are Maestro's '30K+' deals created and '50 Billion' impressions delivered (no period) and Multilocal's '21+ SSP curation marketplaces', while the large numbers on the pages (PubMatic's 821 billion daily ad impressions, Criteo's $2.2B annual ad spend, OpenX's 200k+ domains, Sovrn's 9K+ daily campaigns) are company-wide, so the proposed dimension 8 thresholds ($500 million or 100 billion impressions in a 12-month window) cannot be applied from public pages for any product in the register. | 11 | synthesis | supported as stated | medium | [15, 256, 223, 266] |
| CUR-CLEARLINE-F08 | For ClearLine curated deals, auction type, priority, floor treatment, exclusivity and request sampling are all unknown on the record opened: no source states any of the five, and the Q2 2026 10-Q defines reserve auctions only at company level as those 'where publishers establish direct deals or private marketplaces with select buyers'. | 11 | synthesis | supported as stated | medium | [107, 213, 214] |
| CUR-CLEARLINE-F11 | TAG's release of 5 March 2026 lists 'Magnite', 'Magnite + DV / MDV+' and 'Magnite CTV / MCTV' among 196 companies awarded 307 seals in 2026 across its four programmes without stating each company's seals, while Magnite's Responsible Advertising page names no certification and describes inventory quality guidelines that 'address inventory authorization' and real-time elimination of non-human traffic with unnamed technology. | 11 | direct record | supported as stated | medium | [242, 243] |
| CUR-LOTAME-F01 | Publicis Groupe recorded in note 3.1 of its 2025 half-year financial report the acquisition in April 2025 of 100% of Lotame Solutions, Inc. for a cash consideration of EUR 127 million with no earn-out, identifiable net assets of EUR 31 million and provisional goodwill of EUR 96 million, the fair values being provisional pending an independent valuation. | 11 | direct record | supported as stated | high | [41] |
| CUR-LOTAME-F04 | Lotame's undated (©2026) capability page sells 'Lotame's Curated Marketplaces' as 'customized connected TV marketplaces, along with PMPs for web and mobile' built from inventory Lotame 'curates' from unnamed 'leading SSPs' and made 'available to DSPs' via PMP, with 'Hand-selected supply partners and inventory' and operational support for 'launch and pacing', and names no SSP, DSP, deal ID, fee, figure or geography in its text. | 11 | vendor assertion | supported as stated | high | [13] |
| CUR-MICROSOFTCURATE-F05 | Microsoft Curate's documentation sets the Auction Type of every curated deal to 'Open Auction', states that curated transactions are not guaranteed deals and never override seller floor or ad-quality settings, and has curated bids compete in the seller's final auction against all other deal and open-exchange bids net of fees, which the fees answer makes the default 'unless you agree otherwise with the applicable curator'. | 11 | vendor assertion | supported as stated | high | [110, 111] |
| CUR-MICROSOFTCURATE-F08 | A Microsoft Curate curated deal line item is set up for one buyer selection that cannot be changed after saving, while the same seller inventory remains biddable on the open exchange. Microsoft Curate's release note of 21 January 2026 documents one multi-buyer mechanism: for DV360 buyers the selection can be a Central Partner ID, which makes a single deal 'automatically eligible for all DV360 Partner IDs' associated with it (one Central Partner ID per deal). The product page's broader 'ability to assign deals to multiple buyers on any DSP' is documented only for DV360 on the Learn pages opened. | 11 | vendor assertion | supported with qualification | high | [110, 111, 121] |
| CUR-MICROSOFTCURATE-F17 | The only assurance record found for the operator, the MRC's continuing accreditation of 'Microsoft Advertising' of 31 January 2025, covers desktop and mobile web paid search clicks and the SIVT functionality applied to them, with a TAG Certified Against Fraud addendum limited to that scope, so it does not cover Microsoft Curate or Monetize. | 11 | direct record | supported as stated | high | [219] |
| CUR-MICROSOFTCURATE-F22 | The curated-deal line item offers Ad Type Banner, Video ('includes audio types as well') and Native and Inventory Type App & Web, App Only or Web Only, and Microsoft's 2026 post adds CTV inventory through Monetize; no Microsoft page documents live execution per environment, and the product's advertised channel scope is not stated on its About page. | 11 | vendor assertion | supported as stated | medium | [110, 220] |
| CUR-PUBMATICCONNECT-F06 | PubMatic names only its own Activate as a buying endpoint for curated deals ('Curated deals can be activated through PubMatic’s Activate solution or your DSP of choice'), asserts integration 'with all major DSPs' without naming one, and no DSP is named in any Connect, auction-package, Comscore or Adloox record opened, where the Adloox steps say only that a DSP and buyer are selected when the deal is created. | 11 | vendor assertion | supported as stated | medium | [223, 222, 119, 12] |
| CUR-PUBMATICCONNECT-F17 | PubMatic's independent assurance on the record is company-level: TAG's release of 4 March 2026 lists PubMatic 'Among the companies that achieved TAG Certification Seals for 2026' without naming the seal or its validation mode, and the FY2025 10-K self-describes fraud controls; no audit or accreditation scoped to Connect deals was found. | 11 | direct record | supported as stated | medium | [242, 231] |
| CUR-A-F35 | TAG offers four seals, each certifying a company rather than a deal or product: Certified Against Fraud (launched 2016; requires that 100% of monetizable transactions are filtered for IVT by a qualified third party, with data-centre IP filtering), Certified Against Malvertising, Brand Safety Certified and Certified for Transparency (announced November 2022; guidelines PDF file-named 2023); recertification for all TAG seals happens in January every year with applications due 31 January; the Brand Safety and Transparency pages list no requirements in their text, and 'TrustNet' appears only on the Transparency page as a distributed-ledger platform 'Powered by technology partner Fiducia', not as a seal. | 11, 12 | direct record | supported with qualification | medium | [308, 152, 155, 154, 153] |
| CUR-A-F34 | The MRC auction standard requires defined, documented and disclosed supply-quality requirements enforced through validated measurement, and encourages third-party MRC-accredited IVT and content-level brand-safety measurement pre- and post-bid, naming 'possible future Made for Arbitrage/MFA or carbon considerations' as not yet covered. | 12 | direct record | supported as stated | high | [151] |
| CUR-AUDIGENT-F15 | Coding Q1 from the opened record: Audigent owns the selection rule; the execution host is the exchange (Google Ad Manager, Index Exchange or Nexxen); the data suppliers are Audigent's publisher and data partners, plus Experian and The Weather Company in the DV360 packages; the contracting seller is the publisher through the exchange (publishers on Google can switch off inventory packaging, which is on by default). The fee recipient on Google is Audigent, paid by Google from the bid. Who may change a deal after purchase is not documented. The mechanism is in-flight enrichment and in-auction selection (inference from 'supply-side activation' and the Google packaging settings), traded as Preferred Deals and Private Auctions on Google at launch. | 12 | inference | supported as stated | medium | [59, 196, 190] |
| CUR-LOTAME-F16 | Coded against Q1, Lotame owns the audience-led selection rule and can change a running deal ('Continuous supply-led optimizations'; operational support for 'launch and pacing'), the execution host is a partner SSP (Google Ad Manager is the only one documented as hosting Lotame as a curator), the data supplier is Lotame with its business customers and data partners, and the contracting seller, fee recipient and transaction defaults are not documented, except that on Ad Manager Google 'will handle billing'; in the one case the buyer's allow-list also set the inventory rule. | 12 | inference | supported as stated | medium | [13, 125, 196, 208] |
| CUR-MICROSOFTCURATE-F21 | On the Microsoft pages opened the only documented post-purchase controls belong to the curator, the seller or Microsoft (margins editable 'at any time'; buyer fixed after saving; flight start dates fixed once passed; deals deactivated or deleted with reports retained; sellers able to block curators; Microsoft-applied sampling), and no page documents buyer notification of changes, an audit right, buyer data export or a remedy for misdescribed inventory or charges. | 12 | vendor assertion | supported as stated | medium | [115, 110, 113] |
| CUR-PUBMATICCONNECT-F22 | PubMatic documents reporting for Connect deals ('deal-level usage reporting' for buyers; 'a near real-time view of your data deals' for publishers; 'Easily access transaction logs for verification' in the console) but no source opened covers change notification, audit rights, data exportability or remedies for misdescribed inventory or charges. | 12 | vendor assertion | supported as stated | medium | [246, 20, 222] |
| CUR-F-F25 | IAB Tech Lab's Deal Sync API 1.0 Curation object carries the curator's domain, a curator deal id and a fee type ('0 = undisclosed', '1 = percentage of spend', '2 = flat fee', '3 = CPM', '4 = no fee is paid for curation services') 'but not what that fee is', and version 1.0 'does not support deal revisions'; no cohort B vendor page opened declares a fee type, and the searches found no curator or SSP confirming production support of the API by 3 November 2026. | abstract, 1, 3, 6, 12 | direct record | supported as stated | high | [96] |
| CUR-A-F10 | None of ads.txt 1.1, sellers.json 1.0 or SupplyChain 1.0 defines a curator role: a case-insensitive search of the three raw specification texts on 3 November 2026 found 'curat' once in ads.txt (about licensing CTV content) and never in sellers.json or SupplyChain, so a curator appears in the files only where it is a paid seller (seller_type INTERMEDIARY or BOTH; an hp=1 node) and a non-payment data or selection provider cannot be represented at all (inference from CUR-A-F09). | abstract, 1, 3, 11, 12 | synthesis | supported as stated | high | [143, 142, 141] |
| CUR-A-F12 | The Deal Sync API 1.0 Curation object carries curator (the canonical domain of the entity that packaged inventory, technology and/or data), cdealid (the deal id in the curator's namespace) and curfeetype, an integer whose values are 0 undisclosed, 1 percentage of spend, 2 flat fee, 3 CPM and 4 no fee is paid for curation services; the object's own definition of curation is 'the selection and organization of inventory using technology and data'. | abstract, 1, 3, 12 | direct record | supported as stated | high | [96] |
| CUR-E-F01 | Microsoft Curate's documentation (ms.date 2026-02-18 and 2025-10-21) sets the Auction Type of every curated deal to 'Open Auction', states that curated transactions are not guaranteed deals and never override seller floor or ad-quality settings, and has curated bids compete in the seller's final auction against all other deal and open-exchange bids net of fees, which the fees FAQ makes the default 'unless you agree otherwise with the applicable curator'. | abstract, 1, 4, 12 | vendor assertion | supported with qualification | high | [110, 111] |
| CUR-E-F04 | Microsoft's Curation Optimization samples curated deals that a buyer has stopped bidding on, is applied by Microsoft without curator action, and can reduce the impressions matched and bid requests a curator and buyer observe on a low-spend deal, which the page attributes to sampling rather than to available supply. | abstract, 1, 4, 12 | vendor assertion | supported as stated | high | [113] |
| CUR-E-F10 | Google Ad Manager's curation (Beta) lets curators bundle Private Auctions and Preferred Deals into curated packages, and Ad Manager defines a Private Auction as an auction-based transaction with buyers allowlisted by company and a Preferred Deal as a fixed-price, pre-auction transaction with specific buyers, so a curated package built on Preferred Deals carries pre-auction access by transaction-type definition. | abstract, 1, 4, 12 | vendor assertion | supported as stated | high | [24, 250] |
| CUR-F-F26 | Across the twelve cohort B register seeds, no vendor page discloses a curation fee with amount or rate, base, payer, collector and period: one names charge types only (Audigent: 'bid floors and curation fees only', beside a 'more fairly priced' assertion), one discloses a data-fee amount with base but no payer, collector or period (Adloox: $0.15 CPM segments inside PubMatic auction packages), three make general assertions only (Lotame 'data/tech fees' cut; Seedtag 'Lower effective CPMs' and 'margin logic'; Adelaide 'a fair price'), one is request-only (Dstillery) and six state nothing (Experian, Comscore, IAS, Multilocal, Semasio, Samba). | abstract, 1, 6, appendix | synthesis | supported with qualification | high | [116, 12, 209, 128] |
| CUR-G-F01 | Stagwell announced Stagwell Curate on 16 July 2026 as 'a new centralized deal marketplace and bespoke inventory curation platform' that is 'part of the Stagwell Media Platform' and 'available immediately to clients across the Stagwell network through their existing agency teams with no DSP migration required', across streaming TV, video, display and audio. | abstract, 1, 8 | vendor assertion | supported as stated | high | [5] |
| CUR-G-F03 | Stagwell's 27 April 2026 release (read in ExchangeWire's reproduced copy) says 'FreeWheel will power Stagwell Curate' and that 'FreeWheel’s Curation Hub and Buyer Cloud will integrate into Stagwell’s media acquisition layer', so the execution host of Stagwell Curate is FreeWheel's curation and buying technology, with the marketplace consolidating 'preferred supply relationships across the Stagwell network into a single, always-on, source of trusted supply'. | abstract, 1, 8 | vendor assertion | supported as stated | medium | [72] |
| CUR-G-F04 | Neither Stagwell's release nor Digiday's report discloses a fee, take rate or margin for Stagwell Curate; the release speaks of 'exclusive pricing when sourcing inventory for every buy' and the April release of 'material reduction in ad tech fees', without figures. | abstract, 1, 8 | synthesis | supported as stated | high | [5, 72] |
| CUR-D-F19 | Magnite's FY2025 10-K (Nasdaq Global Select Market, MGNI) states that revenue is generally a percentage of the ad spend that runs through its platform, uses 'take rate' only qualitatively, and describes ClearLine as providing 'integrated curation capabilities' with a 'Magnite Curator Marketplace, a self-service solution integrated within ClearLine'; no numeric take rate or curated-deal share appears in the portion read. | abstract, 2, 5 | direct record | supported as stated | high | [212] |
| CUR-D-F20 | PubMatic's FY2025 10-K (Nasdaq Global Market, PUBM; 8,263,239 Class B shares outstanding) states that it generates revenue through fees charged to publishers that are generally a percentage of the value of ad spend, describes Connect as 'our solution that provides additional data and insights to buyers' and Activate as allowing buyers 'to execute direct deals on our platform across our publisher inventory', names 'data partners and curators' as a customer group, and warns that insiders have substantial control through the dual-class structure; 'take rate', 'curation' and 'curated' do not appear in the portion read. | abstract, 2, 5 | direct record | supported as stated | high | [231] |
| CUR-B-F10 | For October to December 2025 (35 contributing of 54, $142 million, 16.7 billion impressions) the aggregate PMP share fell to 79.2% (OMP 20.8%) while the median rose to 92.2%, CTV was 100% PMP on both measures, web plus mobile was 64.9% aggregate and 87.4% median, and spend-weighted CPMs were $10.72 for PMP against $4.25 for OMP. | abstract, 2, 6, 7 | independent measurement | supported as stated | high | [100, 309] |
| CUR-B-F11 | Between Q2 and Q4 2025 the ANA's median PMP share rose by 4.4 points (87.8% to 92.2%) while the aggregate fell by 3.4 points (82.6% to 79.2%), and the contributor panel changed from 21 to 35 advertisers, so the two series cannot be read as one trend and neither identifies multi-publisher curated deals. | abstract, 2, 7 | synthesis | supported as stated | high | [98, 100, 66] |
| CUR-B-F14 | The benchmark glossary defines a private marketplace as 'A curated programmatic deal environment where buyers transact with selected publishers, often offering premium inventory', names only OMP and PMP as marketplace types, and has no entry for curation, deal ID, preferred deal or programmatic guaranteed, so the ANA series cannot isolate multi-publisher curated deals. | abstract, 2, 7 | direct record | supported as stated | high | [310, 287] |
| CUR-B-F18 | Jounce's State of the Open Internet 2025 shows curation (exchange-controlled and curator-controlled demand) at 72.7% of available RTB auctions, the open auction at 18.6% and publisher-controlled private marketplaces at 8.7%, with the caption '4 of every 5 auctions is an attempt by the sell-side to control demand', and states that DSP-controlled demand 'is far more than 18.6% of programmatic ad spend'; AdExchanger dates the analysis to March 2025 on 1.5 billion bid requests from 44 exchanges. | abstract, 2, 7 | independent measurement | supported as stated | high | [101, 64] |
| CUR-B-F31 | In the sources screened by 3 November 2026 (the ANA 2023 study and benchmark editions to Q1 2026, Jounce's 2023 and 2025 editions and its April 2025 fee report as reported, both ISBA/PwC studies, Adalytics' blog index and three reports, and one academic thesis record), this review found no independent measurement with a published method that tests a named curated deal's quality promise or all-in cost against the same inventory in the open auction, and no public series that identifies multi-publisher curated transactions on a constant definition, so H1, H2a and H3 remain unresolved on this stream. | abstract, 2, 7, 11, appendix | synthesis | supported as stated | medium | [310, 101, 311] |
| CUR-E-F37 | No cohort A product documents a buyer-facing list of the sellers or domains inside a named curated deal on the pages opened: Microsoft exposes the deal name only, Google's agency seat shows name, pricing and a curator label (and, for publisher Marketplace Packages, a description and publisher name), Index and FreeWheel assert dashboards and reporting, OpenX names example publishers in packages, and Yieldmo asserts URL-level visibility before activation without an identified example deal. | abstract, 3, 12, appendix | synthesis | supported as stated | medium | [110, 248, 25, 11, 258] |
| CUR-I-F01 | On 17 April 2025 the E.D. Va. court found (Dkt 1410, 115 pages) that Google had monopolised the worldwide open-web display publisher ad server market (Count I) and the worldwide open-web display ad exchange market (Count II) and unlawfully tied DFP and AdX (Count IV), and dismissed Count III (advertiser ad network market); it entered no remedy and set a remedies schedule. | abstract, 4, 9 | direct record | supported as stated | high | [78, 156] |
| CUR-I-F06 | On 2 September 2026 the court (Order Dkt 1857; Memorandum Opinion Dkt 1858, 106 pages) rejected all structural remedies (AdX divestiture, open-sourcing DFP's final auction logic, contingent divestiture of 'DFP Remainder') and accepted most proposed behavioural remedies as modified; the opinion was filed under seal and unsealed on 16 September 2026 (Dkt 1860). | abstract, 4, 9 | direct record | supported as stated | high | [156, 80] |
| CUR-I-F13 | As of CourtListener's docket copy last updated 29 October 2026, no final judgment had been entered and no notice of appeal had been filed in United States v. Google LLC (E.D. Va.); the last known filing was the joint proposed final judgment of 2 October 2026 (Dkt 1861). | abstract, 4, 9, appendix | direct record | supported as stated | medium | [80] |
| CUR-I-F26 | Inference: the US competition record addresses the auction mechanics of one ad server and one exchange (DFP and AdX) for open-web display 'Indirect Demand' (open and private auctions); none of the liability opinion, the remedies opinion or the proposed judgment mentions curators, curation, curated deals or third-party deal IDs (the one occurrence of 'curate' is remedies-trial testimony, quoted at Dkt 1858 p. 85, that advertisers expect their DSPs to 'curate or prioritize one ad exchange over another'), so the record documents observed, court-found first-look, last-look and floor advantages for Google's exchange and a proposed restoration of publishers' per-source floor control, and establishes nothing about the default priority, floors or exclusivity of curated deals on any exchange. | abstract, 4, 9, appendix | inference | supported with qualification | medium | [79, 156, 78] |
| CUR-INDEXMARKETPLACES-F05 | Index's knowledge base documents that, 'unless otherwise specified', Marketplace bids compete in the publisher's auction net of all Marketplace fees and that incremental demand wins only if its net bid is higher than any other net bid; that rule nets fees and does not settle priority, because the package page has the creator choose a First or Fixed auction with no default and defines a fixed price auction as having priority over all other auction bid types, and no page opened states an exclusivity default or a request-sampling rule. (Corrected in the final live sweep of 4 November 2026; see the corrections ledger.) | abstract, 4, 12 | vendor assertion | supported as stated | medium | [198, 109] |
| CUR-E-F35 | Under the H4 coding rule, read with the register's own distinction between a default and a setting the curator chooses (configurable), the essay's claim that a curated deal carries no default priority, floor advantage or exclusive supply holds for Microsoft Curate (open-auction packaging, seller floors respected, one buyer per deal on supply that stays open in the open exchange) and is unresolved for the other fourteen exchange products: Google Ad Manager's curators can bundle Preferred Deals, fixed-price and pre-auction by transaction-type definition, and Index's package creators can choose a fixed price auction, which Index defines as having priority over all other auction bid types, but neither documents a default for curated packages; Commerce Grid's 'Prioritizes high-value deals in auctions' describes deprioritising low-value curated deals in auction processing (request sampling); the other eleven document no priority or floor default. (Corrected in the final live sweep of 4 November 2026; see the corrections ledger.) | abstract, 4, 12, appendix | inference | supported with qualification | medium | [111, 250, 24, 266] |
| CUR-FINAL-F01 | Index's 'Create a Marketplace Package' page (last updated 26 October 2026) has the package's creator select a First or Fixed auction type for each package, states no default between them, and defines a fixed price auction as 'an auction that has priority over all other auction bid types'; it states no exclusivity default and no other rule ranking a package's bids, so Index documents a configurable route to priority rather than a default. | abstract, 4, 12, appendix | vendor assertion | supported as stated | high | [14] |
| CUR-D-F16 | Oracle's end-of-life FAQ (Version 1, 17 June 2024) states that Oracle will end all Oracle Advertising products and support effective 30 September 2024, listing Moat Analytics, Oracle Contextual Intelligence, the Oracle Marketing Cloud Data Management Platform, Data Append, Digital Audiences (including OnRamp), Offline Targeting, Moat Reach and Cross-Device services with that date, that Oracle will enter no new agreements effective immediately and will end relationships with data providers; this is a product retirement (market exit), not an acquisition or insolvency. | abstract, 5 | direct record | supported as stated | high | [34] |
| CUR-D-F18 | Nexxen International Ltd. (renamed from Tremor International in January 2024) cancelled its AIM depositary interests on 17 February 2025 and its shares have traded solely on the Nasdaq Global Market under NEXN since 18 February 2025; its FY2025 20-F names four beneficial owners above 5% (entities affiliated with Mithaq Capital SPC, JB Capital Partners L.P., News Corporation and Toscafund Asset Management LLP) holding approximately 52.3% of the shares in aggregate, warns that they may exert significant influence, and names no single majority holder; it describes platform fees based on a percentage of spend, flat fees or fixed CPMs, and refers to the Amobee acquisition in 2022 without a price in the portion read. | abstract, 5 | direct record | supported with qualification | high | [30] |
| CUR-D-F25 | On the strict H5 rule (control documented from filings or official releases), the curation data suppliers with an opened ownership record that were documented as independently controlled fell from four of eleven in 2023 (Lotame, LiveRamp, IAS, DoubleVerify; six unknown; Oracle not independent) to two of ten active suppliers at the cutoff (IAS under Novacap funds, DoubleVerify pending sale to Nielsen; three unknown; Oracle exited), because Lotame and LiveRamp passed to Publicis; on the looser reading that treats a private supplier with no documented operating-group parent as independent, the fall is from ten of eleven to five of ten, with Audigent (Experian), Lotame and LiveRamp (Publicis), Semasio (Samba TV) and Adloox (Scope3, then Peer39) leaving independence and Oracle leaving the market. | abstract, 5, appendix | synthesis | contradicted | medium | [49, 41, 38, 36, 44, 34, 50] |
| CUR-INDEXMARKETPLACES-F10 | Index's zero-fee statements cover Index's own charges only: 'Index does not charge data fees' and 'no hidden tolls or platform taxes' on the Data Vendor Ecosystem (7 October 2025), 'no incremental expense for partners using Marketplaces' (23 January 2024) and packages 'all without any added platform fees' (product page), while the same releases say every dollar flows to the data vendor and the publisher's media and the knowledge base documents a partner fee, a vendor fee and an Index revenue share in the deal path. | abstract, 6 | vendor assertion | supported as stated | high | [188, 187, 200] |
| CUR-H-F03 | In the ANA 2023 study's matched data, private-marketplace deals with more than 500 domains carried a $5.52 CPM against $2.75 for the open marketplace while delivering 19.8% MFA impressions against 26.8%, which the report describes as 'quality similar to the OMP deals, but at double the cost'; deals with 1-10 domains cost $6.03 and 11-500 domains $8.49; the report's own footnote says multi-publisher 'auction packages' 'rarely carry auction priority' and their value proposition is 'inventory curation'. | abstract, 6, 7, 12 | independent measurement | supported as stated | medium | [26] |
| CUR-H-F22 | TripleLift's TL Direct page (undated, footer 2026) states 'No TripleLift curation fees', 'Free 1P data onboarding and lookalike modeling', access to '3k+ off-shelf behavioral and contextual segments at no extra cost' and 'Flexible Pricing Models' (fixed-rate, floor-rate or hybrid deals), and says nothing about TripleLift's exchange take rate, margins, revenue share or data-partner fees. | abstract, 6, 11 | vendor assertion | supported as stated | high | [117] |
| CUR-H-F01 | In the ANA's 2023 Complete Report (21 marketers, $123 million, 35.5 billion impressions, September 2022 to January 2023), transaction costs, 'primarily DSP and SSP costs', took 29% of the ad dollar entering a DSP and loss of media productivity 35%, leaving 36 cents; on the report's $1,000 sequential waterfall the DSP platform cost is $80 (8%), the SSP platform cost $130 (13%), DSP data plus DSP additional costs together $80 (the retrospective table in the Q1 2025 report gives the 2023 medians as data 6.0% and additional 2.0%), and seller revenue $710 (71%). | abstract, 6, 12 | independent measurement | supported as stated | high | [26, 63] |
| CUR-H-F30 | In the sources screened by 3 November 2026, no exchange or curator publishes a rate card for curation fees or its exchange take on curated deals: the public record consists of fee types and stacking rules (Google, Microsoft, the Deal Sync API), zero-fee claims for one party's charge (Index data and platform fees, TripleLift curation fees), one segment price ($0.15 CPM, Adloox via Qwarry on PubMatic), a fee-type distribution from a client-only sample (Jounce, via reporting), corporate fee descriptions without rates in the FY2025 10-Ks of Magnite and PubMatic, a derived blended ratio for The Trade Desk, a 2020 Google statement of a 15% DV360 standard rate, and Amazon's statement of the fee basis without the rate. | abstract, 6, 12, appendix | inference | supported with qualification | medium | [212, 231, 281, 115, 24] |
| CUR-H-F31 | Within a single transaction perimeter the only documented stacking orders found are Microsoft's (buyer fees removed; curator margin and curator tech fee or MASC deducted; seller auction service charge deducted; remainder to the seller) and Google's (curator fee 'taken out of the bid price before it enters the auction', Google's own share unstated), while Jounce describes in its public State of the Open Internet 2025 how an exchange's fee is embedded in the declared floor and, via AdExchanger, the SSP accruing and paying the curator's fee out of a floor that already embeds it; no public source gives all four series (curation fee, data fee, SSP take, DSP take) with payer, collector, base and period for one named curated deal. | abstract, 6, 12, appendix | synthesis | supported with qualification | medium | [115, 216, 290, 64] |
| CUR-C-F17 | Across the 19 cohort A, B and C product rows screened in this stream (10 A, 8 B, 1 C) plus one listed comparator product (OpenXBuild), every quantified quality or performance claim found is a vendor assertion (one commissioned Forrester TEI study measures platform economics, not deal quality): 10 cohort rows carry a quantified claim (PubMatic Connect, Microsoft Curate, Maestro by Equativ, TL Direct, YMax, SmartPMP, Adelaide high-attention PMPs, Lotame Curated Marketplaces, Multilocal curation, Seedtag NeuroX Curation) and 9 carry none in the documents opened (ClearLine, Index Marketplaces, Nexxen Curated Marketplace, FreeWheel Curation Hub, Google Ad Manager Curation, Experian Curated Deals, Comscore-Certified Deal IDs, IAS Curation, Stagwell Curate); OpenXSelect was not screened, because the Hyundai figures in OpenX's release of 16 June 2026 attach to OpenXBuild custom decisioning; none of the claims names an independent measurer with a published method, a sample size or a measurement period, and the comparators are the same campaign on the open market (PubMatic), another supply path on the plan (OpenX), DSP-level data on open exchange inventory (Audigent), a rival targeting method on the same inventory (Yieldmo), 'standard media' or Adelaide's own benchmarks (Adelaide), a campaign target (PubMatic, Audigent/Alcon), an unsourced industry benchmark (Lotame/PHD) or none (Microsoft, Lotame launch, Multilocal), so no promise tested reaches the H3 'supported' or 'contradicted' rule and H3 stays unresolved on this stream's evidence. | abstract, 7, 10, 11, appendix | synthesis | supported with qualification | high | [119, 120, 121, 125] |
| CUR-J-F21 | Across the protocols and vendor interfaces screened, the highest state reached on evidence from a party other than the author is documented production use reported by the vendors and the trade press relaying them: PubMatic claims 30 live end-to-end AgenticOS campaigns by April 2026 and names customers in several releases, and Magnite names one buy (H/L for the Northern California Toyota Dealers Association on LG's home screen, October 2026); of the accounts read, only PubMatic's release of 9 June 2026 states an executed selection decision, that agents 'working across the platform' 'autonomously selected CTV inventory' for Havas's Movistar campaign in Spain, while the Geloso (December 2025) and LG accounts document agent-led activation without saying what the agents selected; AAMP has reference connectors and a sandbox pilot, OpenProposal is a draft, the Prebid Sales Agent is an alpha connector with no named deployment, OpenX/Chalice names a customer (Hyundai) with no period, and InMobi, Stagwell, Kargo, Apostra and Microsoft are announced, beta or unnamed; so H7 is 'supported with qualification' on one platform-level, interested-party record that names no specific agent, and 'unresolved' if the registered rule is read to require evidence from a party without a stake. (Corrected in the final live sweep of 4 November 2026; see the corrections ledger.) | abstract, 8, 10, 12, appendix | synthesis | supported with qualification | medium | [93, 120, 172, 168, 171] |
| CUR-FINAL-F04 | PubMatic's AgenticOS launch release of 5 January 2026 describes the December 2025 Butler/Till campaign for Geloso Beverage Group's Clubtails as 'an early agentic campaign on AgenticOS' in which, 'Guided by natural-language input through Claude', AgenticOS 'autonomously recommended campaign tactics, executed the buy, and optimized performance in real time within predefined parameters'; it does not say which inventory was selected, by which agent or by which rule. | abstract, 10 | vendor assertion | supported as stated | high | [312] |
| CUR-J-F15 | On 14 October 2026 LG Ad Solutions and Magnite announced a three-year extension under which 'LG Ad Solutions is making its Home Screen inventory available for agentic buying through Magnite's Seller Agent', with independent agency H/L as 'the first buyer to activate LG Home Screen inventory' for the Northern California Toyota Dealers Association in a campaign timed to the fall football season; the release names no protocol and does not say what the seller agent decides, while AdExchanger's same-day report adds that H/L's buyer agent and Magnite's seller agent set up and activated the brief inside ClearLine over MCP and that LG and H/L approved the agent-led plan before it went live, and Magnite's own blog of 6 January 2026 describes the seller agent as supporting AdCP tests without stating a version. | abstract, 10, 12 | vendor assertion | supported with qualification | medium | [172] |
| CUR-FINAL-F02 | PubMatic's release of 9 June 2026 (Madrid) announces its first agentic campaign in Spain, with Havas for Movistar, part of Telefónica Group, executed on AgenticOS through Activate, 'PubMatic’s direct-to-supply media buying platform'; it says agents 'working across the platform' took a natural-language brief processed through Claude and 'autonomously selected CTV inventory, applied frequency controls and brand safety parameters, and optimised bids in real-time' within the client's guardrails, on in-stream CTV with 'direct access to Spain's premium CTV publishers'; it reports CPM 18% below target and impressions 23% above target, gives no campaign dates, names no specific agent among AgenticOS's 'more than 20 agents', quotes Havas, Telefónica and PubMatic, and cites no independent measurement. | abstract, 10, 12, appendix | vendor assertion | supported as stated | medium | [313] |
| CUR-FINAL-F05 | Of the 21 scored products, 7 carry a band; of the 122 counted cells, 92 are graded D (vendor marketing, press release or case study), 26 C, 2 B and 2 A; all 7 banded rows change band under at least one missing-evidence scenario; in four banded exchange rows (Commerce Grid, Maestro by Equativ, Microsoft Curate and TL Direct) pricing transparency is the only counted Standing cell, so it carries the group's 30% weight alone; counting only cells graded C or better, four products clear the four-cell admission rule (Maestro by Equativ, Google Ad Manager Curation, Index Marketplaces and Microsoft Curate) and no cohort carries bands; three admitted rows rest on unconfirmed availability (Google Ad Manager Curation in Beta, Stagwell Curate live on its operator's claim, WPP Media Premium Marketplace with no page newer than 9 February 2023), and without the Beta Google row the exchange field DQ falls from 46.7 to 45.4 and Microsoft Curate moves from Mid to Strong. | abstract, 11 | synthesis | supported as stated | high | [314] |

### The corrections

Every substantive change after the verification passes, with what questioned it, what was checked, the verdict, the change and how it was verified. It includes the brief's fact-check, the outside review, every verifier correction applied to a claim, the register amendments, the scoring corrections and the final live sweep of the published edition.

**Corrections ledger**

| ID | Original claim or score | Challenge | Evidence checked | Verdict | Change made | Reason | Affected |
|---|---|---|---|---|---|---|---|
| F-C1-01 | Jounce: curated deal IDs 73% of bidding opportunities, single-publisher 8.7%, open auction 18.6%, spring 2025 | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Jounce's 2025 'The State Of The Open Internet' report (p.31) shows a chart 'Mix Of Available RTB Auctions': Curation 72.7% (labelled 'Exchange-controlled demand & curator-controlled demand'), Open Auction 18.6% ('DSP-controlled demand'), Private Marketplaces 8.7% ('Publisher-controlled demand'), captioned '4 of every 5 auctions is an attempt by the sell-side to control demand'. Jounce stresses this is a share of auctions/ad opportunities presented to DSPs, NOT of spend: 'DSP-controlled demand is far more than 18.6% of programmatic ad spend.' The chart is not dated on the page; the report's other charts use March 2025 data. AdExchanger (Anthony Vargas, 1 May 2025) gives the sample as 'In March, ... Jounce Media analyzed 1.5 billion bid requests sent from 44 ad exchanges' and reports 73% multi-seller curated deals, 8.7% single-seller deals, 18.6% pure open auction, of 'bidding opportunitie | supported with qualification | 72.7%; share of available RTB auctions presented to DSPs, March 2025, 1.5 billion requests from 44 exchanges, global; not spend; no 2026 edition | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C1-02 | "Curators cannot mint new bid requests" (Chris Kane) | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | The quoted wording was not found in: the AdExchanger 1 May 2025 article (Kane is quoted on fee structures and static-price deals only), Jounce's 2025 State of the Open Internet (zero hits for 'mint'), Playwire's 10 Apr 2025 summary, or web searches. The closest primary passage is SOI 2025 p.30, which describes curators as decorating existing bid requests rather than creating supply. | unresolved | wording not found in Jounce's publication; treated as the essay's paraphrase | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C1-03 | ANA PMP share 64.5% (first period), 87.8% (Q2 2025), above 92% (Q4 2025), CTV 100% | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | All three figures are MEDIANS across contributing advertisers, not aggregate shares, and 64.5% is NOT the first period. Series (median PMP share \| aggregate PMP share \| period \| contributing/participating marketers \| spend \| impressions): 2023 study 54.2% \| 40.9% \| Sep 2022-Jan 2023 \| 21/21 \| $123M \| 35.5bn. Dec 2024 report (first benchmark period) 56.2% \| 65.9% \| Apr-Sep 2024 \| 21/37 \| $235M \| 38.5bn. Q1 2025 report (21 May 2025) 64.5% \| 67.2% \| 1 Nov 2024-31 Mar 2025 \| 23/39 \| $242M \| 41.9bn; CTV 99.5% via deal IDs. Q2 2025 (14 Aug 2025) 87.8% \| 82.6% \| Apr-Jun 2025 \| 21/39 \| $146M \| 18.6bn; CTV median 100%, aggregate 99.5%. Q3 2025 (5 Nov 2025) 88.8% \| 81.6% \| Jul-Sep 2025 \| 21/39 \| $142M \| 17.5bn; CTV 100%/99.2%. Q4 2025 (25 Feb 2026 release) 92.2% \| 79.2% \| Oct-Dec 2025 \| 35/54 \| $142M \| 16.7bn; CTV 100% both; web+mobile median 87.4%, aggregate 64.9%. The Q2 2025 press release's 'PM | supported with qualification | medians across contributing advertisers; 64.5% is the Q1 2025 report; aggregate fell from 82.6% to 79.2% over Q2 to Q4 2025 | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C1-04 | ANA PMP CPM $7.15 against $4.41 open marketplace | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Q2 2025 press release (14 Aug 2025): 'PMP CPMs averaging $7.15 vs. $4.41 for OMP.' In the report tables these are the Q2 2025 MEDIAN CPMs across contributing advertisers (Q2 2025 8-page PDF median rows: OMP CPM $3.28 -> $4.41, PMP CPM $9.37 -> $7.15; the Q1 2026 report's quarterly table also lists Q2 2025 CPM PMP $7.15 / CPM OMP $4.41). The aggregate (spend-weighted) Q2 2025 CPMs in the Q3 2025 report's table are PMP $7.43 vs OMP $3.64. Later quarters (aggregate): Q3 2025 PMP $9.88 / OMP $3.55; Q4 2025 PMP $10.72 / OMP $4.25; Q1 2026 report table PMP $7.77 / OMP $3.58 (total CPM $4.42). | supported with qualification | Q2 2025 medians (21 advertisers); aggregate $7.43 against $3.64 | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C1-05 | ANA's definition of private marketplace | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | None of the opened benchmark PDFs (Q1, Q2, Q3, Q4 2025; Q1 2026) defines PMP or OMP in-text; each refers readers to a separate 'Benchmark Methodology and Glossary' document, which was not opened. The Q1 2025 report's CTV section equates PMP with deal-ID transactions: '99.5 percent of the CTV ad spending was executed via Deal IDs in private marketplaces (PMPs)'. No report separates curated multi-seller deals or programmatic guaranteed from PMP; the Q2 2025 report once writes 'As PMPs and Programmatic Direct continue to grow'. The reports treat the PMP share itself as the measure of 'curation' (Q3 2025: 'Curation is reshaping the landscape. A massive 81.6 percent of all spend now flows through Private Marketplaces'). | unresolved | no in-text definition; equated with deals executed via deal IDs; glossary document not opened | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C1-06 | ANA participants from 21 (2023) to 86 (Q1 2026) | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Q1 2026 press release (NEW YORK, May 27, 2026) and the Q1 2026 report: 'Participating Marketers increased from 54 to 86 and active data contributors to 66, with impressions and spend reaching 20.9B impressions and $160M in ad spending'. Footnote: 'Participating Marketers are the ones who have requested their suppliers to get access to their LLD and Active Participants are the ones who have been granted access to their LLD with the permission to use it for the Benchmark by their suppliers.' The 2023 study had 21 marketers, all of whom contributed data. | supported with qualification | 86 participating, 66 contributing data; like for like is 21 to 66 | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C1-09 | Two ANA MFA figures: $13 billion and $10 billion a year | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Both are ANA figures from different documents. $13 billion = First Look (June 2023): 15% of the $88 billion global open-web programmatic market (Jounce 'State of the Open Internet 2023' estimate). $10.1 billion = Complete Report (December 2023) re-calculation, footnote 2 (p.13) and footnote 10 (p.33): 'The dollar estimate for MFA spend has been re-calculated since the First Look study (reported there as $13 billion). Deepsee.io, who we used to identify MFA sites, only provides data on web domains. MFA spend is now calculated to be approximately $10.1 billion -- 15 percent of $67.3 billion. That's the $88 billion global market for open web programmatic advertising minus $20.3 billion for CTV (which does not have MFA activity).' The Complete Report's $22 billion opportunity is itemised as 'Approximately $10 billion from a re-evaluation of Made for Advertising websites' plus 'Approximately | supported with qualification | $13 billion = First Look, 19 June 2023 (15% of $88 billion); $10.1 billion = Complete Report, 5 December 2023 (15% of $67.3 billion); both extrapolations | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C1-10 | ANA waste figures ($20 billion, $22 billion, $88 billion base) | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | First Look (June 2023, as restated in the Complete Report): 'Open web programmatic advertising is estimated to be an $88 billion global market... we are confident that there will be opportunities to drive a total of at least $20 billion in efficiency gains for open web programmatic advertising.' Complete Report (Dec 2023): 'It can now be confirmed that there is the opportunity for $22 billion in efficiency gains' ($10bn MFA + $12bn other recommendations); 'of the $88 billion in open web spending, some $22 billion...'; 'only 36 cents of every ad dollar that enters a DSP effectively reaches the consumer' (transaction costs 29%, loss of media productivity 35%). The benchmark later re-based the figure: $21.6bn (Q1 2025, 20.8% of a $104bn 2024 market), $26.8bn (Q2 and Q3 2025, 21.8% median TrueCPM opportunity of a $123bn 2025 market, Madison and Wall estimate). | supported with qualification | "at least $20 billion" (First Look); $22 billion of $88 billion (Complete Report); benchmark re-based to $21.6 billion of $104 billion and $26.8 billion of $123 billion | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C1-12 | Contextual paper: curation a $1B business, $5B in two to three years, median fees 11 to 14% | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | No primary or trade source found. Not in the ANA Q2 2025, Q3 2025, Q4 2025 or Q1 2026 benchmark reports (their 'Curation' sections contain no market-size or fee-median figures; the Q2 2025 PDF has no '$1 billion'/'$5 billion'/'fee' text). Not in Jounce's 2025 State of the Open Internet. Not in AdExchanger (1 May 2025), Playwire (10 Apr 2025), ppc.land, Digiday's 24 Mar 2025 curation piece, or five web searches. The only fee median in the public record is Jounce's 14% median margin-based fee (see C1-13); no 11% figure surfaced anywhere. | unresolved | no source located in the ANA reports, Jounce's publication or trade reporting; not to be reused | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C1-13 | About half of curated deal IDs carry no fee beyond exchange take | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Jounce Media's April 2025 Supply Path Benchmarking Report (client-only), as summarised by Playwire (10 Apr 2025): 'Zero Fee Curation (50% of bid requests)' carrying only the exchange's standard open-auction take rate; margin-based fees 34% (median 14% take rate); fixed CPM fees 10% (median $0.48); static price deals 7%; 'median effective margin is 30%, and median effective CPM is $0.51'. AdExchanger (1 May 2025): 'Roughly half of all curated deal IDs carry no curation-related fees, according to Jounce'; 'Around 34% of curated deals carry margin-based fees... the median margin-based fee is 14%'; 'nearly 40% of margin-based deals have between a 5% and 10% fee'; sample 'In March... 1.5 billion bid requests sent from 44 ad exchanges'. | supported with qualification | Jounce April 2025 report via Playwire and AdExchanger: 50% zero fee, 34% margin-based (median 14%), 10% fixed CPM (median $0.48), 7% static price; reporting | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C1-14 | ISBA/PwC programmatic supply chain study, 2020 | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | ISBA article 'Time for change and transparency in programmatic advertising', dated 6 May 2020: the 'ISBA Programmatic Supply Chain Transparency Study', published by ISBA in association with the Association of Online Publishers (AOP), work carried out by PwC. Sample: 15 advertisers, 8 agencies, 5 DSPs, 6 SSPs, 12 publishers, approximately GBP 0.1 billion of UK programmatic media spend. Findings: publishers receive 51% of advertiser spend on average; 15% of advertiser spend is an 'unknown delta' (about one-third of supply-chain costs) that could not be attributed. The hub page /knowledge/programmatic-supply-chain-transparency-study returned 404 today (www); the '?page=1' variant and an executive summary at /node/654 are indexed but were not opened. | supported with qualification | 6 May 2020; ISBA with AOP, carried out by PwC; 15 advertisers, about £0.1 billion; 15% unknown delta | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C1-15 | ISBA/PwC second study, 2022 | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | ISBA knowledge page 'Second Programmatic Supply Chain Transparency Study', published 18 January 2023, 'conducted by ISBA and PwC' (AOP's Richard Reeves quoted, not credited as author). Findings: impression match rate 58% (vs 12% in 2020); unknown delta 3% (vs 15%); advertiser spend reaching publishers up 8%; study took nine months; results 'should not be taken as representative of the broader programmatic ecosystem'. Links a summary PDF 'ISBA PwC programmatic supply chain study II (summary) - 18 January 2023'; the full report is not linked. | supported with qualification | published 18 January 2023 on 2022 fieldwork; summary only; unknown delta 3% | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C1-16 | WFA/MediaSense supply chain study, 2024 | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | No such report found. WFA's 2024 item is 'Spotlight: ANA Programmatic Media Supply Chain Study' (16 February 2024), a members-only video in which the ANA study's authors discuss their findings; WFA is listed as author of the spotlight only. WFA's knowledge hub now redirects to a gated Stravito portal. WFA also lists 'Webinar+: Programmatic supply-chain transparency' (11 June 2020, members-only) following the ISBA/PwC study. MediaSense has a WFA partnership on media transformation; no programmatic supply-chain report co-authored by them surfaced in two searches. | contradicted | no such study; the WFA's 2024 item is a members-only spotlight on the ANA study | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C1-17 | 4A's transparency or MFA work, Q1 2024 | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Nothing matching a Q1 2024 4A's report was found on aaaa.org. The 4A's own MFA publications are its copy of the joint definition 'Leading Trade Groups Define "Made For Advertising" Websites' (Published 09.26.2023), which notes 'The 4As also refers to MFA sites as "Made for Arbitrage" sites', and reposts of trade articles (Digiday's MFA oral history; Adweek's 'new parameters' piece). AdExchanger reported (2024) that the 4A's hosts monthly buyer/seller meetings to define MFA criteria. The only 2026 aaaa.org programmatic item is an event (08.11.2026) on the first independent agency adopting the ANA Benchmark/TrueKPI framework. | contradicted | none found; the 4A's MFA output is the joint definition of 26 September 2023 | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C2-02 | Lotame acquired by Publicis | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Publicis Groupe release datelined Paris, 03/06/2025, 'Publicis to acquire Lotame': 'Lotame will be positioned as part of Epsilon, accessible to all Publicis clients and teams'; 'This acquisition remains subject to customary conditions to closing' and 'is expected to close early in the second quarter of 2025'; no terms disclosed. No Publicis release stating the closing date was found. Publicis's H1 2025 results release (16 July 2025) counts Lotame among 'Acquisitions (Mars, Influential, BR Media, Lotame, ...)' contributing net revenue in H1 2025, i.e. closed by 30 June 2025. lotame.com (2026) carries an 'Epsilon Lotame' logo lockup and the footer '©2026 Lotame Solutions, Inc.' | supported with qualification | announced 6 March 2025; counted among H1 2025 acquisitions, so closed by 30 June 2025; now Epsilon Lotame | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C2-03 | LiveRamp to Publicis at a $2.167 billion enterprise value, closing to verify | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | The transaction exists and has CLOSED. Publicis Groupe release (Paris, 05/17/2026): acquisition 'for a total enterprise value of $2.167 billion in an all-cash transaction', 'based on an acquisition price of $38.50 per share', 'a total equity value of $2.546 billion', 'a 29.8% premium to LiveRamp's closing share price on May 15th, 2026', 'expected to close before year-end 2026'. LiveRamp 8-K (Item 5.07, event 17 Aug 2026): stockholders adopted the Merger Agreement dated 16 May 2026 (51,578,202 for / 60,073 against / 53,553 abstain). LiveRamp 8-K filed 2 Nov 2026 (Items 2.01, 3.01, 3.03, 5.01, 5.02, 5.03): 'On November 2, 2026 (the "Closing Date"), MMS USA Holdings, Inc., a Delaware corporation ("Parent"), completed the previously announced acquisition of LiveRamp Holdings, Inc.'; each share 'converted into the right to receive $38.50 in cash, without interest'; shares removed from trading | supported with qualification | announced 17 May 2026 ($38.50 a share); approved 17 August 2026; completed 2 November 2026 by MMS USA Holdings, Inc.; delisted (8-K) | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C2-05 | Oracle Advertising, including Grapeshot and Moat, closed 30 September 2024 | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Oracle's own 'Oracle Advertising EOL Frequently Asked Questions' (August 12, 2024, Version 3.0, oracle.com PDF): 'Oracle will End-Of-Life (EOL) all Oracle Advertising products and support effective September 30, 2024. This applies to the following product families, and supporting integrations:' Product families listed, each with EOL date September 30, 2024: Moat Analytics; Oracle Contextual Intelligence; Oracle Marketing Cloud Data Management Platform; Data Append, Data Enrichment & Optimization; Digital Audiences (including OnRamp); Offline Targeting/Direct Mail; Moat Reach; Cross-Device services. 'End-Of-Life notifications were issued via email week of June 17'. 'Effective immediately, Oracle will not enter into any new agreements with customers for Oracle Advertising products and services.' The name 'Grapeshot' does not appear; the contextual product is named 'Oracle Contextual Intell | supported with qualification | all Oracle Advertising products ended that day (FAQ v3.0, 12 August 2024); Grapeshot appears as Oracle Contextual Intelligence | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C2-06 | Scope3 rebranded Apostra on 23 September 2026 as an agentic media platform on AdCP | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Apostra blog post 'Introducing Apostra' dated September 23, 2026: 'Scope3 is becoming Apostra.' Self-description: 'Apostra gives advertisers, media companies, and their agents one place to buy and sell every kind of advertising, anywhere in the world.' and 'It's built on AdCP, the open protocol we helped create that lets agents talk to each other.' Neither the post nor apostra.com uses the phrase 'agentic media platform' (homepage: 'Advertising is changing. What matters shouldn't.' / 'Apostra gives advertisers and their agents direct access to everywhere brands can reach consumers.'). Sustainability: the Apostra site links 'Looking for Sustainability → Scope3 by Apostra'; scope3.apostra.com/about carries the 'Scope3 by Apostra' lockup and 'Copyright © 2021-2026 Scope3, PBC.' Footer on apostra.com: 'Copyright © 2026 Apostra, Inc.' | supported with qualification | "Scope3 is becoming Apostra", built on AdCP; "agentic media platform" is not Apostra's phrase; Scope3 by Apostra continues | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C2-07 | Xandr Curate launched April 2021; Xandr platforms owned by Microsoft | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | (a) Xandr Curate: Business Wire release (ID 20210426005974; Business Wire blocked, read via MarTech Series republication dated April 27, 2021): 'Xandr, AT&T's advanced advertising company, has today announced the launch of Xandr Curate', 'a platform offering data-layered curated deals across premium multi-seller inventory in Xandr's marketplace.' (b) Microsoft Advertising blog, December 20, 2021 (Rik van der Kooi): 'Earlier today, Microsoft announced that we entered into an agreement to acquire Xandr from AT&T.' Microsoft 2022 Annual Report: 'On June 6, 2022, we acquired Xandr, Inc., a technology platform with tools to accelerate the delivery of our digital advertising solutions.' (c) Today's name: Microsoft Advertising solutions page H1 'Microsoft Curate' ('Create curated deals to sell programmatically.'; sign-in at curate.xandr.com; deals 'available via all major demand-side platforms | supported with qualification | launch 27 April 2021; Microsoft completed 6 June 2022; now Microsoft Curate; Invest media buying ended 28 February 2026 per the 14 May 2025 notice | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C2-08 | Equativ and Sharethrough merged in 2024 | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Equativ release datelined 'New York, Paris, London, Montreal, June 12, 2024': 'Equativ, the global independent ad tech company, today announces its merger with Sharethrough, one of the top independent omnichannel ad exchanges.' It names no acquirer, no regulatory condition and no closing date, and refers to 'the unified entity'/'the combined entity'. Equativ release, 'NEW YORK, June 9, 2025': announces 'the full unification of Sharethrough into its global brand and operations' ... 'following Equativ's acquisition of Sharethrough in June 2024'. Equativ blog, July 7, 2025: 'Our merger with Sharethrough is now complete, and as we come together as one team united under the Equativ brand'. Surviving brand: Equativ. No Equativ release giving a 2024 closing date was found. | supported with qualification | announced 12 June 2024; no closing release found; Sharethrough brand retired 9 June 2025 | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C2-10 | Verve operates supply-side and curation products | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Verve release dated Sep 18, 2024 names two exchanges: 'a rebranding of Verve's Smaato ad exchange (now Brand+ Marketplace)' and 'Verve's PubNative exchange (now Performance+ Marketplace)'; Brand+ 'equips advertisers with audience curation tools they need to activate ID and ID-less targeting'. Verve's 2026 Brand+ page ('Brand+ Marketplace (formerly Smaato)', footer '© 2026 Verve Group, Inc.') lists curation as features, not a separately named product: 'Intelligent curation', 'AI-powered deal curation that helps unlock undervalued supply and audiences', 'Curate smarter deals', and a 'Deals Library' ('Activate curated, audience-enriched deals.'). Verve's April 21, 2025 Audigent release calls Verve 'the leading mobile app SSP in North America' and refers to 'Audigent's curation solutions', naming no Verve curation product. | supported with qualification | Brand+ and Performance+ marketplaces (formerly Smaato and PubNative); curation as features, no separately named product | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C3-03 | Remedies record | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Yes: Order of 2 Sept 2026 (Dkt. 1857) and 106-page Memorandum Opinion of 2 Sept 2026 (Dkt. 1858), filed under seal and UNSEALED by order of 16 Sept 2026 (Dkt. 1860). REJECTED: divestiture of AdX, open-sourcing of DFP's final auction logic, contingent divestiture of 'DFP Remainder' ('structural remedies are neither realistic nor needed'). ACCEPTED, as modified: injunction against enforcing any policies or contract terms tying DFP to AdX; Google barred from reimplementing First Look and Last Look and required to deprecate and not reimplement Unified Pricing Rules in DFP for all indirect transaction types; publishers may set different floor prices across buying tools and exchanges and Google may not discount DFP fees for uniform floors; AdX real-time bids made available to rival publisher ad servers and to Prebid for all indirect transactions; DFP-Prebid integration including routing of pro | supported with qualification | remedies memorandum opinion 2 September 2026, Dkt 1858; divestiture rejected; behavioural remedies accepted as modified; not a judgment | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C3-04 | Final judgment | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | No final judgment on the docket. Last known filing is Dkt. 1861 of 2 Oct 2026: 'Joint NOTICE of Parties' Proposed Final Judgment' with attachment 'Proposed Order Proposed Final Judgment' (41 pages). The proposal keeps bracketed, proponent-labelled disputes, e.g. AdX-Prebid and AdX-publisher-ad-server interoperability due '[Plaintiffs: 6] [Google: 12] months from the Effective Date'; Section IV.B: 'will expire on the sixth anniversary of the Effective Date'; Effective Date = 60 days after entry. Section VI.I: Google 'must not reimplement either First Look or Last Look'; VI.J: Google 'must deprecate and must not reimplement Unified Pricing Rules in DFP for Indirect Demand'; VI.H: no policies or contract terms tying DFP to AdX real-time bids. | supported with qualification | none entered; joint proposed final judgment of 2 October 2026 with bracketed disputes | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C3-05 | Appeal | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | No notice of appeal appears on the district docket through its 29 Oct 2026 update; no final judgment has been entered from which to appeal. The proposed final judgment anticipates appeal: Effective Date deadlines are tolled while 'stayed pursuant to an order of this Court, the United States Court of Appeals for the Fourth Circuit, or the United States Supreme Court', and Google's bracketed text makes effectiveness '[s]ubject to the outcome of any motion to stay this Final Judgment pending appeal'. Reporting (NatLawReview, 18 Sept 2026) says 'appeals are likely' and the Open Markets Institute has urged plaintiffs to appeal. | supported with qualification | none noticed by 29 October 2026; nothing to appeal yet | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C3-07 | Other private suits citing the judgment | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Verified from court records or the plaintiff's own release: (1) OpenX Technologies, Inc. and OpenX Ltd. v. Google LLC, E.D. Va. No. 1:25-cv-01282, complaint filed 4 Aug 2025 (caption 'Document 1 Filed 08/04/25', 87 pages, jury demanded; copy hosted by PPC Land). (2) PubMatic v. Google, E.D. Va., filed 8 Sept 2025 (PubMatic 8-K Ex. 99.1: 'today filed a lawsuit in the U.S. District Court for the Eastern District of Virginia'; case number 1:25-cv-1482 is reporting-only). (3) Magnite v. Google, E.D. Va., announced 16 Sept 2025 (Magnite 8-K Ex. 99.1: 'filed a lawsuit in the U.S. District Court for the Eastern District of Virginia'). (4) Equativ SAS, Equativ Inc., Sharethrough Inc. and Sharethrough USA v. Google LLC: complaint filed in E.D. Va. 13 Oct 2025, transferred to S.D.N.Y. 8 Jan 2026 as No. 1:26-cv-00140 before Judge P. Kevin Castel (docket). (5) Index Exchange Inc. v. Google LLC, E.D. | supported with qualification | OpenX 4 August 2025; PubMatic 8 September 2025; Magnite 16 September 2025; Equativ 13 October 2025 (to S.D.N.Y. 8 January 2026); Index Exchange 10 November 2025, terminated 18 December 2025; they cite the liability opinion, not a judgment | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C3-09 | Later Google cookie statement | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | 'Update on Plans for Privacy Sandbox Technologies', Anthony Chavez, 17 Oct 2025: Google 'decided to retire the following Privacy Sandbox technologies': Attribution Reporting API, IP Protection, On-Device Personalization, Private Aggregation (incl. Shared Storage), Protected Audience, Protected App Signals, Related Website Sets, SelectURL, SDK Runtime, Topics; it will 'continue to support' CHIPS and FedCM and 'maintain Private State Tokens'. On cookies it restates the April position: 'we will maintain our current approach to offering users third-party cookie choice in Chrome.' No mention of a standalone prompt. | supported with qualification | 17 October 2025 retires most Privacy Sandbox APIs; cookie position restated unchanged | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C3-11 | UK CMA designation and ad-tech requirements | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | CMA case page 'Google's general search and search advertising services': SMS investigation launched 14 Jan 2025; final designation decision published 10 Oct 2025 covering general search and search advertising; conduct-requirement consultation 28 Jan 2026 (data portability, publisher, fair ranking, user choice); publisher CR imposed 3 June 2026; fair ranking and data portability CRs imposed 17 June 2026; further consultation on a user-choice CR launched 23 Sept 2026 (latest entry). No ad-tech, open-display, ad-exchange or deal-mechanics requirement appears. | supported with qualification | strategic market status of 10 October 2025 covers search only; ad-tech case 51145 still open | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C3-12 | 2026 EC or CMA step on exchanges or deal mechanics | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | EC: the only 2026 entries on the AT.40670 case page are publication of the provisional non-confidential decision on 15 Jan 2026 and (via CURIA) the OJ notice of Google's annulment action on 12 Jan 2026; no 2026 Commission decision on Google's November 2025 proposal or on structural/behavioural remedies was found. CMA: the ad-tech investigation case page (updated 19 May 2026) records no decision, commitments or closure; the SMS conduct requirements of June/Sept 2026 are search-only. | unresolved | none found beyond publication of the decision on 15 January 2026 | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C4-05 | Current OpenRTB 2.x version | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | GitHub releases (API, published_at): 2.6-202606 on 2026-06-11 (latest, 'Programming Liveness'); 2.6-202505 on 2025-05-20; 2.6-202501 on 2025-01-27; 2.6-202409 on 2024-09-23. The spec text: version number only increments on breaking changes since 2.6-202211. | supported with qualification | 2.6-202606 of 11 June 2026; the 2025 revisions touched neither Pmp nor Deal | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C4-06 | OpenRTB 3.0 or AdCOM deal object | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | OpenRTB Specification v3.0 (March 2022) has 'Object: Deal' under Item with id (string; required), flr (float), flrcur (string; default USD), at (integer), wseat (string array), wadomain (string array), ext. Item also carries a `private` attribute. The Deal object lives in the OpenRTB 3.0 spec, not in AdCOM. | supported with qualification | Deal object sits in OpenRTB 3.0 (March 2022), not AdCOM | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C4-07 | Other Deal fields relevant to curation | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Object: Deal (2.6-202606) = id, bidfloor, bidfloorcur, at, wseat, wadomain, guar (integer, default 0), mincpmpersec (float), durfloors (object array of DurFloors, Section 3.2.35), ext ('Placeholder for exchange-specific extensions to OpenRTB.'). No curator, packager or fee attribute exists anywhere in 2.6; 'curated' appears only as 'exchange curated string names' (Metric.type/vendor, Device.carrier) and in the Channel definition. | supported with qualification | bidfloorcur, wadomain, guar, mincpmpersec, durfloors; no curator or fee field | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C4-08 | ads.txt current version | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Spec version table: '1.1 \| April 2022 public comment July 2022 final release \| Added the "managerdomain" and "ownerdomain" variables'. iabtechlab.com/standards/ads-txt/ (page 'Last updated: September 14, 2026') lists ads.txt 1.1 in its Version History with an August 2022 date and the banner 'Ads.txt/App-Ads.txt Version 1.1 To Support Owner and Manager Domains is Ready to Implement'. app-ads.txt is 1.0. | supported with qualification | 1.1; final July 2022 per the spec, August 2022 per the Tech Lab page | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C4-10 | SupplyChain object current version | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | 'OpenRTB SupplyChain object' in the openrtb repo defines `ver` (major.minor) with 1.0 as the value; the document carries no date of its own. The Tech Lab sellers.json page dates both sellers.json and SupplyChain to July 31, 2019. | supported with qualification | 1.0; document undated; 2019 per the Tech Lab page | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C4-12 | IAB Tech Lab curation guidance | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | (1) Deals API v1.0 ('Deal Sync API Specification', repo IABTechLab/deal-api, file deal1.0.md): press release Dec 4, 2025 'Open for Public Comment Through January 31, 2026'; standards page https://iabtechlab.com/standards/dealsapi/ says 'Last updated: February 6, 2026' and links the spec as 'Final'. It defines Object: Curation with `curator` (string), `cdealid` (string), `curfeetype` (int: 0 undisclosed, 1 percentage of spend, 2 flat fee, 3 CPM, 4 no fee), plus Deal.`seller`, `origin`, `auxdata`; section 'Origin, Curator, and Seller'. (2) Blog 'Curated Audiences and launching the Curation Framework Working Group' (Hillary Slattery, Dec 16, 2024): rebrands Seller Defined Audiences as Curated Audiences and announces a Curation Framework Subgroup / 2025 Curation Workstream under the Addressability WG; no spec published. (3) 2025 articles/webinar: 'Separating the Truth from the Hype in Curati | supported with qualification | Deals API 1.0, final 6 February 2026, with a Curation object (curator, cdealid, curfeetype); no standalone curation standard | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C4-13 | Prebid curation guidance | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | A site:prebid.org search for 'curation OR curated' returned no prebid.org pages (only third-party vendor docs, e.g. Permutive's guide to routing cohort signals through Prebid.js RTD modules to SSP curation marketplaces). docs.prebid.org/agents/salesagent.html does not mention curation or curator. | unresolved | none surfaced; absence weakly bounded | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C4-14 | Prebid Sales Agent implements AdCP 3.1.1, beta, repository state 27 September 2026 | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Repo github.com/prebid/salesagent (description: 'A media sales agent that implements the AdCP Media Buy protocol'; default branch main; 39 stars; no GitHub releases). README: '> **Status: alpha.** A pre-1.0 implementation of a pre-release protocol (AdCP 3.1.1), under active development' and 'This implementation targets **AdCP spec version 3.1.1** via the `adcp==6.6.0`' Python SDK; later 'Because this is alpha software tracking a beta protocol, pin the version you deploy'. pyproject.toml: name 'adcp-sales-agent', version '2.0.0', dependency 'adcp==6.6.0'. Latest commit on main: 1c202df2, 2026-10-28T17:46:21Z ('chore(deps): bump uv, pillow, certifi and virtualenv past their advisories (#2458)'). | supported with qualification | AdCP 3.1.1 correct; software is alpha; latest commit 28 October 2026 | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C4-15 | AdCP versions | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | GitHub releases (published_at): v3.3.0-beta.3 2026-10-31 (pre-release); v3.3.0-beta.2 2026-10-30 (pre-release); v3.1.31 2026-10-30; v3.3.0-beta.1 2026-10-29 (pre-release); v3.2.8 2026-10-28 (marked Latest); v3.1.30 2026-10-28; v3.2.7 2026-10-24; v3.2.6 2026-10-23; v3.3.0-beta.0 2026-10-22 (pre-release); v3.2.5 2026-10-16; v3.1.29 2026-10-16. adcontextprotocol.org: 'AdCP 3.2 is now generally available'. docs.adcontextprotocol.org states release 3.2.1 with wire pin '3.2'. | supported with qualification | 3.2.8 latest (28 October 2026); 3.1.31 (30 October); 3.3.0-beta.3 (31 October) | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C4-16 | AdCP agent types | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | README.md frames AdCP as protocol SURFACES, not agent types: Negotiation layer = Media Buy, Creative, Signals, Accounts, Governance, Brand, Sponsored Intelligence; Decisioning and serving layer = Trusted Match Protocol (Context Match, Identity Match); plus a 'Curation \| Media inventory curation \| Coming soon' row. docs.adcontextprotocol.org uses agent nouns: 'Media buying agents', 'Creative agents', 'Signals agents', 'Governance agents', 'Orchestrators', 'buyer agent', 'creative agent', and 'sales-agent endpoints' (brand.json). The homepage names no agent types. | supported with qualification | protocol surfaces including Media Buy and Signals; "Curation" listed as coming soon | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C4-18 | MRC viewability guidelines | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | MRC Standards and Guidelines page lists: 'Viewable Ad Impression Measurement Guidelines' issued Jun 2014, updated Aug 2015 (file: 081815 Viewable Ad Impression Guideline_v2.0_Final.pdf); 'Mobile Viewable Ad Impression Measurement Guidelines' issued Jun 2016 (file: 062816 Mobile Viewable Guidelines Final.pdf); no CTV-specific viewability guideline. Nearest CTV document: 'OTT/CTV and SSAI Digital Video Measurement Guidelines' issued Aug 2021 (file: 083021 SSAI and OTT Guidance FINAL.pdf). | supported with qualification | Viewable Ad Impression 2.0 (2015), mobile (2016); no CTV viewability guideline | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C4-19 | MRC invalid-traffic guidelines | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | MRC standards page: 'Invalid Traffic Detection and Filtration Guidelines & Addendum / Invalid Traffic (IVT 2.0) Detection and Filtration Standards Addendum' issued Oct 2015, updated Jun 2020 (file: IVT Addendum Update 062520.pdf), plus 'Invalid Traffic Detection and Filtration Interim Update Memo' issued Apr 2024 (file: 2024_IVT_Interim_Updates_FINAL.pdf). | supported with qualification | IVT 2.0 Addendum of 25 June 2020 with the April 2024 interim memo; Digital Advertising Auction Transparency Standards, January 2026 | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C4-20 | TAG programmes relevant to supply paths | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | tagtoday.net/certifications lists four programmes: Certified Against Fraud (https://www.tagtoday.net/fraud, 'Anti-Fraud Program Guidelines', launched 2016), Certified Against Malvertising (https://www.tagtoday.net/threat-sharing, 'Anti-Malvertising Program Guidelines'), Brand Safety Certified (https://www.tagtoday.net/brand-safety, 'Brand Safety Program Guidelines'), Certified For Transparency (https://www.tagtoday.net/transparency, 'Transparency Program Guidelines', seal announced Nov 2022); plus Platinum status. No guideline version numbers or years are shown. 'TrustNet' does not appear on the homepage or certifications page (only an image file path on the homepage). | supported with qualification | four seals; no TrustNet programme page | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C5-01 | Adelaide High Attention Markets, 7 March 2023 | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Adelaide's own release, datelined New York, March 7, 2023, announces global High Attention Markets (headline says 'High Attention Marketplaces', body says 'High Attention Markets') with Google Display & Video 360, Xandr (part of Microsoft Advertising), Magnite, PubMatic, Teads, Sovrn, TRUSTX and Sharethrough. It lists 20 countries plus the Nordics ('20+ countries', incl. US, Canada, Japan, Germany, UK, Australia, Brazil, India; 'nearly 5 trillion monthly impressions'). PMPs are reached through seats on major DSPs; DV360, Xandr and MediaMath are said to have curated high-AU marketplaces. | supported with qualification | date and partners confirmed; Adelaide's current page lists a changed SSP roster | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C5-03 | AU in Index Marketplaces, January 2026 | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Index Exchange newsroom release 'Adelaide Joins Index Marketplaces, Bringing Attention-Based Media Quality to Sell-Side Curation', New York, NY, Jan 07, 2026: AU 'is now available in Index Marketplaces' and can be used 'without added fees or custom integrations'. Index's own description in that release: 'Index Marketplaces is Index Exchange's omnichannel platform for curated, signal-rich supply activation and sell-side decisioning.' Index Marketplaces itself was announced Jan 23, 2024 (Toronto) as 'an omnichannel platform aimed at helping partners transact more efficiently via Index's robust supply footprint and advanced, modern infrastructure', with Adelaide already among the global launch partners; the Oct 07, 2025 data-vendor release calls it 'a cornerstone of the sell-side decisioning movement' and says 'Index does not charge data fees'. | supported with qualification | 7 January 2026; Index Marketplaces announced 23 January 2024 | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C5-04 | Amazon DSP AU tiers, 8 October 2024 | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | 8 October 2024 belongs to The Trade Desk, not Amazon DSP. Adelaide's release 'Brands Achieve Significant Gains in Media Quality and Performance with Adelaide's Attention Pre-Bid Segments' (New York, October 8, 2024) describes 'attention pre-bid targeting within The Trade Desk' and advertisers choosing 'high, average, or low AU'; it does not mention Amazon DSP. AdExchanger the same day: Adelaide announced 'attention-based segments are available for pre-bid targeting in The Trade Desk', with 'three different levels of attention targetable within TTD's platform: low, medium and high'. Amazon DSP appears only in the June 2026 release. | contradicted | 8 October 2024 is The Trade Desk | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C5-05 | The Trade Desk 8 October 2024; Amazon DSP 9 June 2026 | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | TTD: Adelaide's Oct 8, 2024 release names the quality levels 'high, average, or low AU' and gives no beta dates; AdExchanger (Oct 8, 2024) reports 'low, medium and high' (top 30% / middle 40% / bottom 30%) and that advertisers 'tested the solution between August and September' per Adelaide. Amazon: Adelaide's release 'Adelaide Brings Attention-Based Pre-Bid Segments to Amazon DSP, Helping Advertisers Reduce Waste and Prioritize High-Quality Inventory' is dated Jun 9, 2026 on Adelaide's news index (the post page itself shows no date); it says 'The launch includes two Adelaide products within Amazon DSP', 'AU Media Quality allows advertisers to target or exclude inventory based on High, Average, and Low AU performance tiers', and the AU Quality Floor 'serves as a waste exclusion filter' (MFA sites plus bottom 10% of AU-scored placements), 'launching exclusively in Amazon DSP'. AdExchanger' | supported with qualification | confirmed; Adelaide's tier names are High, Average and Low | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C5-06 | Stagwell Curate as reported by Digiday, July 2026 | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Digiday, Sam Bradley, 'Stagwell is building its own AI media curation marketplace', July 16, 2026: Stagwell is 'set to launch Stagwell Curate, an AI platform' pulling curated CTV, OLV, display and audio inventory into a central reservoir; 'The system uses AI agents built in Claude'; a 'domain health check' runs on 'data pulled from The Trade Desk's OpenSincera'; it 'relies upon Magnite and Freewheel's publisher relationships'; TTD's OpenPath is cited as the kind of marketplace Stagwell wants to rely on less; no earnings statement is mentioned. Stagwell's own release the same day (ACCESS Newswire, New York, July 16, 2026, 'Stagwell Launches Stagwell Curate') calls it 'a new AI-powered marketplace and platform designed to reach audiences through premium, trusted media partnerships', says 'Stagwell Curate is available immediately to clients across the Stagwell network', 'Partners including | supported with qualification | Digiday 16 July 2026, "set to launch"; Stagwell's release the same day, "available immediately" | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| F-C5-07 | Google Ad Manager curation feature | Brief fact-check, 2 November 2026: an independent checker opened the primary source. | Ad Manager Help page 'Configure curation settings (Beta)' (answer/15308224), under the Help topic 'Curation (Beta)' (topic/15272903, two articles). It says 'Curation partners use this information to create data segments' that buyers target in their agency seat and turn into a Deal ID; 'curation partners can also create packages of inventory to share with a buyer'; 'Curation partners can set floor prices for these curated packages'; partners set a single fee visible to buyers that Google pays from the bid price; a buyer subscribes to the partner's Deal ID in the agency seat. Settings path: 'Navigate to Signals, then Curation settings' (controls: Allow data sharing, Share secure signals, Allow inventory packaging). 'Features in Beta phase might not be available in your network.' Launch: Google's Ad Manager blog, Nov 04, 2024, Peentoo Patel: 'launching curation, a new capability that simpli | supported with qualification | Curation (Beta), launched 4 November 2024; Marketplace Packages (Beta) among the transaction types | Shared rules: settle each claim at the source that settles its type; record what stays unverified. | RESEARCH-BRIEF.md v1.1 |
| R1-CUR-01 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-01 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. The brief now states its own cutoff and that it replaces the shared rules' 2 October 2026 date. Because the author directed on 2 November 2026 that the paper be written at once, the freeze is Tuesday 3 November 2026, not the planned 4 December; status-sensitive claims are rechecked at publication. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-02 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-02 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. A curated deal is a deal ID with a documented selection or enrichment rule; mechanism, publisher multiplicity (publisher entities counted apart from selling accounts and owners), transaction type and buyer eligibility are four fields. Multi-publisher curation is the primary subset; Roku Curate is the single-publisher test case. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-03 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-03 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. Cohorts follow the purchased service: A exchange-hosted deal-building tools, B standardised curated deal products, C managed client marketplaces; operator type and ownership are attributes; a frozen four-step decision tree; features, aliases and activation variants never create rows. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-04 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-04 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. Hypotheses, eligibility definitions and a provisional register freeze before workstream U; the eligible register, masks and anchors freeze before scoring; later additions are versioned amendments. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-05 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-05 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. Masks are fixed per cohort: A all twelve; B and C all except auction mechanics, so supply-path integrity now applies to C. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-06 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-06 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. The opening no longer asserts "most of programmatic" or that the impression is unchanged; the essay's position is labelled a hypothesis and the absence claim is bounded to the sources screened. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-07 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-07 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. H1 reports each series separately (geography, population, unit, definition, period), separates matched-panel from composition changes, and stays unresolved unless a representative series identifies multi-publisher curated transactions. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-08 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-08 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. H2 split into H2a (all-in net buyer cost under documented matched comparison) and H2b (disclosure of each charge with payer, collector, base, period, gross or net). | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-09 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-09 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. H3 tests each quality promise as threshold attainment or improvement over a pre-curation comparator, with a registered margin per metric; an inconclusive comparison is unresolved. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-10 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-10 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. H4 is a descriptive access question with four states (default, configurable, recommended, observed) plus unknown; the essay's no-default-advantage claim is evaluated against the finite register. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-11 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-11 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. H5 compares two supplier registers (2023 and cutoff) with operating supplier and ultimate controlling group recorded separately, one control definition, owner business type as an attribute; rebrands do not change counts. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-12 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-12 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. H6 codes composition, selection rule and provenance separately for every eligible A, B and C product (complete, partial, none, access unknown), distinguishing capability, observed disclosure and delivery verification, with counts against the full denominator. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-13 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-13 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. H7 defines the agent (decision, inputs, authority, permitted actions, execution location, feedback) and requires third-party production evidence naming a deployment, period, executed decision and transaction context; announcements, demos and author-built tests do not count. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-14 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-14 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. Q1 records rule owner, execution host, data supplier, contracting seller, fee recipient and the party able to change the deal; new Q5 and research question 13 cover post-purchase control. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-15 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-15 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. Every charge records payer, collector, amount or rate, denominator, period, gross or net and inclusion; "no additional fee" is settled from evidence; reconciliation only within one transaction perimeter. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-16 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-16 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. Every cell records a disclosure mode; public documentation supports a capability score under its cap; a specific deal needs an identified example; low scores only after the registered search; unobserved behaviour is n/e. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-17 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-17 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. Dimension 3 records source, provider, collection method, intended use, permissions and legal basis separately; scores documentation, not compliance; 0 is a verified material contradiction, not only a court finding. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-18 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-18 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. Dimension 4 scores evidence quality independently of result (5 replication, 4 one independent study, 3 vendor or commissioned with method, 1 claims, 0 demonstrated invalidity); effect size reported separately. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-19 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-19 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. Dimension 6 publishes breadth descriptively on a non-overlapping taxonomy and scores documented execution within the advertised scope captured in U (100% / 50 to under 100% / under 50% / documented failure). | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-20 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-20 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. Dimension 7 counts distinct, deduplicated buying endpoints with documented activation; "any DSP" without names is n/e; 0 needs documented absence or withdrawal. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-21 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-21 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. Dimension 8 freezes one unit, perimeter and window per cohort after U and before scoring, with numeric boundaries, boundary treatment and a stale-data rule in the rules file; company-wide and unperiodised figures are context. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-22 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-22 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. Dimension 10 scores the charges the product controls or collects (amount or rate, base, payer, collector, period); third-party charges are attributes; the four series stay distinct in the deal-path record. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-23 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-23 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. Dimension 11 records assured entity, product, metric, environment, period, standard and funding; input accreditation and company certification are attributes unless their scope covers the service; 0 needs a defined current failure. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-24 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-24 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. Composition to 2, measured quality evidence to 4, assurance scope to 11, authorisation and traceability to 12; resale and request duplication reported separately over the observed sample. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-25 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-25 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. Sibling evidence reused with paper version, claim id, source, recheck date and disposition; sibling scores and input accreditation never transfer to the curated product. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-26 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-26 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted with a limitation. The conflict declaration is frozen as provisional on 2 November 2026 and reconfirmed at publication; Verve, Samba and Semasio products enter under ordinary rules; no separate reviewer is available, so the paper says so and publishes a sensitivity run excluding interest-marked rows from each field mean. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-27 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-27 (design) | decided against shared rules v3 and ROUND-1-CHANGES precedents | structural | Accepted. Research questions 8 and 9 reworded. | design finding | RESEARCH-BRIEF.md v2; data/hypotheses.json; data/scoring-rules.json |
| R1-CUR-28 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-28 (fact) | primary source re-opened by a checker or the browser pane (private/factcheck/round1-V*.json) | supported as stated | Fiducia glossary: PMP defined as a selected-publisher deal environment; only OMP and PMP as marketplace types; publisher multiplicity and programmatic guaranteed unspecified. Brief and H1 now say the ANA series cannot isolate multi-publisher curation. | an outside model’s review is a lead, not evidence | RESEARCH-BRIEF.md v2 |
| R1-CUR-29 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-29 (fact) | primary source re-opened by a checker or the browser pane (private/factcheck/round1-V*.json) | supported as stated | Fiducia methodology: estimated SSP fee rates applied where SSP log-level data or identifiers are unavailable. Workstream H distinguishes observed from imputed fees. | an outside model’s review is a lead, not evidence | RESEARCH-BRIEF.md v2 |
| R1-CUR-30 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-30 (fact) | primary source re-opened by a checker or the browser pane (private/factcheck/round1-V*.json) | supported as stated | Google Authorized Buyers Help (Beta): curated packages can bundle Private Auctions and Preferred Deals; fees as percentage of media or CPM, visible or hidden; nothing on who pays. Entered under H4 and workstream E. | an outside model’s review is a lead, not evidence | RESEARCH-BRIEF.md v2 |
| R1-CUR-31 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-31 (fact) | primary source re-opened by a checker or the browser pane (private/factcheck/round1-V*.json) | supported with qualification | SupplyChain 1.0 lists payment participants; stronger than "need not appear": every node must be in the payment flow, so a non-payment provider cannot be represented. Definitions updated. | an outside model’s review is a lead, not evidence | RESEARCH-BRIEF.md v2 |
| R1-CUR-32 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-32 (fact) | primary source re-opened by a checker or the browser pane (private/factcheck/round1-V*.json) | supported with qualification | Deal Sync API Specification 1.0: curfeetype 0 undisclosed to 4 no fee; no deal revisions in 1.0 ("some attributes may be subject to change"); the guidance text calls the attribute curationfee. Dimension 10 now scores a declared type of 4 as 2 and of 0 as 1; adoption recorded apart from specification support. | an outside model’s review is a lead, not evidence | RESEARCH-BRIEF.md v2 |
| R1-CUR-33 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-33 (fact) | primary source re-opened by a checker or the browser pane (private/factcheck/round1-V*.json) | supported with qualification | Prebid Deals documentation recommends line-item priority above open-market bids; generic ad-server guidance with no curated-deal mention; docs.prebid.org has no curation page. Entered under H4 and workstream A as configuration guidance. | an outside model’s review is a lead, not evidence | RESEARCH-BRIEF.md v2 |
| R1-CUR-34 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-34 (fact) | primary source re-opened by a checker or the browser pane (private/factcheck/round1-V*.json) | supported as stated | Microsoft Curation Optimization: Microsoft-controlled sampling of curated deals with low bidding. Request sampling added to the mechanism field, Q1 and dimension 5. | an outside model’s review is a lead, not evidence | RESEARCH-BRIEF.md v2 |
| R1-CUR-35 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-35 (fact) | primary source re-opened by a checker or the browser pane (private/factcheck/round1-V*.json) | supported with qualification | MRC Cross-Media Audience Measurement Standards (Phase I Video, September 2019) set a TV and OTT viewability assumption and defer to the viewability guidelines; DoubleVerify's CTV viewability accreditation is scoped to certified applications. Workstream A now says there are applicable CTV requirements but no separately titled guideline. | an outside model’s review is a lead, not evidence | RESEARCH-BRIEF.md v2 |
| R1-CUR-36 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-36 (fact) | primary source re-opened by a checker or the browser pane (private/factcheck/round1-V*.json) | supported with qualification | Publicis half-year report: Lotame acquired for 100% in April 2025; Equativ dates the Sharethrough acquisition to June 2024; neither gives a legal closing day. H5 register updated. | an outside model’s review is a lead, not evidence | RESEARCH-BRIEF.md v2 |
| R1-CUR-37 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-37 (fact) | primary source re-opened by a checker or the browser pane (private/factcheck/round1-V*.json) | supported with qualification | Microsoft announced on 14 May 2025 that Invest media buying would end on 28 February 2026 and reaffirmed it on 7 October 2025; no Microsoft page after that date confirms execution. Brief says "announced", not "ended". | an outside model’s review is a lead, not evidence | RESEARCH-BRIEF.md v2 |
| R1-CUR-38 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-38 (fact) | primary source re-opened by a checker or the browser pane (private/factcheck/round1-V*.json) | supported with qualification | The quoted PubMatic sentence is not on the buyers page (browser read, 2 November 2026); Activate's page describes direct-to-supply buying without curation, so Activate is a comparator. Magnite's ClearLine Activate (9 September 2026) is an activation variant, not a row. | an outside model’s review is a lead, not evidence | RESEARCH-BRIEF.md v2 |
| R1-CUR-39 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-39 (fact) | primary source re-opened by a checker or the browser pane (private/factcheck/round1-V*.json) | supported as stated | Criteo Commerce Grid ("the world's first commerce SSP") and InMobi Buyer Hub (agency access 31 August 2026) added as cohort A candidates; neither page states US availability. | an outside model’s review is a lead, not evidence | RESEARCH-BRIEF.md v2 |
| R1-CUR-40 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-40 (fact) | primary source re-opened by a checker or the browser pane (private/factcheck/round1-V*.json) | supported with qualification | Multilocal (cureon) and Dstillery (ID-free Deal IDs) added as cohort B candidates; OAREX's backing is a $4 million receivables-backed facility, not ownership; Dstillery's FAQ is dated 2022 yet names Microsoft Curate, so it was edited later and needs a status check. | an outside model’s review is a lead, not evidence | RESEARCH-BRIEF.md v2 |
| R1-CUR-41 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-41 (fact) | primary source re-opened by a checker or the browser pane (private/factcheck/round1-V*.json) | supported as stated | OpenX release of 16 June 2026: Chalice AI models run as always-on bidding agents inside OpenX for Hyundai; vendor performance claims. Entered under H7 and workstream J; OpenXBuild in the comparator table. | an outside model’s review is a lead, not evidence | RESEARCH-BRIEF.md v2 |
| R1-CUR-42 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-42 (fact) | primary source re-opened by a checker or the browser pane (private/factcheck/round1-V*.json) | supported as stated | Roku Curate (27 April 2026): Roku's own inventory with six named data partners; activation by Roku or via an unnamed DSP. Registered as the single-publisher boundary case. | an outside model’s review is a lead, not evidence | RESEARCH-BRIEF.md v2 |
| R1-CUR-43 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-43 (fact) | primary source re-opened by a checker or the browser pane (private/factcheck/round1-V*.json) | supported as stated | Stagwell release of 16 July 2026: "available immediately"; Media Machine buying agents "will connect" to Curate. Cohort C entry and workstream J updated; no agent connection in production on that record. | an outside model’s review is a lead, not evidence | RESEARCH-BRIEF.md v2 |
| R1-CUR-44 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-44 (fact) | primary source re-opened by a checker or the browser pane (private/factcheck/round1-V*.json) | supported as stated | Answer E-000132/2026 by Executive Vice-President Ribera, 12 March 2026 (read in the browser; the fetch tool returned an empty body): the Commission was consulting market participants on Google's compliance plan and had not reached a final conclusion on a structural remedy. Workstream I updated. | an outside model’s review is a lead, not evidence | RESEARCH-BRIEF.md v2 |
| R1-CUR-45 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-45 (fact) | primary source re-opened by a checker or the browser pane (private/factcheck/round1-V*.json) | supported with qualification | "Considering next steps" is the CMA ad-tech case's timetable phase (estimated December 2025 to July 2026; page updated 19 May 2026; status Open), not its status; Google's mobile platform was designated with SMS on 22 October 2025. Workstream I keeps the Competition Act case apart from both SMS designations. | an outside model’s review is a lead, not evidence | RESEARCH-BRIEF.md v2 |
| R1-CUR-46 | Brief v1.1 as reviewed | OpenAI review round 1 finding CUR-46 (fact) | primary source re-opened by a checker or the browser pane (private/factcheck/round1-V*.json) | supported as stated | Remedies opinion p. 103: final judgment to take effect 60 days after entry, except that work to appoint the Monitor and form the Technical Committee starts on entry; six-year term; behavioural remedies within 15 months. Q2 updated. | an outside model’s review is a lead, not evidence | RESEARCH-BRIEF.md v2 |
| V-A-01 | OpenRTB 2.6's implementation notes (section 7.3) describe prioritising Deal ID bids above open-auction bids and assigning preferential floors as publisher-side configurations rather than defaults, and recommend that any prioritisation other than by price, and any preferential floor for a seat, carry a Deal ID. | Refute-mode verifier for A.json re-opened the cited source | 7.3 'Warning' reads verbatim: 'The exchange side may be configured by a publisher to prioritize all Deal ID bids above open auction bids, or to force a price auction between them with different floors by class of bid. There are multiple common practices that depend on how the publisher prefers to sell inventory with Deal ID.' The Policy Recommendations ('A Deal ID should be utilized for any situation where the auction may be awarded to a bid not on the basis of price alone. Any prioritization of bids other than by price should have a Deal ID.'; 'A Deal ID is recommended for all situations where a preferential floor may be assigned to a seat entity.') and the Anti-Patterns passage match the stream verbatim. The word 'default' does not appear in 7.3. The section also sets out use cases the stream did not record: Case-3 'Open Bidding with Deal ID as Value-added Markers' (Deal ID represents | supported with qualification | OpenRTB 2.6's implementation notes (section 7.3) say an exchange 'may be configured by a publisher to prioritize all Deal ID bids above open auction bids, or to force a price auction between them with different floors by class of bid', call these 'multiple common practices', recommend that any prioritisation other than by price and any preferential floor for a seat carry a Deal ID, and describe use cases in which prioritisation is expected (a 'First Look Trading Agreement' locked to one buyer; a 'Direct Option Deal' on private inventory) alongside one in which Deal IDs are 'Package Tokens' with their own floors and 'Winner is decided by bid price'; the notes never use the word default. | scope, unit, date or wording narrowed to what the source carries | CUR-A-F05 |
| V-A-02 | On 16 December 2024 IAB Tech Lab announced a 2025 Curation Workstream under its Addressability Working Group and the rebranding of Seller Defined Audiences as Curated Audiences, with 'visibility of curation or enrichment partners' and data-leakage prevention as topics and 'plans for testing the framework in 2025'; by 3 November 2026 the Tech Lab's own site search for 'curation' returned no specification, standard or guideline titled with the word, and the only specification-grade output naming a curator is the Deal Sync API 1.0, which a September 2025 member perspective attributes to the 'Curation Sub Working Group'. | Refute-mode verifier for A.json re-opened the cited source | The blog post ('Published: December 16, 2024', Hillary Slattery) carries every quoted sentence; its body says 'Curation Framework Subgroup' while the title says 'Working Group'. The site search shows 10 results on page 1 with a 'Next Page' link; page 2 ('Page 2 of 2') shows 10 more, among them 'Curated Audiences' (listed 24 February 2022), which is the standards page https://iabtechlab.com/sda/: 'Last updated: August 25, 2025'; 'In December 2024, Seller Defined Audiences was renamed Curated Audiences'; 'Curated Audiences (formerly Seller Defined Audiences or SDA) is an addressability specification incubated within Project Rearc'; its download link is the February 2022 'Taxonomy and Data Transparency Standards to Support Seller-defined Audience and Context Signaling' PDF, no version number stated. That page also says: 'Work to establish and evolve a Curation Framework starting with standa [252] | supported with qualification | On 16 December 2024 IAB Tech Lab announced a 2025 Curation Workstream under its Addressability Working Group and the rebranding of Seller Defined Audiences as Curated Audiences, with 'visibility of curation or enrichment partners' and data-leakage prevention as topics and 'plans for testing the framework in 2025'; by 3 November 2026 the Tech Lab's site search for 'curation' (20 results over two pages) returned no new specification, standard or guideline: the only standards page with 'Curated' in its title is the renamed Seller Defined Audiences page (last updated 25 August 2025, still serving the February 2022 specification), and the only specification-grade output naming a curator is the Deal Sync API 1.0, whose deals-related standards work the Tech Lab's own Curated Audiences page places in the 'Curation Sub Working Group' with 'a release expected in Q4 2025', as the September 2025 One | scope, unit, date or wording narrowed to what the source carries | CUR-A-F20 |
| V-A-03 | TAG offers four seals, each certifying a company rather than a deal or product: Certified Against Fraud (launched 2016; requires that 100% of monetizable transactions are filtered for IVT by a qualified third party, with data-centre IP filtering; recertification every January), Certified Against Malvertising, Brand Safety Certified and Certified for Transparency (announced November 2022; guidelines PDF dated 2023); the Brand Safety and Transparency pages list no requirements in their text, and 'TrustNet' appears only on the Transparency page as a distributed-ledger initiative 'Powered by technology partner Fiducia', not as a seal. | Refute-mode verifier for A.json re-opened the cited source | Certifications page: headings Certified Against Fraud, Certified Against Malvertising, Brand Safety Certified, Certified For Transparency; body text 'Certified Against Malware Program'; 'TAG launched its Certified Against Fraud Program in 2016'; 'Since 2014, TAG has partnered with industry leaders to design and strengthen the Certified Against Malware Program'; 'Announced in November 2022, the Certified for Transparency seal recognises trusted businesses'; 'TAG member companies that receive all of TAG's available seals achieve the coveted Platinum Member status'; 'TrustNet' does not appear. Fraud page: 'One of the key requirements for companies looking to achieve CAF is ensuring that 100% of their monetizable transactions are filtered for IVT detection and removal by a qualified third party.'; 'Effective immediately, TAG will allow publishers to fulfill the requirement ... by leveraging  [152] | supported with qualification | TAG offers four seals, each certifying a company rather than a deal or product: Certified Against Fraud (launched 2016; requires that 100% of monetizable transactions are filtered for IVT by a qualified third party, with data-centre IP filtering), Certified Against Malvertising, Brand Safety Certified and Certified for Transparency (announced November 2022; guidelines PDF file-named 2023); recertification for all TAG seals happens in January every year with applications due 31 January; the Brand Safety and Transparency pages list no requirements in their text, and 'TrustNet' appears only on the Transparency page as a distributed-ledger platform 'Powered by technology partner Fiducia', not as a seal. | scope, unit, date or wording narrowed to what the source carries | CUR-A-F35 |
| V-B-04 | For April to June 2025 (21 contributing of 39, $146 million, 18.6 billion impressions) the median PMP share rose to 87.8% and the aggregate to 82.6%, with median CPMs of $7.15 for PMP against $4.41 for OMP and spend-weighted CPMs of $7.43 against $3.64; CTV reached 44.2% of spend and the ANA re-based its efficiency opportunity to $26.8 billion. | Refute-mode verifier for B.json re-opened the cited source | Q2 2025 press release (14 August 2025): 'PMP share of spend surged from 64.5% to 87.8%, with PMP CPMs averaging $7.15 vs. $4.41 for OMP' and 'CTV's share of programmatic spend jumped from 30.4% to 44.2%'; it labels none of these as medians and gives no participant, spend or impression counts. Q3 2025 report p. 7 table, Q2 2025 column: Participating 39, Active 21, Total Impressions 18.6 B, Total Ad Spending $146 M, OMP 17.4% / PMP 82.6% (total), OMP CPM $3.64 / PMP CPM $7.43 (total); its median table (p. 12) gives PMP 87.8% (Q2) to 88.8% (Q3) across all environments, 79.9% to 81.4% web and mobile, 100% to 100% CTV, which establishes 87.8% as a median. Q1 2026 report p. 5 table lists CPM PMP $7.15 and CPM OMP $4.41 under Q2 2025 with no basis label. Q3 2025 report footnote 6: 'The $26.8 billion reflects the 21.8 percent median Q3 TrueCPM Opportunity for a global $123 billion estimated 2025 | supported with qualification | For April to June 2025 (21 contributing of 39, $146 million, 18.6 billion impressions) the median PMP share rose to 87.8% (a median per the Q3 2025 report's median table) and the aggregate to 82.6%; PMP CPMs were $7.15 against $4.41 for OMP (the ANA's release calls these 'averaging' and the Q1 2026 table carries them without a basis label; the gated Q2 2025 report is the only place a median label could be confirmed) and spend-weighted CPMs were $7.43 against $3.64; CTV reached 44.2% of spend and the ANA put its efficiency opportunity at $26.8 billion (21.8% of a $123 billion Madison and Wall base that includes DOOH and digital audio). | scope, unit, date or wording narrowed to what the source carries | CUR-B-F08 |
| V-B-05 | ANA fee components in the series: buy-side transaction costs were 16.8% of ad spending in the April to September 2024 data (DSP platform 8.6%, data and additional costs 8.2%), private-marketplace DSP platform costs ran 1.1 points below open-marketplace costs, CTV DSP fees averaged 8.78% 'with little difference between marketplace type', and in Q4 2025 median SSP platform costs fell to 12.7%. | Refute-mode verifier for B.json re-opened the cited source | December 2024 update p. 11: 'buy-side transaction costs have risen slightly to 16.8 percent, with DSP platform costs increasing to 8.6 percent, while data/additional costs held steady at 8.2 percent. A deeper look by marketplace type reveals that private marketplace DSP platform costs are 1.1 percent lower than those for the open marketplace.' p. 8: 'Average CTV DSP fees are 8.78 percent, with little difference between marketplace type', with a chart showing 8.91%, 8.90%, 8.78%, 8.75%. Q4 2025 report p. 7: 'SSP platform costs fell to 12.7 percent, while seller revenue rose'. But the Q1 2025 report p. 10 restates the same CTV figure as a dollar median: 'DSPs charged marketers $3.47 less for access to CTV inventory, with a median value decreasing from $8.78 in Q4 2024 to $5.31.' | supported with qualification | ANA fee components in the series: buy-side transaction costs were 16.8% of ad spending in the April to September 2024 data (DSP platform 8.6%, data and additional costs 8.2%); private-marketplace DSP platform costs were '1.1 percent lower' than open-marketplace costs (the report's wording, read here as 1.1 points); the December 2024 report put average CTV DSP fees at '8.78 percent, with little difference between marketplace type', a figure the Q1 2025 report restates as a median of $8.78 falling to $5.31; and in Q4 2025 median SSP platform costs fell to 12.7%. | scope, unit, date or wording narrowed to what the source carries | CUR-B-F16 |
| V-B-06 | Jounce's April 2025 Supply Path Benchmarking report, as reported by Playwire and AdExchanger, found that about half of curated deal IDs in the March 2025 sample carried no fee beyond the exchange's standard take, 34% carried a margin-based fee (median 14%), 10% a fixed CPM fee (median $0.48) and 7% a static price, with a median effective margin of 30% and a median effective CPM of $0.51 per Playwire; the report itself is client-only. | Refute-mode verifier for B.json re-opened the cited source | Playwire, 10 April 2025: headings 'Zero Fee Curation (50% of bid requests)', 'Margin-Based Curation (34% of bid requests)', 'CPM-Based Curation (10% of bid requests)', 'Static Price Curation (7% of bid requests)'; body 'Half of all curated deal IDs carry no additional fee beyond the exchange's standard open auction take rate'; 'the median fee for margin-based curation is 14%'; 'Approximately 36% of margin-based fees fall in the 5-10% range'; 'the median fee for CPM-based curation is $0.48'; 'The median effective margin for curated deals is 30%, with the median effective CPM at $0.51'; it names the 'April 2025 Supply Path Benchmarking Report' and states no sample month. AdExchanger, 1 May 2025: 'Roughly half of all curated deal IDs carry no curation-related fees, according to Jounce'; 'Around 34% of curated deals carry margin-based fees'; 'the median margin-based fee is 14%'; 'nearly 40% | supported with qualification | Jounce's April 2025 Supply Path Benchmarking report, as reported by Playwire (10 April 2025) and AdExchanger (1 May 2025), found that about half of curated deal IDs (Playwire's headings say 'of bid requests') carried no fee beyond the exchange's standard take, 34% a margin-based fee (median 14%), 10% a fixed CPM fee (median $0.48) and 7% a static price, with a median effective margin of 30% and a median effective CPM of $0.51 per Playwire; neither account states the sample month, AdExchanger's March 2025 dating applies to its 1.5 billion-request auction-mix analysis, and the report itself is client-only. | scope, unit, date or wording narrowed to what the source carries | CUR-B-F20 |
| V-B-07 | Balocco's 2023 Erasmus thesis (abstract) finds that private marketplaces 'cause adverse selection' in open real-time bidding 'without raising advertiser utility' and are a weak signalling device, and argues that publishers' move from the waterfall to header bidding removed exchanges' incentive to keep sell-side fees low, explaining the rise in sell-side fees known as the 'Ad Tech-Tax'; the full text could not be opened. | Refute-mode verifier for B.json re-opened the cited source | Record page (Pure, Erasmus University Rotterdam; published 8 September 2023; supervisors Ting Li and Eric van Heck; ISBN 978-90-5892-669-2): the abstract, per the fetch tool's reading, says premium contracts (PMPs) 'lead to adverse selection in the open market (RTB)' 'while not increasing advertisers' utility', that PMPs are 'not the effective signaling device they are often assumed to be', that replacing the Waterfall with Header Bidding 'eliminates all incentives for ad exchanges to keep sell-side fees low', and that this is linked to the rise in sell-side fees 'commonly referred to as the Ad Tech-Tax'. The record lists a final-dissertation PDF (3.06 MB) at pure.eur.nl/files/97663628/dissertationtemplateprintnew_-_64b7a6fbed6c7.pdf; a download with a browser user agent returned HTTP 403 with a Cloudflare 'Just a moment...' challenge page. | supported with qualification | Balocco's 2023 Erasmus thesis (abstract) finds that premium contracts (PMPs) 'lead to adverse selection in the open market (RTB)' 'while not increasing advertisers' utility' and are 'not the effective signaling device they are often assumed to be', and argues that publishers' replacement of the waterfall with header bidding 'eliminates all incentives for ad exchanges to keep sell-side fees low', which the abstract links to the rise in sell-side fees 'commonly referred to as the Ad Tech-Tax'; the full text (a PDF listed on the record) returned HTTP 403 behind a Cloudflare challenge. | scope, unit, date or wording narrowed to what the source carries | CUR-B-F30 |
| V-B-08 | One row per published figure on deal-based trading, fees, waste and samples (73 rows), each with its source_id, period and claim status. | Refute-mode verifier for B.json re-opened the cited source | All ANA rows match the quarter's own report as re-opened (sample counts, spend, impressions, OMP/PMP shares, CPMs, TrueAdSpend, MFA, fee shares); the $21.6 billion row is '20.8 percent' of $104 billion in the Q1 2025 report (p. 3); the $26.8 billion row is 21.8% of $123 billion per the Q3 2025 report footnote 6; Jounce, Playwire/AdExchanger, ISBA/PwC and Adalytics rows match. Three rows carry a source or label that the cited source does not supply: (1) 'PMP share of tracked spend, median 87.8%' cites CUR-B-S05, whose text ('PMP share of spend surged from 64.5% to 87.8%') has no median label; the label is in the Q3 2025 report's median table (CUR-B-S06, p. 12: 87.8% to 88.8%). (2) 'PMP CPM vs OMP CPM, median $7.15 vs $4.41' cites CUR-B-S05, which says 'PMP CPMs averaging $7.15 vs. $4.41'. (3) 'CTV DSP fees 8.78%' is a percentage in CUR-B-S02 but restated as a $8.78 median in the Q1 2025 r [66, 98, 66, 102, 98, 66] | supported with qualification | Change the source_id of the 87.8% median row to CUR-B-S06 (Q3 2025 report median table) with CUR-B-S05 as the restating release; relabel the Q2 2025 CPM row as 'PMP CPM vs OMP CPM ($7.15 vs $4.41; "averaging" per the release, basis unlabelled in the Q1 2026 table)'; add to the 8.78% row's 'unstated' cell 'unit: 8.78 percent in the December 2024 report, $8.78 median in the Q1 2025 report'; add to the MFA 1.1% row's 'unstated' cell 'the Q2 2025 release gave a Q2 median of 0.8%'. Add to the $21.6bn and $26.8bn rows that the Madison and Wall base covers web, mobile, CTV, DOOH and digital audio (US $52 billion 2024, global taken as twice the US). | scope, unit, date or wording narrowed to what the source carries | CUR-B data_table estimates |
| V-B-09 | Tidy ANA benchmark series: seven period rows with median and aggregate PMP share, CTV PMP share, marketplace CPMs and sample. | Refute-mode verifier for B.json re-opened the cited source | 2023 and Apr-Sep 2024 rows match the Q1 2025 report's p. 8 table (median 54.2 / 56.2, aggregate 40.9 / 65.9; CPMs 6.68, 2.23, 5.83, 2.75 and 5.67, 3.26, 7.46, 3.35; 21/21 and 37/21; $123M/$235M; 35.5bn/38.5bn). Q1 2025 row matches the same table. Q2 2025 row: 87.8 median (Q3 report median table), 82.6 aggregate, CTV 100 median / 99.5 aggregate (Q3 report median and total tables), $7.43/$3.64 aggregate (Q3 report), 39/21, $146M, 18.6bn all confirmed; $7.15/$4.41 are 'averaging' in the release and unlabelled in the Q1 2026 table. Q3 2025 row: 88.8 / 81.6, CTV 100 / 99.2, $9.88/$3.55 confirmed. Q4 2025 row: 92.2 / 79.2, 100 / 100, $10.72/$4.25, 54/35, $142M, 16.7bn confirmed. Q1 2026 row: 86/66, $160M, 20.9bn, $7.77/$3.58 (unlabelled) confirmed. The Q1 2026 table's 'CPM Total' row ($3.74, $5.81, $6.86, $5.55, $4.42) equals the 2023 spend-weighted CPM under 'Q1 2025' and the Q2-Q4 median tot | supported with qualification | Relabel the Q2 2025 row's cpm_pmp_median_usd and cpm_omp_median_usd cells as '7.15 (release: "averaging"; basis unlabelled in the Q1 2026 table)' and '4.41 (same)', or move them to a note; keep every other cell. | scope, unit, date or wording narrowed to what the source carries | CUR-B data_table ana_pmp_series |
| V-C-10 | Teads' release of 17 August 2023 says its high-attention PMPs (DV360, The Trade Desk, Yahoo) include placements 'an average attention score 31% higher than Adelaide's Teads AU benchmarks' and lists outcome lifts for unnamed advertisers (12.5%, 32.7%, 41.6%, 110.8%, 8x) measured by Adelaide and unnamed third-party brand measurement partners, alongside projections ('would improve overall campaign performance by 31%'). | Refute-mode verifier for C.json re-opened the cited source | 'NEW YORK – August 17, 2023'; 'premium placements with an average attention score 31% higher than Adelaide's Teads AU benchmarks'; DSPs 'Google Display & Video 360, The Trade Desk, and Yahoo'; 'Adelaide has measured AU across more than 9.3 billion Teads impressions, 65 advertiser campaigns, 22 ad formats' and 'well over 20,000 domains', 'demonstrating the relationship between AU and full-funnel outcomes measured by third-party brand measurement partners'; 'Since launching the Teads Attention Program just over a year ago'. Cases: multinational hospitality company 32.7% increase in Familiarity and 12.5% higher Unaided Awareness, projection 'would improve overall campaign performance by 31%'; global dairy company 13.15% increase in Favorability; national clothing retailer 41.6% higher Store Visitation Rate, projection 'would drive significant brand lift, including 54.5% higher Unaided Aware | supported with qualification | Teads' release of 17 August 2023 (hosted by Adelaide) says its high-attention omnichannel PMPs, bought through Google Display & Video 360, The Trade Desk and Yahoo, include placements with 'an average attention score 31% higher than Adelaide's Teads AU benchmarks', and lists outcome lifts for five unnamed advertisers measured by Adelaide and unnamed 'third-party brand measurement partners' (12.5% higher unaided awareness and 32.7% higher familiarity; 13.15% higher favorability; 41.6% higher store visitation; 455.6% higher site visits and 110.8% higher checkout actions; 8x higher in-store visits and a 74% higher conversion rate) alongside three projections ('would improve overall campaign performance by 31%'; '54.5% higher Unaided Awareness' at a 30 AU threshold; 'could improve overall campaign performance by 67%'), with no periods, samples or baselines. | scope, unit, date or wording narrowed to what the source carries | CUR-C-F09 |
| V-C-11 | Multilocal's homepage asserts '20-40% increase in performance', 'Expect a 30–50% boost', '24–28% boost in yield', '75% Higher CTR than projected', '38% Increase in Video Completion Rate (VCR)' and 'Cut your campaign's carbon footprint by 40–60%' 'compared to traditional programmatic ads', with no comparator (except carbon), period, measurer or method; its Maglr case studies returned 'Account expired'. | Refute-mode verifier for C.json re-opened the cited source | 'a 20-40% increase in performance' (under 'Proven Results'); 'Expect a 30–50% boost' ('Better Campaign Performance'); 'See a 24–28% boost in yield' ('Maximise Revenue'); tile '75%' 'Avg. reduction in cost-effeciency'; tile '75%' 'Higher CTR than projected'; tile '40%+' 'Lift in performance'; tile '38%' 'Increase in Video Completion Rate (VCR)'; 'Cut your campaign's carbon footprint by 40–60% compared to traditional programmatic ads' ('Greener Campaigns'); 'over 150 countries' and 'more than 190 countries' on the same page; also '20+ Years in Advertising & Media' alongside a 2019 founding date and '5 years experience', '35,000+ domains and apps', '62,000 audience data points', 'Global Team of 100+ Curation Experts'. No period, measurer or method for any figure. | supported with qualification | Multilocal's homepage asserts ranges and single figures with no comparator (except carbon), period, measurer or method: '20-40% increase in performance', 'Expect a 30–50% boost', '24–28% boost in yield', '40%+ Lift in performance', '75% Avg. reduction in cost-effeciency' (sic), '75% Higher CTR than projected', '38% Increase in Video Completion Rate (VCR)' and 'Cut your campaign's carbon footprint by 40–60% compared to traditional programmatic ads', while its hosted case studies (Maglr) returned 'Account expired' on 3 November 2026 so none of the case figures surfaced in search could be opened. | scope, unit, date or wording narrowed to what the source carries | CUR-C-F12 |
| V-C-12 | Equativ's Maestro by Equativ curation page publishes 'up to a 40% reduction in planning time', '500+ self-serve clients, 30K+ deals created, and 50 billion impressions delivered', a Forrester SSP Wave (Q4 2024) Curation rating of 5/5 and an unquantified Omnicom Media Group testimonial, but no viewability, MFA, IVT, suitability or attention figure. | Refute-mode verifier for C.json re-opened the cited source | The page's wording is 'Cuts planning time by up to 40%' (the fetch tool flagged that the page does not say '40% reduction'); 'Self-Serve Clients: 500+', 'Deals Created: 30K+', 'Impressions Delivered: 50 Billion'; 'Curation Forrester SSP Wave (Q4 2024)' with '5/5'; Omnicom Media Group testimonial attributed to Sebastián Orduvini, Product Specialist: 'leading to better CTR, increased views and broader reach'; IPG Mediabrands testimonial (Piotr Wiktor) without figures; also 'Analyzes 400K+ sites to surface premium inventory', 'over 500K publishers, apps, and leading media channels', '100th' percentile software NPS and an unattributed 'Best Digital Product Innovation: Winner'. No viewability, MFA, IVT, suitability or attention figure; no period beyond Q4 2024 for the Wave. | supported with qualification | Equativ's Maestro by Equativ curation page publishes operational and scale figures ('Cuts planning time by up to 40%'; '500+' self-serve clients, '30K+' deals created and '50 Billion' impressions delivered, unperiodised), a Forrester SSP Wave (Q4 2024) rating of 5/5 on the Curation criterion, and an Omnicom Media Group testimonial of 'better CTR, increased views and broader reach' without figures, but no viewability, MFA, IVT, suitability or attention figure for a curated deal. | scope, unit, date or wording narrowed to what the source carries | CUR-C-F13 |
| V-C-13 | Across the 20 cohort A, B and C product rows screened, every quantified quality or performance claim is a vendor assertion (one commissioned Forrester TEI measures platform economics); 11 rows carry a quantified claim and 9 none; no promise reaches the H3 supported or contradicted rule, so H3 stays unresolved on this stream's evidence. | Refute-mode verifier for C.json re-opened the cited source | Recount of the stream's own tables against the opened sources: the quality_claim_register has 20 rows (11 A, 8 B, 1 C), 11 coded 'yes' and 9 'no'; the studies table has 30 rows, 29 vendor_assertion and 1 commissioned_measurement; metric tags recount to viewability 3, attention 3, ctr 5, cpm 2, outcome 17. All comparator phrasings quoted in the claim are verbatim in their sources. But the OpenX release names only OpenXBuild in its text ('deploy custom AI models and bidding agents directly within OpenXBuild'; 'OpenXBuild's Real Time Bidstream API'); 'OpenXSelect' appears only in site navigation. The brief places OpenXBuild in the comparator table and workstream J, not cohort A, and the stream's own F02 limitation says the mechanism is 'not a standardised curated deal product'. No OpenXSelect document was opened in this stream, so OpenXSelect was not screened rather than screened with a cla | supported with qualification | Across the 19 cohort A, B and C product rows screened in this stream (10 A, 8 B, 1 C) plus one listed comparator product (OpenXBuild), every quantified quality or performance claim found is a vendor assertion (one commissioned Forrester TEI study measures platform economics, not deal quality): 10 cohort rows carry a quantified claim (PubMatic Connect, Microsoft Curate, Maestro by Equativ, TL Direct, YMax, SmartPMP, Adelaide high-attention PMPs, Lotame Curated Marketplaces, Multilocal curation, Seedtag NeuroX Curation) and 9 carry none in the documents opened (ClearLine, Index Marketplaces, Nexxen Curated Marketplace, FreeWheel Curation Hub, Google Ad Manager Curation, Experian Curated Deals, Comscore-Certified Deal IDs, IAS Curation, Stagwell Curate); OpenXSelect was not screened, because the Hyundai figures in OpenX's release of 16 June 2026 attach to OpenXBuild custom decisioning; none | scope, unit, date or wording narrowed to what the source carries | CUR-C-F17 |
| V-D-14 | Publicis Groupe announced an agreement to acquire Lotame on 6 March 2025 (closing expected early in Q2 2025, no price disclosed) and recorded in its 2025 half-year report the acquisition of 100% of Lotame Solutions, Inc. in April 2025 for a cash consideration of 127 and a provisional goodwill of 96 (table figures; the report's unit header was not captured). | Refute-mode verifier for D.json re-opened the cited source | Publicis half-year financial report 2025, note 3.1: 'In April 2025, acquisition of Lotame Solutions, Inc for 100%, an independent player specialized in data, identity and technology as part of the artificial intelligence strategy of Publicis.' The acquisitions table is headed '(in millions of euros)' and gives for Lotame: cash consideration 127, earn-out commitments none, total consideration transferred 127, total identifiable net assets acquired 31, provisional goodwill 96. The same table lists BR Media 150 and Captiv8 143. The 6 March 2025 release (CUR-D-S04) was not opened here; its date is confirmed by the brief's fact-check record C2-02. [40] | supported with qualification | Publicis Groupe announced an agreement to acquire Lotame on 6 March 2025 (closing expected early in Q2 2025, no price disclosed) and recorded in note 3.1 of its 2025 half-year financial report the acquisition of 100% of Lotame Solutions, Inc. in April 2025 for a cash consideration of EUR 127 million, identifiable net assets of EUR 31 million and provisional goodwill of EUR 96 million (table stated in millions of euros). | scope, unit, date or wording narrowed to what the source carries | CUR-D-F02 |
| V-D-15 | Scope3 announced on 23 September 2026 that it is becoming Apostra, a platform 'built on AdCP' for advertisers and their agents, with the sustainability business continuing as 'Scope3 by Apostra' and the site footer naming Apostra, Inc.; the post says nothing about legal structure, funding or ownership, so the event is coded as a rebrand with unknown legal form. | Refute-mode verifier for D.json re-opened the cited source | Apostra post dated 'September 23, 2026': 'Scope3 is becoming Apostra.'; 'It's built on AdCP, the open protocol we helped create that lets agents talk to each other.'; footer 'Copyright (c) 2026 Apostra, Inc. All rights reserved.' No sentence in the post says the sustainability business continues as Scope3 by Apostra; that name appears only as the footer link 'Looking for Sustainability -> Scope3 by Apostra'. Nothing on funding, ownership, legal structure, investors or a new entity. | supported with qualification | Scope3 announced on 23 September 2026 that it is becoming Apostra, a platform 'built on AdCP' for advertisers and their agents; the post's footer names Apostra, Inc. and links to 'Scope3 by Apostra' for sustainability, but the post itself says nothing about the sustainability business, legal structure, funding or ownership, so the event is coded as a rebrand with unknown legal form. | scope, unit, date or wording narrowed to what the source carries | CUR-D-F05 |
| V-D-16 | Adloox changed hands twice: Scope3 announced its acquisition of Adloox on 5 November 2024 (per Adloox's own post) and Peer39 announced on 3 June 2026 that it had acquired Adloox, with the Adloox team joining Peer39; neither post discloses terms, and Peer39's post does not name the seller, so at the cutoff Adloox is coded as controlled by Peer39 rather than by Apostra (Scope3). | Refute-mode verifier for D.json re-opened the cited source | Adloox post dated 'Nov 5, 2024' under the headline, body: 'On 4 November 2024, Scope3 announced the acquisition of Adloox to bolster the company's media sustainability solutions'; 'Over the coming quarters, both Scope3 and Adloox will be working to integrate our teams and technology'; no terms; no mention of Peer39 or Apostra. Peer39 post 'Published June 3, 2026': 'Peer39 acquired Adloox to give advertisers consistent, independent quality signals across every major platform'; 'The Adloox team is joining Peer39'; Scope3 appears once, in the DV360 FAQ ('until the names have been updated from the Scope3 segments'); no seller, price, terms or closing date. | supported with qualification | Adloox changed hands twice: Scope3 announced its acquisition of Adloox on 4 November 2024 (per Adloox's own post of 5 November 2024) and Peer39 announced on 3 June 2026 that it had acquired Adloox, with the Adloox team joining Peer39; neither post discloses terms, and Peer39's post does not name the seller, so at the cutoff Adloox is coded as controlled by Peer39 rather than by Apostra (Scope3). | scope, unit, date or wording narrowed to what the source carries | CUR-D-F06 |
| V-D-17 | Outbrain Inc. closed its acquisition of Teads from Altice Teads on 3 February 2025 for a cash payment of $625 million (subject to adjustments) and 43.75 million shares of common stock, with two Altice Teads designees to join the board, and renamed itself Teads Holding Co. effective 6 June 2025 (Nasdaq ticker OB to TEAD on or about 10 June 2025); Altice Teads' post-closing percentage is not stated. | Refute-mode verifier for D.json re-opened the cited source | Outbrain 8-K: 'On February 3, 2025, the parties consummated the closing of the Acquisition' under the Share Purchase Agreement 'as amended by the SPA Amendment'; consideration 'a cash payment of $625 million, subject to certain customary adjustments' and '43.75 million shares of Common Stock'; seller 'Altice Teads S.A.'; 'the contemplated appointment of two designees of Altice Teads to the Board'; no post-closing percentage. The 8-K also states: 'On August 1, 2024, Outbrain entered into a Share Purchase Agreement'. The name-change 8-K (CUR-D-S11) was not opened here; the brief's fact-check C2-09 confirms 6 June 2025. [42] | supported with qualification | Outbrain Inc., which entered into a Share Purchase Agreement with Altice Teads S.A. on 1 August 2024, closed its acquisition of Teads on 3 February 2025 for a cash payment of $625 million (subject to adjustments) and 43.75 million shares of common stock, with two Altice Teads designees to join the board, and renamed itself Teads Holding Co. effective 6 June 2025 (Nasdaq ticker OB to TEAD on or about 10 June 2025); Altice Teads' post-closing percentage is not stated. | scope, unit, date or wording narrowed to what the source carries | CUR-D-F10 |
| V-D-18 | DoubleVerify announced on 6 August 2026 an agreement and plan of merger with Neptune BidCo US Inc. and its subsidiary Wallace Merger Sub Inc.; stockholders adopted the agreement at a special meeting on 29 October 2026 (approximately 94% of votes cast, over 75% of outstanding shares), DV expects closing by the first quarter of 2027, has suspended earnings calls and withdrawn guidance, and reported Q3 2026 revenue of $205.6 million (+9%); the transaction was pending at the cutoff and the filings opened do not name Parent's owners or the per-share price. | Refute-mode verifier for D.json re-opened the cited source | Q3 2026 release (8-K exhibit, 30 October 2026): 'Total revenue of $205.6 million, an increase of 9%'; 'On October 29, 2026, DV held a special meeting of stockholders'; 'Approximately 94% of votes cast were in favor of the adoption of the Merger Agreement, representing over 75% of the issued and outstanding shares of DV common stock as of the close of business on September 28, 2026'; 'continues to expect the transactions contemplated by the Merger Agreement to close by the first quarter of 2027'; earnings calls suspended and guidance withdrawn for the pendency; and, contrary to the claim's last clause, 'Parent is the parent company of the Nielsen Company (US), LLC.' Exhibit 99.1 of the 6 August 2026 8-K (the filing the stream cites as CUR-D-S13; the stream opened the 8-K body only): datelined 'New York, NY - August 6, 2026'; 'DoubleVerify shareholders will receive $13.60 per share in cash [51, 52] | supported with qualification | DoubleVerify announced on 6 August 2026 an agreement and plan of merger under which Neptune BidCo US Inc., the parent company of The Nielsen Company (US), LLC, will acquire DV for $13.60 per share in cash (enterprise value approximately $2.15 billion); stockholders adopted the agreement at a special meeting on 29 October 2026 (approximately 94% of votes cast, over 75% of outstanding shares), DV now expects closing by the first quarter of 2027 (the August release said by the end of Q4 2026), has suspended earnings calls and withdrawn guidance, and reported Q3 2026 revenue of $205.6 million (+9%); the transaction was pending at the cutoff. | scope, unit, date or wording narrowed to what the source carries | CUR-D-F11 |
| V-D-19 | Omnicom pro forma combined revenue in excess of $25 billion at completion of the Interpublic acquisition (table row, claim_status 'estimate'). | Refute-mode verifier for D.json re-opened the cited source | The release states 'pro forma combined revenue in excess of $25 billion', a parties' pro forma figure. | supported with qualification | Replace the row's claim_status 'estimate' with 'n/a' (or 'actual' with 'pro forma, parties' figure' in the definition column); 'estimate' is not a value in the schema vocabulary (actual \| ytd \| forecast \| n/a). | scope, unit, date or wording narrowed to what the source carries | CUR-D-F13 (disclosed_take_rates row Omnicom) |
| V-D-20 | Nexxen International Ltd. (renamed from Tremor International in January 2024) cancelled its AIM depositary interests in February 2025 and its shares trade solely on the Nasdaq Global Market under NEXN; its FY2025 20-F names four beneficial owners above 5% (entities affiliated with Mithaq Capital SPC, JB Capital Partners L.P., News Corporation and Toscafund Asset Management LLP) and no controlling shareholder, describes platform fees based on a percentage of spend, flat fees or fixed CPMs, and refers to the Amobee acquisition in 2022 without a price in the portion read. | Refute-mode verifier for D.json re-opened the cited source | Nexxen FY2025 20-F (characters 0 to 200,000 read): rename 'to Nexxen International Ltd., in January 2024'; 'On February 17, 2025, the Company's AIM-listed depository interests representing the Company's old Ordinary Shares' were cancelled and 'on February 18, 2025, the Company's New Ordinary Shares began trading on Nasdaq under the stock ticker "NEXN"'; cover page: Nasdaq Global Market; risk factor: the four largest beneficial owners, 'Mithaq Capital SPC, JB Capital Partners L.P., News Corporation and Toscafund Asset Management LLP', hold 'in the aggregate, approximately 52.3% of our Shares' and 'may be able to exert significant influence'; 'We generate revenue through platform fees based on either (i) a percentage of spend, (ii) flat fees or (iii) fixed CPMs'; Amobee: only 'to fund a portion of the purchase price of the Amobee acquisition in 2022'; 2025 segment figures Video $242.0 mill | supported with qualification | Nexxen International Ltd. (renamed from Tremor International in January 2024) cancelled its AIM depositary interests on 17 February 2025 and its shares have traded solely on the Nasdaq Global Market under NEXN since 18 February 2025; its FY2025 20-F names four beneficial owners above 5% (entities affiliated with Mithaq Capital SPC, JB Capital Partners L.P., News Corporation and Toscafund Asset Management LLP) holding approximately 52.3% of the shares in aggregate, warns that they may exert significant influence, and names no single majority holder; it describes platform fees based on a percentage of spend, flat fees or fixed CPMs, and refers to the Amobee acquisition in 2022 without a price in the portion read. | scope, unit, date or wording narrowed to what the source carries | CUR-D-F18 |
| V-D-21 | Comscore's FY2025 10-K (NASDAQ Global Select Market, SCOR) records a recapitalisation that closed on 29 December 2025, exchanging and retiring all Series B preferred stock and reducing the holders' director designation rights; the portion read names neither the exchanging holders nor any 'controlled company' status, and refers to Proximic by Comscore without any 'curated' or 'Comscore-Certified' wording. | Refute-mode verifier for D.json re-opened the cited source | Comscore FY2025 10-K (first 100,000 characters): 'On September 26, 2025, we entered into separate Stock Exchange Agreements with the holders of our Series B Convertible Preferred Stock'; issuance approved 19 December 2025 and 'subsequently closed on December 29, 2025'; 'The Recapitalization Transaction resulted in the exchange and retirement of all shares of Series B Preferred Stock' and 'a reduction in the holders' director designation rights, among other things'; holders exchanged into 'shares of Series C Convertible Preferred Stock', 'common stock' and 'a fixed cash payment of $2.0 million in June 2028'; the narrative names no holder and has no 'controlled company' statement in that portion (XBRL tags reference Charter Communications Holding Company LLC, Pine Investor LLC, Qurate Retail and a Liberty entity); 'Proximic(R) by Comscore' and 'Activation Solutions (branded as Proximic)' a | supported with qualification | Comscore's FY2025 10-K (NASDAQ Global Select Market, SCOR) records a recapitalisation agreed on 26 September 2025 and closed on 29 December 2025 that exchanged and retired all Series B preferred stock held by Charter Communications Holding Company, LLC, Liberty Broadband Corporation and Pine Investor, LLC (a Cerberus-advised vehicle), each receiving 4,223,621 Series C preferred shares, 3,286,825 common shares and a fixed $2.0 million cash payment due in June 2028, and reduced the holders' board rights to one designee each on a seven-member board with a 16.66% voting threshold; the 10-K portion read names no 'controlled company' status and refers to Proximic by Comscore without any 'curated' or 'Comscore-Certified' wording. | scope, unit, date or wording narrowed to what the source carries | CUR-D-F22 |
| V-D-22 | On the strict H5 rule (control documented from filings or official releases), the curation data suppliers with an opened ownership record that were documented as independently controlled fell from four of eleven in 2023 (Lotame, LiveRamp, IAS, DoubleVerify; six unknown; Oracle not independent) to two of ten active suppliers at the cutoff (IAS under Novacap funds, DoubleVerify pending sale; three unknown; Oracle exited), because Lotame and LiveRamp passed to Publicis; on the looser reading that treats a private supplier with no documented operating-group parent as independent, the fall is from ten of eleven to four of ten, with Audigent (Experian), Lotame and LiveRamp (Publicis), Semasio (Samba TV) and Adloox (Scope3, then Peer39) leaving independence and Oracle leaving the market. | Refute-mode verifier for D.json re-opened the cited source | Checked against the stream's own data tables rather than an external source (the finding is derived from them). supplier_register_2023, role = curation data supplier: 11 rows; independent yes 4 (Lotame, LiveRamp, IAS, DoubleVerify), no 1 (Oracle), unknown 6 (Audigent, Semasio, Scope3, Adloox, Comscore, Multilocal). supplier_register_cutoff: 11 rows; yes 2 (IAS, DoubleVerify pending), no 5 (Audigent, Lotame, LiveRamp, Semasio, Adloox), unknown 3 (Apostra, Comscore, Multilocal), exit 1 (Oracle), so 10 active. The strict counts (four of eleven to two of ten) match the tables. The loose count treats unknown rows with no documented operating-group parent as independent: 2023 gives 4 + 6 = 10 of 11, as stated; the cutoff gives 2 + 3 = 5 of 10, not 'four of ten'. The finding's own list of leavers (Audigent, Lotame, LiveRamp, Semasio, Adloox: five) applied to ten also gives five. | contradicted | On the strict H5 rule (control documented from filings or official releases), the curation data suppliers with an opened ownership record that were documented as independently controlled fell from four of eleven in 2023 (Lotame, LiveRamp, IAS, DoubleVerify; six unknown; Oracle not independent) to two of ten active suppliers at the cutoff (IAS under Novacap funds, DoubleVerify pending sale to Nielsen; three unknown; Oracle exited), because Lotame and LiveRamp passed to Publicis; on the looser reading that treats a private supplier with no documented operating-group parent as independent, the fall is from ten of eleven to five of ten, with Audigent (Experian), Lotame and LiveRamp (Publicis), Semasio (Samba TV) and Adloox (Scope3, then Peer39) leaving independence and Oracle leaving the market. | the source does not carry the claim as first written | CUR-D-F25 |
| V-D-23 | None of the seven annual reports read (Magnite, PubMatic, Criteo, Comscore, The Trade Desk and Viant 10-Ks for FY2025; Nexxen's FY2025 20-F) discloses, in the portion read, a numeric take rate or a deal, PMP or curated share of its own activity; each describes its compensation as a fee that is generally a percentage of spend (with flat fees or fixed CPMs at Nexxen and Viant, and negotiated SSP, technology and DSP fees at Criteo), so filings supply no SSP-level series for H1 and no product-level charge for H2b. | Refute-mode verifier for D.json re-opened the cited source | In the portions opened here (first 100,000 characters of the Magnite, PubMatic, Viant and Comscore 10-Ks; characters 0 to 200,000 of the Nexxen 20-F) the fee-basis sentences are as quoted and no numeric take rate or deal, PMP or curated share appears; Magnite's 10-K uses 'take-rate' qualitatively ('biddable transactions ... carry a higher take-rate compared to reserve auctions'). The Criteo and The Trade Desk 10-Ks were not re-opened. Each portion read ends before the MD&A and the financial statements (Magnite 347,401; PubMatic 333,216; Viant 334,303; Comscore 271,377; Nexxen 339,143 characters in total), so the absence is bounded to the Business and risk-factor sections. | supported with qualification | None of the seven FY2025 annual filings read (Magnite, PubMatic, Criteo, Comscore, The Trade Desk and Viant 10-Ks; Nexxen's 20-F) discloses, in the Business and risk-factor sections read (the first 100,000 characters of each 10-K and 200,000 of the 20-F, which end before the MD&A and financial statements), a numeric take rate or a deal, PMP or curated share of its own activity; each describes its compensation as a fee that is generally a percentage of spend (with flat fees or fixed CPMs at Nexxen and Viant, and negotiated SSP, technology and DSP fees at Criteo), so those sections supply no SSP-level series for H1 and no product-level charge for H2b. | scope, unit, date or wording narrowed to what the source carries | CUR-D-F26 |
| V-E-24 | Microsoft Curate's documentation (ms.date 2026-02-18 and 2025-10-21) sets the Auction Type of every curated deal to 'Open Auction', states that curated transactions are not guaranteed deals and never override seller floor or ad-quality settings, and requires curated bids to compete, net of fees, in the seller's final auction against all other deal and open-exchange bids. | Refute-mode verifier for E.json re-opened the cited source | Create page (ms.date 2026-02-18, updated_at 2026-02-18T04:56Z): 'Auction Type - Auction Type is always set to "Open Auction" for curators.' Curated Deals page (ms.date 2025-10-21): 'Curated transactions are not guaranteed deals and will not override any seller account configurations, including ad quality settings or floor price settings.' and 'All bids submitted for curated transactions will compete in the seller's final auction with all other deal and open exchange bids, net of all applicable fees and charges.' The fees FAQ on the same page adds an exception the stream did not carry: 'If you are a seller, all bids for curated transactions will be submitted to your final auction net of such fees or charges unless you agree otherwise with the applicable curator'. | supported with qualification | Microsoft Curate's documentation (ms.date 2026-02-18 and 2025-10-21) sets the Auction Type of every curated deal to 'Open Auction', states that curated transactions are not guaranteed deals and never override seller floor or ad-quality settings, and has curated bids compete in the seller's final auction against all other deal and open-exchange bids net of fees, which the fees FAQ makes the default 'unless you agree otherwise with the applicable curator'. | scope, unit, date or wording narrowed to what the source carries | CUR-E-F01 |
| V-E-25 | Curators on Microsoft Curate take no margin by default; a margin set as a percentage of the buyer's bid or a fixed CPM is deducted, together with Microsoft's curator tech fee, before the net bid enters the seller's auction; fees apply only when the curated bid wins; the platform enforces a maximum margin rate whose value the pages do not state; and no page opened says whether the buyer sees the margin. | Refute-mode verifier for E.json re-opened the cited source | Margins page (ms.date 2025-10-21): 'Curators do not take margin by default and therefore must select a margin type and set a non-zero margin value if they want to profit on the transaction.'; the three ordered steps (buyer fees removed = Curator Revenue; Curator Margin deducted and Curator Tech Fees taken; remainder = Curator Net Media Cost 'and is the bid sent to the seller auction, where seller fees are deducted before all bids are ranked in the final auction'); 'the Microsoft Curate platform enforces a maximum margin rate when curators configure a curated deal' with no value given; no statement on buyer visibility. Curated Deals page: fees apply 'only if the bid for the curated deal wins the impression in the seller's final auction'. The floors page qualifies the margin rule: on seller deals with a fixed floor 'your curator margin is effectively the minimum margin you may earn' and 't | supported with qualification | Curators on Microsoft Curate take no margin by default; a margin set as a percentage of the buyer's bid or a fixed CPM is deducted, together with Microsoft's curator tech fee, before the net bid enters the seller's auction, and on seller deals with a fixed floor the configured margin is a minimum because the curator takes the difference between the curator and seller floors; fees apply only when the curated bid wins; the platform enforces a maximum margin rate whose value the pages do not state; and no page opened says whether the buyer sees the margin. | scope, unit, date or wording narrowed to what the source carries | CUR-E-F03 |
| V-E-26 | A Microsoft Curate curated deal is built for one selected buyer that cannot be changed after the line item is saved, the same seller inventory remains biddable in the open exchange, and the documentation names Microsoft Invest and 'all major DSPs' as buying routes without naming any third-party DSP. | Refute-mode verifier for E.json re-opened the cited source | Create page: 'You cannot change the selected buyer after the curated deal line item is created and saved.' Curated Deals page: 'With curation, buyers can still bid on seller inventory via open exchange, but buyers that target curated deals may also bid on seller inventory via curated transactions.' About page (ms.date 2025-10-21): 'Curated deals can be bought via Microsoft Invest and all major DSPs.'; no third-party DSP is named on any of the six Curate pages. Microsoft's blog post of May 14, 2025 (opened by the verifier; the stream relied on the brief): 'we will no longer support media buying through our DSP, Microsoft Invest, starting on February 28, 2026.' and 'we will continue to support access to Microsoft and partner inventory through third-party DSPs'; the same post describes Microsoft Curate as continuing ('through Microsoft Curate, we're empowering publishers to monetize proprie | supported with qualification | A Microsoft Curate curated deal is built for one selected buyer that cannot be changed after the line item is saved, the same seller inventory remains biddable in the open exchange, and the documentation (ms.date 2025-10-21) names Microsoft Invest and 'all major DSPs' as buying routes without naming any third-party DSP, although Microsoft said on 14 May 2025 that it would 'no longer support media buying through our DSP, Microsoft Invest, starting on February 28, 2026', so at the cutoff the only named route is retired by Microsoft's own statement and no documented route remains. | scope, unit, date or wording narrowed to what the source carries | CUR-E-F05 |
| V-E-27 | On Google Ad Manager the curation partner enters a package or segment fee as a percentage of media or a CPM, may mark it visible or hidden from the buyer, and Google pays the partner out of the bid price; the Ad Manager publisher page calls the fee 'fully transparent to the buyer' while the Authorized Buyers agency page says the agency 'may or may not see the fee value depending on the curation partner's visibility settings'. | Refute-mode verifier for E.json re-opened the cited source | Curation partners page: 'Enter the "Fee" as either percentage of media or CPM.' and '(Optional) Mark your package fees as visible or hidden from buyer.' Publisher page: 'Curation partners define the fee of the data segment or curated package.', 'It is a single fee that is fully transparent to the buyer.' and, in full, 'Google then handles the fee payment to the curation partner (taken out of the bid price before it enters the auction).' The stream's quoted passage drops 'before it enters the auction'. Agency page: 'The agency may or may not see the fee value depending on the curation partner's visibility settings.' and 'The agency doesn't pay anyone directly, as Google manages fee payments out of the bid price and pays the curation partner.' | supported with qualification | On Google Ad Manager the curation partner enters a package or segment fee as a percentage of media or a CPM, may mark it visible or hidden from the buyer, and Google pays the partner out of the bid price, which the publisher page says is 'taken out of the bid price before it enters the auction'; the Ad Manager publisher page calls the fee 'fully transparent to the buyer' while the Authorized Buyers agency page says the agency 'may or may not see the fee value depending on the curation partner's visibility settings'. | scope, unit, date or wording narrowed to what the source carries | CUR-E-F11 |
| V-E-28 | Index's Marketplaces product page offers media owners incremental demand 'without any added platform fees', claims 'Real-time dashboards and audit logs' and intelligence applied 'in the opening milliseconds of the auction', and states no deal type, priority, floor, exclusivity rule, fee amount or DSP name. | Refute-mode verifier for E.json re-opened the cited source | The phrase 'without any added platform fees' sits in the page intro, which addresses the sell side generally: 'Activate premium data and other powerful media solutions across the entire open internet to create high-performing deal packages—all without any added platform fees.' The 'For media owners: Earn more from every impression' section uses a variant, 'without added fees or extra work'. 'Real-time dashboards and audit logs' is a heading under 'Why build an Index Marketplace?', described as 'For performance tracking and visibility.' 'Apply intelligence in the opening milliseconds of the auction to achieve greater scale and performance.' No deal type, priority, floor, fee amount or DSP name; 'exclusive' appears only in partner-news headlines. | supported with qualification | Index's Marketplaces product page offers sell-side users deal packaging 'without any added platform fees' and media owners incremental demand 'without added fees or extra work', claims 'Real-time dashboards and audit logs' and intelligence applied 'in the opening milliseconds of the auction', and states no deal type, priority, floor, exclusivity rule, fee amount or DSP name. | scope, unit, date or wording narrowed to what the source carries | CUR-E-F22 |
| V-E-29 | FreeWheel's Curation Hub pages describe a self-service inventory curation and deal management tool that builds data-enriched packages from FreeWheel publisher partners using standardised content metadata and audience signals and activates them 'as deals across any DSP', naming only FreeWheel Buyer Cloud, with reconciled reporting; the pages state no deal type, auction type, priority, floor, exclusivity rule, fee or the SSP on which the marketplace runs. | Refute-mode verifier for E.json re-opened the cited source | 'This self-serve advertising platform simplifies inventory curation and deal management'; 'Self-service tools to discover valuable impressions, curate supply, and activate deals across any DSP.'; 'activate them seamlessly as deals across any demand-side platform (DSP) including Buyer Cloud.'; 'Standardized content metadata and audience signals integrated for scalable, differentiated packaging and targeting.'; 'reconciled reporting to maximize performance and safeguard profitability.'; 'Direct access to FreeWheel publisher partners and the marketplace chosen by major publishers'; and 'Integration into unified ad decisioning among FreeWheel publishers.' No deal type, auction type, priority, floor, exclusivity or fee statement; no SSP named. [260] | supported with qualification | FreeWheel's Curation Hub pages describe a self-service inventory curation and deal management tool that builds data-enriched packages from FreeWheel publisher partners using standardised content metadata and audience signals and activates them 'as deals across any DSP', naming only Buyer Cloud, with reconciled reporting and 'Integration into unified ad decisioning among FreeWheel publishers'; the pages state no deal type, auction type, priority, floor, exclusivity rule or fee and do not name the SSP. | scope, unit, date or wording narrowed to what the source carries | CUR-E-F27 |
| V-E-30 | Criteo's Commerce Grid page calls the product 'the world's first commerce SSP', trades 'via open exchange and via deal IDs' with 'seamless PMP activation', asserts that it 'Prioritizes high-value deals in auctions, reducing unnecessary traffic', that 'Fees only apply if curated deals win impressions', activation 'through any DSP' with '60+ third party DSPs plugged directly', and consented Commerce Audiences data; it states no auction type, floor, exclusivity rule, fee rate or DSP name, and the Commerce Grid documentation page returned 404. | Refute-mode verifier for E.json re-opened the cited source | Product page: 'Say hello to the world's first commerce SSP'; 'via open exchange and via deal IDs'; 'Create high-value deals with seamless PMP activation'; 'Prioritizes high-value deals in auctions, reducing unnecessary traffic for better results.'; 'Fees only apply if curated deals win impressions, ensuring fair and visible transactions.'; 'Make off-site audience reach programs available through any DSP'; 'premium demand from 60+ third party DSPs plugged directly into Commerce Grid'; 'Built on consented data, Commerce Audiences ensure privacy'; no DSP named, no auction type, floor or fee rate. The stream's five-step URL was not retried, but the same knowledge base has live pages the stream did not reach. Curation FAQ (no date): 'a tech fee charged by Criteo to the third-party curator' and 'a margin to be charged by the curator for their inventory enhancement'; 'all bids for curated trans | supported with qualification | Criteo's Commerce Grid product page calls the product 'the world's first commerce SSP', trades 'via open exchange and via deal IDs' with 'seamless PMP activation', asserts that it 'Prioritizes high-value deals in auctions, reducing unnecessary traffic', that 'Fees only apply if curated deals win impressions', activation 'through any DSP' with '60+ third party DSPs plugged directly', and consented Commerce Audiences data, and names no DSP, auction type, floor, exclusivity rule or fee rate; the five-step deal guide the stream tried returned 404, but Commerce Grid's knowledge base is live and documents a Criteo tech fee charged to the curator and a curator margin, curated bids submitted to the final auction net of those fees and charged only on a win, deprioritisation of low-value curated deals in auction processing, one DSP seat per demand deal, and Deal Sync to DV360 and The Trade Desk. | scope, unit, date or wording narrowed to what the source carries | CUR-E-F29 |
| V-E-31 | Across the 17 product-by-transaction-type rows in the access register (15 products; Google Ad Manager coded for Private Auction, Preferred Deal and Marketplace Packages), auction type is documented for 5 rows (4 default, 1 configurable), priority for 4 (all default), floor treatment for 6 (all configurable), buyer exclusivity for 7 (3 default, 4 configurable) and request sampling for 1 (default), and 9 of the 15 products have every attribute unknown. | Refute-mode verifier for E.json re-opened the cited source | The 85 access_register rows were checked against the opened pages: every passage cited for Microsoft Curate, Google Ad Manager (three types), Index, Maestro, TL Direct and Commerce Grid is on the cited page, and the access_register_counts arithmetic is internally consistent (each attribute sums to 17). Two things qualify the counts. (1) Commerce Grid's priority row is coded 'default' on the marketing phrase 'Prioritizes high-value deals in auctions'; the Commerce Grid knowledge base describes the mechanism as deprioritising 'unnecessary information' so that 'each auction focuses on processing the most valuable curated deals', which is request sampling in the brief's terms, and its glossary says 'Only one DSP seat is permitted per demand deal.' (2) The Index first-price passage in the auction_type row is carried in substance; the fetch returned the glossary wording ('the winner pays the h | supported with qualification | Across the 17 product-by-transaction-type rows in the access register (15 products; Google Ad Manager coded for Private Auction, Preferred Deal and Marketplace Packages), on the pages the stream opened auction type is documented for 5 rows (4 default, 1 configurable), priority for 4 (all default), floor treatment for 6 (all configurable), buyer exclusivity for 7 (3 default, 4 configurable) and request sampling for 1 (default), and 9 of the 15 products have every attribute unknown; Commerce Grid's knowledge base, which the stream did not reach, documents deprioritisation of low-value curated deals in auction processing and one DSP seat per demand deal, so with it priority is documented for 3 rows, request sampling for 2 and buyer exclusivity for 8 (4 default, 4 configurable), and the all-unknown count stays at 9. | scope, unit, date or wording narrowed to what the source carries | CUR-E-F34 |
| V-E-32 | Under the H4 coding rule, the essay's claim that a curated deal carries no default priority, floor advantage or exclusive supply holds for Microsoft Curate (open-auction packaging, seller floors respected, no margin by default, one buyer per deal but the same supply biddable in the open exchange), fails for a Google Ad Manager curated package built on Preferred Deals (fixed-price and pre-auction by transaction-type definition) and on Commerce Grid's own assertion that it 'Prioritizes high-value deals in auctions', and is unresolved for the other twelve products. | Refute-mode verifier for E.json re-opened the cited source | The Microsoft and Google branches rest on verbatim passages (see CUR-E-F01, F05, F10). The Commerce Grid branch rests on a marketing sentence; the Commerce Grid knowledge base describes the mechanism as deprioritising low-value curated deals in auction processing ('deprioritize unnecessary information'; 'ensure each auction focuses on processing the most valuable curated deals'), which under the brief is request sampling, not a priority advantage for the curated deal over open-auction bids. | supported with qualification | Under the H4 coding rule, the essay's claim that a curated deal carries no default priority, floor advantage or exclusive supply holds for Microsoft Curate (open-auction packaging, seller floors respected, no margin by default, one buyer per deal but the same supply biddable in the open exchange), fails for a Google Ad Manager curated package built on Preferred Deals (fixed-price and pre-auction by transaction-type definition), is unresolved for Commerce Grid, whose 'Prioritizes high-value deals in auctions' its knowledge base describes as deprioritising low-value curated deals in auction processing (request sampling) rather than ranking them above open-auction bids, and is unresolved for the other twelve products. | scope, unit, date or wording narrowed to what the source carries | CUR-E-F35 |
| V-E-33 | For the curation-fee series, three of fifteen cohort A products (Microsoft Curate, Google Ad Manager Curation, Index Marketplaces) document the fee type, base, collector and payment route without amounts; three (TripleLift, Index's marketplaces page for media owners, Commerce Grid) make a general assertion of no or win-only fees without a schedule; nine disclose nothing on the pages opened or could not be opened; and buyer visibility of the fee is curator-controlled on Google and Index and unstated on Microsoft. | Refute-mode verifier for E.json re-opened the cited source | The Microsoft, Google and Index codings are verbatim-supported (see CUR-E-F03, F11, F18, F20). TripleLift's 'No TripleLift curation fees' is on the page with no schedule. Index's 'without any added platform fees' addresses the sell side generally, not media owners specifically (CUR-E-F22). Commerce Grid's product page carries only the win-only assertion, but its knowledge base documents the fee types (a Criteo tech fee charged to the curator; a curator margin), the base (bids submitted 'net of such fees or charges'), the collector and the win-only rule, with no amounts; Deal Sync adds that the 'Curation fee amount can be adjusted' after synchronisation, and nothing states buyer visibility. Maestro's help page names 'additional costs or curator margins' that depend on 'your access level (with or without margin)' without a base. | supported with qualification | For the curation-fee series, four of fifteen cohort A products (Microsoft Curate, Google Ad Manager Curation, Index Marketplaces and, through its knowledge base rather than its product page, Commerce Grid) document the fee type, base, collector and payment route without amounts; two (TripleLift, and Index's marketplaces page addressed to the sell side) make a general assertion of no fees without a schedule; nine disclose nothing on the pages opened or could not be opened; and buyer visibility of the fee is curator-controlled on Google and Index and unstated on Microsoft and Commerce Grid. | scope, unit, date or wording narrowed to what the source carries | CUR-E-F36 |
| V-E-34 | No cohort A product names more than one DSP in text on the pages opened: Microsoft names Microsoft Invest plus 'all major DSPs', FreeWheel names Buyer Cloud plus 'any DSP', Commerce Grid counts '60+ third party DSPs', Google refers to 'all DSPs integrated with agency deals', Index to 'all DSPs', Sovrn shows DSP logos only as image file names, and the other eight name none. | Refute-mode verifier for E.json re-opened the cited source | On the vendor pages the stream opened the claim holds: 'Curated deals can be bought via Microsoft Invest and all major DSPs.'; 'across any demand-side platform (DSP) including Buyer Cloud.'; '60+ third party DSPs plugged directly into Commerce Grid'; 'all DSPs integrated with agency deals'; 'the ability to reach all DSPs'. Commerce Grid's knowledge base, not opened by the stream, names two third-party DSPs with documented activation functionality: 'Deal Sync is available to a limited number of integrated DSP partners.', 'One hour after deal creation for a deal to surface in the DV360 UI', 'Three hours after deal creation for a deal to surface in the TTD UI', and the FAQ's 'DSP Name (Ex: DV360, TTD)'. Microsoft's post of 14 May 2025 retires the Invest route from 28 February 2026. | supported with qualification | On the vendor pages the stream opened no cohort A product names more than one DSP in text: Microsoft names Microsoft Invest (whose media buying Microsoft said on 14 May 2025 it would no longer support from 28 February 2026) plus 'all major DSPs', FreeWheel names Buyer Cloud plus 'any DSP', Commerce Grid's product page counts '60+ third party DSPs', Google refers to 'all DSPs integrated with agency deals', Index to 'all DSPs', Sovrn shows DSP logos only as image file names, and the other eight name none; Commerce Grid's knowledge base, however, documents Deal Sync to DV360 and The Trade Desk with deal surfacing one and three hours after creation, which is two named third-party activation routes with documented functionality. | scope, unit, date or wording narrowed to what the source carries | CUR-E-F38 |
| V-E-35 | On every page opened for the fifteen products, the only documented post-purchase controls belong to the curator or publisher (Microsoft: margins editable at any time, buyer fixed; Google: fee amount changeable on live segments, packages editable except type, archiving cuts off subscribed buyers; Index: partner-chosen fee disclosure), and no page documents change notification to the buyer, an audit right, data exportability for the buyer or a remedy for misdescribed inventory or charges. | Refute-mode verifier for E.json re-opened the cited source | All four cited passages are verbatim on the pages ('Margins can be edited at any time for a deal.'; 'For live data segments, you can change the fee amount, but you cannot change the fee type.'; 'any buyers currently transacting on packages you archive will no longer be able to transact.'; 'Partners can choose the level of fee transparency that they disclose for reporting metrics.'). Commerce Grid's Deal Sync page adds a curator-side control of the same kind: after synchronisation 'Curation fee amount can be adjusted'. No page opened documents buyer notification, audit, export or remedies. | supported with qualification | On every page opened for the fifteen products, the only documented post-purchase controls belong to the curator or publisher (Microsoft: margins editable at any time, buyer fixed; Google: fee amount changeable on live segments, packages editable except type, archiving cuts off subscribed buyers; Index: partner-chosen fee disclosure; Commerce Grid: curation fee amount adjustable after a deal is synced to the DSP), and no page documents change notification to the buyer, an audit right, data exportability for the buyer or a remedy for misdescribed inventory or charges. | scope, unit, date or wording narrowed to what the source carries | CUR-E-F39 |
| V-F-36 | Audigent's only pricing statement names the charge types and nothing else: 'PMP pricing is determined by bid floors and curation fees only', with no amount, rate, base, payer, collector or period on the pages read on 3 November 2026. | Refute-mode verifier for F.json re-opened the cited source | 'PMP pricing is determined by bid floors and curation fees only.' appears verbatim on the homepage and on the CognitivePMP Marketplace page (under 'Better Pricing'), with no amount, rate, base, payer, collector or period. The SmartPMP product page, which the stream also opened, additionally describes SmartPMPs as 'more fairly priced PMP alternatives' under a 'Better Pricing' heading, a general pricing assertion. | supported with qualification | Audigent's only statement naming charge types is 'PMP pricing is determined by bid floors and curation fees only' (homepage and CognitivePMP Marketplace page), with no amount, rate, base, payer, collector or period; the SmartPMP page adds only a general assertion ('more fairly priced PMP alternatives') on the pages read on 3 November 2026. | scope, unit, date or wording narrowed to what the source carries | CUR-F-F03 |
| V-F-37 | Audigent's marketplace page offers 'Search 1000+ Comscore CognitivePMPs' built on Comscore Predictive Audiences and says the approach is used 'across 100,000+ campaigns each month'; both are unsourced, unperiodised vendor counts. | Refute-mode verifier for F.json re-opened the cited source | 'Search 1000+ Comscore CognitivePMPs™' is the description text of the 'Comscore CognitivePMP™ Marketplace' entry in the site-wide top navigation (FOR ADVERTISERS); it is not in the marketplace page's hero, headings or buttons, and the same string appears in the navigation of the SmartPMP page. 'Comscore Predictive Audiences' likewise appears only as the navigation description of the CognitivePMP link. 'across 100,000+ campaigns each month' is body text on the marketplace page and on the homepage. 'Curated deals by their very nature are enriched with first-party, contextual, and cognitive data.' is on the marketplace page under 'Rich With Data'. No date or copyright year on any of the three pages. | supported with qualification | Audigent's site-wide navigation describes the Comscore CognitivePMP Marketplace as 'Search 1000+ Comscore CognitivePMPs™' and the CognitivePMP link as 'Comscore Predictive Audiences', and the marketplace page says the approach is used 'across 100,000+ campaigns each month'; both are unsourced, unperiodised vendor counts (listed deal IDs and company-wide campaigns) that score as context only. | scope, unit, date or wording narrowed to what the source carries | CUR-F-F04 |
| V-F-38 | Adelaide's undated resource page says its high-AU PMPs run across PubMatic, Microsoft Advertising (Xandr), TripleLift, Magnite and OpenX with high-, medium- and low-AU sets as separate deal IDs, no thresholds and no fee statement, and performance claims of up to 74% (Teads), 77.3% unaided awareness (Kantar) and up to 69% (Audi) whose methods are not on the page; counterevidence says Index Exchange is absent from Adelaide's own roster on the page. | Refute-mode verifier for F.json re-opened the cited source | The SSP sentence reads 'available for digital display, OLV, and CTV across SSPs, including PubMatic, Microsoft Advertising (Xandr), TripleLift, Magnite, and OpenX.' 'Placements are grouped into high-, medium-, and low-AU sets, representing varying levels of quality.' Figures: Teads 'up to 74% higher conversion rates, 111% higher checkout rates, and 456% higher site visits'; Kantar 'A third-party Kantar study proved that attention PMPs drove 77.3% higher unaided awareness and a 49% increase' (consideration); Audi 'up to 69% higher programmatic conversions while reducing cost per conversion by 53%' and 'a 60% increase in cumulative conversion rate on both Open Exchange and PMP deals versus standard bidding tactics'. DSPs: 'Teads clients have seen impressive results using high-AU PMPs across Google Display & Video 360, The Trade Desk, and Yahoo DSP'; 'DSPs like Yahoo, Adobe, and Viant have | supported with qualification | Adelaide's undated resource page (© 2019–2026) says its high-AU PMPs run 'across SSPs, including PubMatic, Microsoft Advertising (Xandr), TripleLift, Magnite, and OpenX', with placements 'grouped into high-, medium-, and low-AU sets' as separate deal IDs, no thresholds and no fee statement ('a fair price' only), and performance claims of up to 74% higher conversion rates, 111% higher checkout rates and 456% higher site visits (Teads), 77.3% higher unaided awareness and 49% higher consideration (a 'third-party Kantar study', funding unstated), and up to 69% higher conversions, 53% lower cost per conversion and a 60% cumulative conversion-rate increase 'versus standard bidding tactics' (Audi), none with a method on the page; Index Exchange appears on the page only in a logo list, not in the SSP sentence. | scope, unit, date or wording narrowed to what the source carries | CUR-F-F08 |
| V-F-39 | Multilocal's homepage claims 'a 20-40% increase in performance', a '40–60%' carbon-footprint cut and a stat block of 75%, 75%, 40%+ and 38% (VCR) with no source, sample or method, and describes its data as a Media Quality Index over '35,000+ domains and apps' and '62,000 audience data points'. | Refute-mode verifier for F.json re-opened the cited source | 'Our curated campaigns deliver a 20-40% increase in performance'; 'Cut your campaign's carbon footprint by 40–60% compared to traditional programmatic ads.'; stat block: '75%' 'Higher CTR than projected', '75%' 'Avg. reduction in cost-effeciency' (sic), '40%+' 'Lift in performance', '38%' 'Increase in Video Completion Rate (VCR)'; 'our proprietary database, powered by our Media Quality Index, covering 35,000+ domains and apps' 'and 62,000 audience data points'; 'ensuring zero chance of your ads'. The page also claims 'Expect a 30–50% boost through smarter, sharper targeting and relentless optimisation.' (advertisers and agencies) and 'See a 24–28% boost in yield' (media owners), and names no source, sample or method for any figure. Footer '© Multilocal Media. All rights reserved.' with no year. | supported with qualification | Multilocal's undated homepage claims 'a 20-40% increase in performance', a '30–50% boost' for advertisers and agencies, a '24–28% boost in yield' for media owners, a '40–60%' carbon-footprint cut 'compared to traditional programmatic ads' and a stat block of 75% (higher CTR than projected), 75% (average reduction in cost-efficiency), 40%+ (lift in performance) and 38% (VCR increase), with no source, sample or method, and describes its data as a Media Quality Index 'covering 35,000+ domains and apps' and '62,000 audience data points' without stating where the audience data comes from. | scope, unit, date or wording narrowed to what the source carries | CUR-F-F13 |
| V-F-40 | Across the twelve cohort B register seeds, no vendor page discloses a curation fee with amount or rate, base, payer, collector and period: one names charge types only (Audigent), one discloses a data-fee amount with base (Adloox), two make general assertions (Lotame, Seedtag), one is request-only (Dstillery) and seven state nothing. | Refute-mode verifier for F.json re-opened the cited source | The headline holds: no page read discloses amount or rate, base, payer, collector and period together. The per-vendor coding is uneven: Adelaide's page says attention-based algorithms 'ensure advertisers secure high-quality media for a fair price', a general pricing assertion of the same kind as Seedtag's 'Lower effective CPMs' and Lotame's 'data/tech fees' line, yet Adelaide is counted among the seven that 'state nothing'; the stream's own table records it as 'none ('fair price' adjective)'. Audigent's SmartPMP page adds 'more fairly priced PMP alternatives' beside its charge-type sentence. Seedtag's page also mentions 'margin logic' in its deal-building flow. Experian, Comscore (both releases), IAS, Multilocal (homepage and cureon), Semasio (homepage) and Samba (homepage and Audiences) carry no fee, price or pricing statement, as claimed. | supported with qualification | Across the twelve cohort B register seeds, no vendor page discloses a curation fee with amount or rate, base, payer, collector and period: one names charge types only (Audigent: 'bid floors and curation fees only', beside a 'more fairly priced' assertion), one discloses a data-fee amount with base but no payer, collector or period (Adloox: $0.15 CPM segments inside PubMatic auction packages), three make general assertions only (Lotame 'data/tech fees' cut; Seedtag 'Lower effective CPMs' and 'margin logic'; Adelaide 'a fair price'), one is request-only (Dstillery) and six state nothing (Experian, Comscore, IAS, Multilocal, Semasio, Samba). | scope, unit, date or wording narrowed to what the source carries | CUR-F-F26 |
| V-F-41 | No cohort B product documents the six provenance fields per segment on the pages read; the best cases name a source type and provider, and none states collection method, intended use, permissions or legal basis. Derivation: 12 products; 0 with all six fields; 7 with source type and a provider or named signal (Audigent, Comscore, Adelaide, IAS, Seedtag, Adloox, Samba); 3 with source type only (Experian, Lotame, Multilocal); 1 with a 2022 statement only (Dstillery); 1 n/e (Semasio, 404). | Refute-mode verifier for F.json re-opened the cited source | Every page read matches the stream's quotations, and none states a collection method, intended use, permissions or legal basis per segment, so the headline (0 of 12 with all six fields) holds. The split in the derivation does not match the stream's own tables: Samba is counted in the derivation among the seven 'with source type and a provider or named signal' but the curator_products row codes it 'source type only', and Semasio is 'n/e (404)' in the derivation but graded D with 'semantic page profiles and audience profiles from external ID sources' (homepage) in curator_capabilities and 'source type only' in curator_products. On the pages, Samba names a source type (smart-TV viewership, web behaviour) and no provider; Semasio's homepage names source types for its targeting products and no ID source. By the tables, the split is 6 with a provider or named signal (Audigent, Comscore, Adelai | supported with qualification | No cohort B product documents the six provenance fields (source, provider, collection method, intended use, permissions, legal basis) per segment on the pages read on 3 November 2026: six name a source type and a provider or named signal (Audigent's named partners and 'opt-in' claim; Comscore's MMX and VMX rankings with Proximic; Adelaide's AU; IAS's and Seedtag's own contextual classifiers; Adloox's attention score), five name a source type only (Experian, Lotame, Multilocal, Samba, Semasio's homepage), one has a 2022 statement only (Dstillery), and none states collection method, intended use, permissions or legal basis. | scope, unit, date or wording narrowed to what the source carries | CUR-F-F27 |
| V-F-42 | Every quality figure found on the cohort B pages is a vendor or partner assertion without a published method, and this review found no independent measurement with a published method of any of the twelve products' deals against a pre-curation comparator; derivation: 17 claims, 17 vendor or partner assertions, 0 with published method. | Refute-mode verifier for F.json re-opened the cited source | Every figure in the quality_claims table is on its cited page with the wording recorded, none with a method, sample or period, so the conclusion stands. The count is short: the pages carry further numeric claims the table omits: Adelaide's 53% lower cost per conversion, 111% higher checkout rates, 456% higher site visits (Teads) and a 60% cumulative conversion-rate increase 'versus standard bidding tactics' (Audi); Multilocal's '30–50% boost' and '75%' average reduction in cost-efficiency; Semasio's '$28' peak ROAS; Peer39's '40% more coverage than other contextual providers' (a candidate page, like Proximic's +96%, which the table does include). With those, the table would hold at least 25 claims, all vendor or partner assertions without a published method. | supported with qualification | Every quality or performance figure found on the cohort B and candidate pages (at least 25 numeric claims, not 17) is a vendor or partner assertion without a published method on the page; the only claim with a stated comparator is Audi's '60% increase in cumulative conversion rate ... versus standard bidding tactics' on Adelaide's page, which names no method; and this review found no independent measurement with a published method of any of the twelve products' deals against a pre-curation comparator on the vendor pages and searches screened by 3 November 2026. | scope, unit, date or wording narrowed to what the source carries | CUR-F-F28 |
| V-G-43 | On Stagwell's own record the rule owner of Stagwell Curate is Stagwell (it applies 'its own intelligence, quality standards and performance signals upstream'), the supply comes through FreeWheel and Magnite, and the supply-quality models are 'powered by The Trade Desk's Sincera'; no third party is named as setting the rules. | Refute-mode verifier for G.json re-opened the cited source | Release: 'enables Stagwell to apply its own intelligence, quality standards and performance signals upstream'; 'Partners including Freewheel and Magnite will provide access to deep, direct publisher relationships'; 'supply-quality models powered by The Trade Desk's Sincera'. 'Claude' and 'OpenSincera' do not appear in the release. Digiday: 'The system relies upon Magnite and Freewheel's publisher relationships'; 'The system uses AI agents built in Claude and data pulled from The Trade Desk's OpenSincera for a "domain health check"'. | supported with qualification | On Stagwell's own record the rule owner of Stagwell Curate is Stagwell (it applies 'its own intelligence, quality standards and performance signals upstream'); the release says partners 'including Freewheel and Magnite will provide access to deep, direct publisher relationships' (future tense, like the agent connection), and the supply-quality models are 'powered by The Trade Desk's Sincera'; no third party is named as setting the rules. | scope, unit, date or wording narrowed to what the source carries | CUR-G-F02 |
| V-G-44 | The connection of Stagwell's Media Machine buying agents to Stagwell Curate is stated in the future tense ('its buying agents will connect directly to Stagwell Curate'), so no agent-executed selection on Curate is documented as production use by the cutoff; no Stagwell results release naming Curate was located (the Q3 2026 release URL returned 404). | Refute-mode verifier for G.json re-opened the cited source | Release: 'its buying agents will connect directly to Stagwell Curate' (future tense confirmed). The Q3 2026 results release exists and is public: Stagwell's 8-K exhibit 'STAGWELL INC. (NASDAQ: STGW) REPORTS RESULTS FOR THE THREE AND NINE MONTHS ENDED SEPTEMBER 30, 2026', dated New York, October 29, 2026, on SEC EDGAR. It contains no 'Curate', 'Stagwell Curate', 'Media Machine', 'FreeWheel', 'Magnite', 'curation' or 'marketplace'; the only agent wording is 'Software products, like our agentic Stagwell Machines, and the Digital Transformation segment are driving our business...'. The 10-Q for the quarter (text read through Note 14, Segment Information; the MD&A was beyond the portion returned) likewise contains none of those terms. | supported with qualification | The connection of Stagwell's Media Machine buying agents to Stagwell Curate is stated in the future tense ('its buying agents will connect directly to Stagwell Curate'), so no agent-executed selection on Curate is documented as production use by the cutoff; Stagwell's Q3 2026 results release (8-K exhibit, 29 October 2026, on SEC EDGAR) names neither Stagwell Curate nor The Media Machine, FreeWheel, Magnite, curation or a marketplace, referring only to 'agentic Stagwell Machines' software products, and the 10-Q's financial notes name none of them either. | scope, unit, date or wording narrowed to what the source carries | CUR-G-F05 |
| V-G-45 | Publicis's 6 March 2025 announcement said Lotame 'will be positioned as part of Epsilon, accessible to all Publicis clients and teams', with closing 'expected to close early in the second quarter of 2025' and price undisclosed; the release does not mention curation, deals, supply or media buying, so Publicis's use of Lotame as a curation asset is not documented on this record. | Refute-mode verifier for G.json re-opened the cited source | Dated March 6, 2025, Paris: 'Lotame will be positioned as part of Epsilon, accessible to all Publicis clients and teams'; 'is expected to close early in the second quarter of 2025.' 'curat', 'deal', 'supply', 'media buying' and 'terms' not found; 'price' only in site navigation. 'marketplace' IS found: 'Lotame is home to one of the largest end-to-end data marketplaces in the world' and 'building out our global data marketplace'. | supported with qualification | Publicis's 6 March 2025 announcement said Lotame 'will be positioned as part of Epsilon, accessible to all Publicis clients and teams', with closing 'expected to close early in the second quarter of 2025' and price undisclosed; the release describes Lotame's 'end-to-end data marketplaces' and 'global data marketplace' (a data marketplace, not a curated-supply or media marketplace) and does not mention curation, deals, supply or media buying, so Publicis's use of Lotame as a curation asset is not documented on this record. | scope, unit, date or wording narrowed to what the source carries | CUR-G-F12 |
| V-G-46 | Walmart Connect introduced Connect Select on 27 April 2026 as 'a curated marketplace within Walmart DSP' that 'At launch' 'includes inventory from VIZIO, premium CTV publishers, including Paramount' and 'Warner Bros. Discovery and supply-side platforms (SSPs), including FreeWheel, Index Exchange, Magnite, and PubMatic', so the supply is multi-publisher and multi-exchange; deals 'are available through Inventory Controls and Deal Desk'. | Refute-mode verifier for G.json re-opened the cited source | Walmart Connect article dated April 27, 2026, Ryan Mayward: heading 'Introducing Connect Select: a curated marketplace for scalable growth'; 'At launch, Connect Select includes inventory from VIZIO, premium CTV publishers, including Paramount, and Warner Bros. ...' (excerpt truncated by the fetch; the SSP list was not returned verbatim); 'Deals are available through Inventory Controls and Deal Desk.'; 'SSP partnerships enhance marketplace efficiencies through post-auction optimizations, driving improved performance, scale ...'; PubMatic quote 'Our integration with Connect Select as a preferred supply partner is delivering curated, high-performing CTV inventory'. No fee, minimum or deal ID. Magnite blog dated April 28, 2026, headline 'Magnite Helps Power Walmart DSP Marketplace Connect Select': 'Connect Select, a curated marketplace designed to simplify ...'; 'Magnite's technology is now | supported with qualification | Walmart Connect introduced Connect Select on 27 April 2026 as 'a curated marketplace' inside Walmart DSP (Magnite's headline: 'Walmart DSP Marketplace Connect Select') that 'At launch' 'includes inventory from VIZIO, premium CTV publishers, including Paramount, and Warner Bros. Discovery' and SSPs the article lists as FreeWheel, Index Exchange, Magnite and PubMatic, so the supply is multi-publisher and multi-exchange on the operator's list; deals 'are available through Inventory Controls and Deal Desk', and the article attributes 'post-auction optimizations' to the SSP partnerships. | scope, unit, date or wording narrowed to what the source carries | CUR-G-F13 |
| V-G-47 | Roku Curate (page dated 27 April 2026) 'combines Roku's proprietary insights and inventory' with 'in-market targeting data from partners like Best Buy Ads, Criteo, Fandango, Fetch, Instacart and Kroger'; partners supply data, not inventory, and 'Roku can then activate directly or send the package to the advertiser's preferred demand-side platform', so it is single-publisher and falls outside the multi-publisher subset. | Refute-mode verifier for G.json re-opened the cited source | Dated Apr 27, 2026, 'What is Roku Curate? A quick guide to Roku's new data-and-inventory packages': 'Roku Curate is a premium advertising package that combines Roku's proprietary insights and inventory with high-value partner data'; partners 'Best Buy Ads, Criteo, Fandango, Fetch, Instacart and Kroger' supply data only; 'Advertisers work with Roku directly to set up their desired Roku Curate media packages.'; 'Roku can then activate directly or send the package to the advertiser's preferred demand-side platform (DSP) for activation.'; 'With a standard Deal ID or a PMP, the publisher sends inventory, while the targeting and decisioning happen at the DSP level.'; 'With Roku Curate, Roku packages the data and media supply for easier activation and lower fees.' No DSP, deal ID or fee amount named. | supported with qualification | Roku Curate (page dated 27 April 2026) 'combines Roku's proprietary insights and inventory' with 'in-market targeting data from partners like Best Buy Ads, Criteo, Fandango, Fetch, Instacart and Kroger'; partners supply data, not inventory; Roku contrasts it with a standard deal ID or PMP, where 'the publisher sends inventory, while the targeting and decisioning happen at the DSP level', saying that with Roku Curate 'Roku packages the data and media supply'; 'Roku can then activate directly or send the package to the advertiser's preferred demand-side platform', so it is single-publisher and falls outside the multi-publisher subset. | scope, unit, date or wording narrowed to what the source carries | CUR-G-F17 |
| V-G-48 | The Trade Desk describes OpenPath (page of 24 February 2026) as 'providing access to high-quality inventory via direct connections between advertisers and a publisher's ad server' and says it 'does not function as an intermediary', with 'hundreds of publishers'; the page states no fee and attributes inventory inclusion to publishers 'making their inventory available', with a '3x higher' win-rate claim without sample or period. | Refute-mode verifier for G.json re-opened the cited source | Dated February 24, 2026: 'providing access to high-quality inventory via direct connections between advertisers and a publisher's ad server'; 'OpenPath does not function as an intermediary, but rather as the most direct route an ad can take within our platform'; 'among the hundreds of publishers making their inventory available through a direct connection' (sentence names New York Post, FUNKE, Valnet, The Arena Group and Vizio); 'fee', '4.5%' and 'percent' do not appear. Footnote 1 verbatim: 'The Trade Desk, platform data, October 2024.' 'Analysis spanning dozens of advertisers in various verticals.' 'Advertiser campaigns saw win rates 3x higher than campaigns not using OpenPath.' Footnotes 2 and 3 cite 'platform data, 2024' for DraftKings and Ovative Group, whose figures render as blanks ('x higher win rate', '% reduction in CPMs'). [297] | supported with qualification | The Trade Desk describes OpenPath (page of 24 February 2026) as 'providing access to high-quality inventory via direct connections between advertisers and a publisher's ad server' and says it 'does not function as an intermediary', with 'hundreds of publishers' (naming New York Post, FUNKE, Valnet, The Arena Group and Vizio); the page states no fee and attributes inventory inclusion to publishers 'making their inventory available', with a '3x higher' win-rate claim footnoted to The Trade Desk platform data for October 2024 across 'dozens of advertisers in various verticals', without a stated method or controls. | scope, unit, date or wording narrowed to what the source carries | CUR-G-F18 |
| V-G-49 | Amazon Ads' undated supply-quality article says it 'recently' launched the Certified Supply Exchange programme to strengthen relationships with major ad exchanges (none named), offers Amazon Publisher Direct access to 'thousands' of invitation-only publishers, includes its quality measures 'at no additional cost', and charges 'a single 1% technology fee' for guaranteed deals outside Amazon. | Refute-mode verifier for G.json re-opened the cited source | Served in Italian, undated ('In che modo Amazon Ads garantisce la qualità dell'offerta su Internet'): 'Di recente abbiamo lanciato il programma Certified Supply Exchange per rafforzare le nostre relazioni con i principali exchange pubblicitari' (no exchange named); 'relazioni dirette e solo su invito con migliaia di publisher autorevoli' via 'Amazon Publisher Direct (APD)'; 'Includiamo tutte queste misure per la qualità senza costi aggiuntivi per i nostri clienti.'; 'Quando gli inserzionisti desiderano stabilire accordi garantiti con specifici publisher all'esterno di Amazon, viene loro addebitata una sola commissione per la tecnologia dell'1%.' The page does not state the base on which the 1% is calculated. 2024 Forrester recognition confirmed. | supported with qualification | Amazon Ads' undated supply-quality article (served in Italian) says it 'recently' launched the Certified Supply Exchange programme to strengthen relationships with major ad exchanges (none named), offers Amazon Publisher Direct access to 'thousands' of invitation-only publishers, includes its quality measures 'at no additional cost', and charges 'a single 1% technology fee' when advertisers set up guaranteed deals with specific publishers outside Amazon; the page states neither the base of that percentage nor the certified exchanges. | scope, unit, date or wording narrowed to what the source carries | CUR-G-F22 |
| V-G-50 | Viant's Direct Access page describes 'Viant's innovative supply path optimization program' that brings 'buyers and sellers closer through direct inventory and first-party data integration', lists 19 CTV and audio publisher logos, and sells IRIS-enabled inventory as 'Independence from SSP-curated deal ID limitations'; no fee, take rate or selection rule is stated. | Refute-mode verifier for G.json re-opened the cited source | Undated page, footer '©2026': 'Viant's innovative supply path optimization program'; 'bringing buyers and sellers closer through direct inventory and first-party data integration'; 'Independence from SSP-curated deal ID limitations'; 'IRIS-enabled publisher inventory is available across select publishers only'; 19 logos under 'Direct Access CTV and Audio Publishers' with no alt text; Disney and NBCUniversal are named in the FAQ text ('integrations with partners like Disney and NBCUniversal'); cost wording: 'Pricing efficiency to allow for more working media', 'Reduce ad tech fees and non-working media spend', 'reducing fees, minimizing fraud risk, and improving campaign performance'. No fee amount, take rate or selection rule. | supported with qualification | Viant's Direct Access page describes 'Viant's innovative supply path optimization program' that brings 'buyers and sellers closer through direct inventory and first-party data integration', shows 19 CTV and audio publisher logos and names Disney and NBCUniversal in its FAQ, and sells IRIS-enabled inventory as 'Independence from SSP-curated deal ID limitations'; it promises to 'Reduce ad tech fees and non-working media spend' but states no fee amount, take rate or selection rule. | scope, unit, date or wording narrowed to what the source carries | CUR-G-F24 |
| V-H-51 | ANA benchmark median transaction-cost series: DSP platform 8.0% (2023), 8.6% (Apr-Sep 2024), 7.9% (Nov 2024-Mar 2025); DSP data 6.0, 4.6, 1.0; DSP additional 2.0, 3.6, 4.5; SSP platform 13.0, 13.0, 13.7, with the SSP top quartile at 16.9% in Q1 2025. | Refute-mode verifier for H.json re-opened the cited source | Q1 2025 report p. 8 table (columns 2023, Q4 2024, Q1 2025; 'Median' rows): DSP Platform Costs 8.0% 8.6% 7.9%; DSP Data Costs 6.0% 4.6% 1.0%; DSP Additional Costs 2.0% 3.6% 4.5%; SSP Platform Costs 13.0% 13.0% 13.7%; Active Marketers 21 21 23. p. 13: 'the median increased by just 0.7 percent to 13.7 percent ... The quartiles are also consistent across the period, with the top of the range reaching 16.9 percent in Q1 2025'. The 2023 column values (8.0, 6.0, 2.0, 13.0) coincide exactly with the 2023 Complete Report's sequential waterfall averages ($80, $60, $20, $130 per $1,000, p. 101); the 2023 report's appendix (p. 123) says DSP and SSP costs 'were calculated as averages across participating advertisers' and no median transaction-cost table was found in it. | supported with qualification | The ANA benchmark's Q1 2025 report tabulates its transaction-cost series as medians across contributing advertisers: DSP platform costs 8.0% (2023 column), 8.6% (April to September 2024) and 7.9% (November 2024 to March 2025); DSP data costs 6.0%, 4.6% and 1.0%; DSP additional costs 2.0%, 3.6% and 4.5%; SSP platform costs 13.0%, 13.0% and 13.7%, with the top of the SSP quartile range at 16.9% in Q1 2025; the 2023 column's values coincide exactly with the 2023 Complete Report's sequential waterfall averages ($80, $60, $20 and $130 per $1,000), and that report describes its cost figures as averages and publishes no medians, so the 2023 'medians' are to be labelled as the Q1 2025 report's re-tabulation of the 2023 study rather than as medians the 2023 study itself reported. | scope, unit, date or wording narrowed to what the source carries | CUR-H-F04 |
| V-H-52 | ANA Q2 2025 release: PMP CPMs $7.15 against $4.41 OMP (medians per the report tables), TrueCPM Index 36.5% from 37.8%, PMP share 64.5% to 87.8%, median SSP count 19 to 17; no fee figure. | Refute-mode verifier for H.json re-opened the cited source | Release of 14 August 2025: 'TrueCPM Index slipped from 37.8% to 36.5%'; 'Median SSP count fell from 19 to 17'; 'PMP share of spend surged from 64.5% to 87.8%, with PMP CPMs averaging $7.15 vs. $4.41 for OMP.' No DSP, SSP or curation fee figure appears. The release says 'averaging'; the median reading comes from the brief's fact-check C1-04, not from a source the stream opened. The Q1 2026 report's p. 5 retrospective table places $7.15 / $4.41 under Q2 2025 in a row whose Q1 2025 column holds $5.83 / $2.75, which are the 2023 study's spend-weighted totals in the Q1 2025 report (p. 8), so that table cannot settle median against total. | supported with qualification | The ANA's Q2 2025 release (14 August 2025) reports 'PMP CPMs averaging $7.15 vs. $4.41 for OMP', a TrueCPM Index of 36.5% (from 37.8%), PMP share of spend 'surged from 64.5% to 87.8%' and a median SSP count down from 19 to 17, and carries no DSP, SSP or curation fee figure; that $7.15 and $4.41 are medians across the 21 contributing advertisers (aggregate $7.43 against $3.64) rests on the brief's fact-check of the Q2 2025 report, which this stream did not open and which should be cited directly. | scope, unit, date or wording narrowed to what the source carries | CUR-H-F08 |
| V-H-53 | ISBA/PwC 2020: publishers 51%, unknown delta 15% (about one-third of supply-chain costs), 15 advertisers / 8 agencies / 5 DSPs / 6 SSPs / 12 publishers, about GBP 0.1 billion, 290 chains and 31 million impressions matched; CMA restates as 49% absorbed by intermediaries. | Refute-mode verifier for H.json re-opened the cited source | ISBA article (06 May 2020): 'publishers receive 51% of advertiser spend on average'; '15% of advertiser spend - an "unknown delta", representing around one-third of supply chain costs - could not be attributed'; '15 advertisers, eight agencies, five Demand Side Platforms (DSPs), six Supply Side Platforms (SSPs) and 12 publishers ... approximately £0.1 billion'; 'end-to-end match 290 chains (31 million impressions)'. CMA Appendix R para. 42 (p. R16): '51% of initial advertiser expenditure, with 49% of the initial spend absorbed by various intermediaries ... Of this 49%, 15% was classified as an unknown delta'. The ISBA/PwC second-study summary (18 January 2023, p. 7, not opened by the stream): 'our 2022 study excludes agency fees, verification tools and ad serving; so to allow direct comparisons, our 2020 results must be restated as follows: unknown delta 15% restated to 17%; publisher ne | supported with qualification | The ISBA/PwC Programmatic Supply Chain Transparency Study (6 May 2020; 15 advertisers, 8 agencies, 5 DSPs, 6 SSPs and 12 publishers; about GBP 0.1 billion of UK programmatic spend; 290 supply chains and 31 million impressions matched end to end) found publishers received 51% of advertiser spend on average and that 15% of advertiser spend was an 'unknown delta', about one-third of supply-chain costs, which the CMA restates as 49% absorbed by intermediaries; ISBA and PwC's second study (18 January 2023) restated those 2020 figures to 57% publisher net revenue and a 17% unknown delta so as to exclude agency fees, verification tools and ad serving, the second study's scope, so the 51% and the CMA's 65% are not on one basis. | scope, unit, date or wording narrowed to what the source carries | CUR-H-F09 |
| V-H-54 | ISBA/PwC second study (18 January 2023; 2022 fieldwork, nine months): match rate 58% (12% in 2020), unknown delta 3% (15%), publisher share 'risen by 8%' without saying points or relative, sample sizes unstated, not representative. | Refute-mode verifier for H.json re-opened the cited source | Knowledge page (18 Jan 2023): 'halved the time required to conduct the study to nine months (vs 18 months for the 2020 study)'; 'improved the ad impression match rate to 58% (vs 12% in 2020) and the unattributable ad spend (AKA the unknown delta) was reduced to 3% (vs 15% in 2020)'; 'The proportion of advertiser spend reaching publishers has risen by 8%.'; 'These promising results should not be taken as representative of the broader programmatic ecosystem.' The linked summary PDF (pp. 7-8), which the stream did not open: 2022 study of 11 advertisers, 7 agencies, 6 DSPs and 6 SSPs, 10 publishers; 1.3bn impressions analysed for 1 September to 31 October 2022; 61m matchable impressions, 58% matched; unknown delta 3% ('and <1% for private marketplaces'); publisher revenue 65% of advertiser spend 'after applying the supply chain costs analysed in this study (including the unknown delta)' agai | supported with qualification | ISBA and PwC's second study (summary published 18 January 2023; 11 advertisers, 7 agencies, 6 DSPs, 6 SSPs and 10 publishers; 1.3 billion impressions analysed for 1 September to 31 October 2022; a nine-month study) reported an impression match rate of 58% (12% in 2020), an unknown delta of 3% (15% in 2020, restated to 17% on the second study's basis; under 1% for private marketplace deals) and a publisher share of advertiser spend of 65% against a restated 57% for 2020, so the knowledge page's 'risen by 8%' is 8 percentage points on a basis that excludes agency fees, verification tools and ad serving; the summary says the results are not representative of the broader programmatic ecosystem. | scope, unit, date or wording narrowed to what the source carries | CUR-H-F10 |
| V-H-55 | Jounce April 2025 Supply Path Benchmarking (client-only; March 2025, 1.5 billion bid requests, 44 exchanges) via Playwire and AdExchanger: 50% zero fee, 34% margin-based (median 14%; 36% in 5-10% and 9% below 5% per Playwire; 'nearly 40%' per AdExchanger), 10% fixed CPM (median $0.48), 7% static price; median effective margin 30%, median effective CPM $0.51 per Playwire; Jounce's public site carries no fee statement. | Refute-mode verifier for H.json re-opened the cited source | Playwire (April 10, 2025): 'Zero Fee Curation (50% of bid requests): Half of all curated deal IDs carry no additional fee beyond the exchange's standard open auction take rate'; 'Margin-Based Curation (34% of bid requests) ... median fee being a 14% take rate'; 'CPM-Based Curation (10% of bid requests): Fixed CPM fees (median $0.48)'; 'Static Price Curation (7% of bid requests)'; 'Approximately 36% of margin-based fees fall in the 5-10% range, and another 9% below 5%'; 'The median effective margin for curated deals is 30%, with the median effective CPM at $0.51'. AdExchanger (May 1, 2025): 'In March, programmatic consultancy Jounce Media analyzed 1.5 billion bid requests sent from 44 ad exchanges'; 'Roughly half of all curated deal IDs carry no curation-related fees'; 'Around 34% of curated deals carry margin-based fees'; 'the median margin-based fee is 14%, and nearly 40% of margin-base | supported with qualification | Jounce Media's April 2025 Supply Path Benchmarking report (client-only; 1.5 billion bid requests from 44 exchanges, analysed in March 2025 per AdExchanger) found, as reported by Playwire (10 April 2025) and AdExchanger (1 May 2025), that 50% of curated deal IDs (Playwire also labels the shares 'of bid requests') carried no fee beyond the exchange's standard open-auction take, 34% a margin-based fee with a median of 14% (Playwire: about 36% of those between 5% and 10% and 9% below 5%; AdExchanger: 'nearly 40%' between 5% and 10%), 10% a fixed CPM fee with a median of $0.48 and 7% a static price, with a median effective margin of 30% and a median effective CPM of $0.51 per Playwire; Jounce's public State of the Open Internet 2025 (pp. 37-38) treats exchange take rates only as mark-ups embedded in declared floors (illustrative 10%, 15% and 20% on a $0.60 floor) and refers to a client-portal | scope, unit, date or wording narrowed to what the source carries | CUR-H-F11 |
| V-H-56 | Magnite FY2025 10-K: take rate defined as 'our fee as a percentage of advertising spend', varies by publisher and transaction type, lower for reserve auctions, higher for managed service, lower take rates negotiated with large sellers, fee concessions more prevalent among CTV publishers; revenue $713,953 / $668,170 thousand; 10% / 14% / 18% reported gross; no numeric take rate. | Refute-mode verifier for H.json re-opened the cited source | Item 1A: 'Our take rate varies by publisher and transaction type.' 'For instance, our take rate tends to be lower for reserve auctions compared to auctions with multiple bidders, and tends to be lower on CTV transactions compared to other channels.' 'Additionally, take rates tend to be higher for our managed service business where we are responsible [for delivering a campaign at a fixed price]'. 'We may also negotiate lower take rates with large sellers to win additional business or share of inventory'. 'These fee concessions may be more prevalent among CTV publishers where inventory is scarce'. 'could drive a decrease in our overall take rate (our fee as a percentage of advertising spend)'. Item 7: 'our revenue reported on a gross basis was 10%, 14%, and 18% of total revenue for the respective periods' (2025, 2024, 2023); revenue $713,953 and $668,170 thousand; 'Revenue increased $45.8 | supported with qualification | Magnite's FY2025 Form 10-K defines its take rate as 'our fee as a percentage of advertising spend', states that it 'varies by publisher and transaction type', 'tends to be lower for reserve auctions compared to auctions with multiple bidders, and tends to be lower on CTV transactions compared to other channels', and is higher for its managed service business, that Magnite 'may also negotiate lower take rates with large sellers', and that fee concessions 'may be more prevalent among CTV publishers where inventory is scarce'; it reports revenue of $713,953 thousand (2025) and $668,170 thousand (2024), with 10%, 14% and 18% of revenue reported on a gross basis in 2025, 2024 and 2023, and discloses no numeric take rate, no total ad spend and no fee for the curation inside ClearLine in the portions read. | scope, unit, date or wording narrowed to what the source carries | CUR-H-F13 |
| V-H-57 | DOJ liability opinion (17 April 2025): 'take rate' usage; '$20 to $40 of every $100'; AdX 'overall take rate of around 20%', 15% or lower / 25% or higher per impression; rivals 'often charged closer to 10%'; exchanges 'between 10% and 20% of revenue from the winning bid'; DFP fees equate to a take rate of 1.3% to 1.7%; Index Exchange 'nominal, at best' effect including a zero take rate. | Refute-mode verifier for H.json re-opened the cited source | p. 18: 'refer to an ad tech tool's percentage-based fees as its "take rate."'; 'Typically, between $20 to $40 of every $100 spent on digital advertising goes to the ad tech companies ... See PTX794 at -738; Tr. ... (LaSala (Google))'. p. 35: 'Although AdX maintained an overall take rate of around 20%, it would lower its take rate below 20% on competitive impressions'. p. 36: 'a 15% or lower take rate for competitive impressions, and a 25% or higher take rate for less competitive impressions, all the while maintaining an average take rate of around 20%'. p. 52: 'often taking between 10% and 20% of revenue from the winning bid'. p. 77: 'AdX's 20% take rate was higher than that of rival ad exchanges like IndexExchange, Magnite, and Xandr, which often charged closer to 10%'. p. 78: 'at one point setting its take rate to zero, but found only a "nominal, at best, effect on win rate."' p. 45: t | supported with qualification | The liability opinion of 17 April 2025 in United States v. Google records that industry participants call an ad-tech tool's percentage fees its 'take rate', that 'between $20 to $40 of every $100 spent on digital advertising goes to the ad tech companies' (citing an exhibit and Google testimony), that AdX maintained 'an overall take rate of around 20%' while varying it between 15% or lower and 25% or higher per impression under sell-side dynamic revenue share, that rival exchanges 'often charged closer to 10%' and exchanges generally take 'between 10% and 20% of revenue from the winning bid', that Index Exchange found 'nominal, at best' win-rate effects from experiments including setting its take rate to zero, and, in a citation parenthetical summarising defence-expert testimony and exhibits, that DFP fees equate to a take rate of '1.7 percent to 1.3 percent' against around 20% 'for an a | scope, unit, date or wording narrowed to what the source carries | CUR-H-F16 |
| V-H-58 | In the sources screened, no exchange or curator publishes a curation rate card or its exchange take on curated deals; the public record is fee types and stacking rules, zero-fee claims, one segment price, a client-only fee-type distribution via reporting, 10-K descriptions without rates, a derived TTD ratio, Google's 2020 15% DV360 rate and Amazon's fee basis. | Refute-mode verifier for H.json re-opened the cited source | The absence claim holds across every cited source as re-opened. Two public items the inventory omits: Jounce's public State of the Open Internet 2025 (pp. 37-38) gives illustrative exchange take rates (10%, 15%, 20%) as floor mark-ups and documents one exchange marking a $0.60 floor to $0.74; and the Deal Sync API 1.0 'auxdata' attribute flags the presence of a fee-bearing non-publisher data layer (0 undisclosed, 1-2 yes, 3-4 no). | supported with qualification | In the sources screened by 3 November 2026, no exchange or curator publishes a rate card for curation fees or its exchange take on curated deals: the public record consists of fee types and stacking rules (Google, Microsoft, the Deal Sync API's curfeetype and its auxdata flag for a fee-bearing data layer), zero-fee claims for one party's charge (Index data and platform fees, TripleLift curation fees), one segment price ($0.15 CPM, Adloox via Qwarry on PubMatic), a fee-type distribution from a client-only sample (Jounce, via reporting) alongside Jounce's public illustration of exchange take rates as floor mark-ups (State of the Open Internet 2025), corporate fee descriptions without rates in the FY2025 10-Ks of Magnite and PubMatic, a derived blended ratio for The Trade Desk, a 2020 Google statement of a 15% DV360 standard rate, and Amazon's statement of the fee basis without the rate. | scope, unit, date or wording narrowed to what the source carries | CUR-H-F30 |
| V-H-59 | Only Microsoft's and Google's stacking orders are documented; Jounce (via AdExchanger) describes the SSP paying the curator from a floor that embeds the fee; no source gives all four series for one named deal. | Refute-mode verifier for H.json re-opened the cited source | Microsoft and Google passages verbatim as in F19 and F20. AdExchanger: 'Curators and SSPs factor their curation fees into the floor price passed to DSPs'; 'an SSP accrues and pays the curator's fee, rather than the DSP'. Jounce's own public State of the Open Internet 2025 (pp. 37-38): 'After accounting for exchange fees ... the bid requests issued to DSPs should declare a floor that is modestly higher than $0.60'; 'Exchange A simply marks up the publisher's floor price to account for its fee'. No source gives all four series for one named curated deal. | supported with qualification | Within a single transaction perimeter the only documented stacking orders found are Microsoft's (buyer fees removed; curator margin and curator tech fee or MASC deducted; seller auction service charge deducted; remainder to the seller) and Google's (curator fee 'taken out of the bid price before it enters the auction', Google's own share unstated), while Jounce describes in its public State of the Open Internet 2025 how an exchange's fee is embedded in the declared floor and, via AdExchanger, the SSP accruing and paying the curator's fee out of a floor that already embeds it; no public source gives all four series (curation fee, data fee, SSP take, DSP take) with payer, collector, base and period for one named curated deal. | scope, unit, date or wording narrowed to what the source carries | CUR-H-F31 |
| V-H-60 | Row 34: 51 (publishers), 15 (unknown delta), % of advertiser spend, 2019-2020 fieldwork. Row 35: unknown delta 3; publisher share 'risen by 8%' (ambiguous); sample not stated; 2022 fieldwork. | Refute-mode verifier for H.json re-opened the cited source | Summary PDF p. 8 highlights: 2020 study 1.3bn impressions, 18 months, 31m matched, 12% match rate, 15% unknown delta '(17% restated)', 51% publisher revenue '(57% restated)'; 2022 study 1.3bn, 9 months, 61m matchable, 58%, 3%, 65%; period 1 September to 31 October 2022; 2020 data 'collected for Q1 2020'; 2022 sample 11 advertisers, 7 agencies, 6 DSPs and 6 SSPs, 10 publishers. Publisher revenue is 'Proportion of advertiser spend that reaches publishers after applying the supply chain costs analysed in this study (including the unknown delta)'. | supported with qualification | Row 34 should read: publishers 51 and unknown delta 15, percent of advertiser spend as published in 2020 (restated by the second study to 57 and 17 on a basis excluding agency fees, verification tools and ad serving), 15 advertisers, about GBP 0.1 billion, data for Q1 2020, 290 chains and 31 million impressions matched; row 35 should read: publishers 65 and unknown delta 3 (under 1 for private marketplace deals), percent of advertiser spend on the second study's basis, against a restated 57 and 17 for 2020, so the publisher share rose by 8 percentage points; 11 advertisers, 7 agencies, 6 DSPs, 6 SSPs and 10 publishers; 1.3 billion impressions for 1 September to 31 October 2022; published 18 January 2023. | scope, unit, date or wording narrowed to what the source carries | CUR-H data table fee_series, rows 34-35 (ISBA/PwC 2020 and 2023 publisher share) |
| V-H-61 | Rows 12/21/30 give 2023-study values 6.0 (data), 13.0 (SSP), 8.0 and 2.0 (DSP platform, additional) as medians across 21 contributing advertisers; the ana_cost_waterfall_series table restates every sequential and index value with its period. | Refute-mode verifier for H.json re-opened the cited source | Every value in ana_cost_waterfall_series was found in the cited report: 29 / 35 / 36 (2023 report pp. 20, 101); 29.8, 26.1, 73.9, 43.9, 41.0, 37.8 (Q1 2025 pp. 4-5, 11); 23.7, 25.9, 26.3, 40.1, 30.8, 36.3, 43.3, 44.6, 33.9 (Q1 2026 pp. 6-8); 36.5 (Q2 2025 release); counts 21/21/23/66 and spend 123/235/242/160 (Q1 2025 p. 8; Q1 2026 p. 5). The 2023 'median' column in the Q1 2025 table coincides exactly with the 2023 report's waterfall averages and the 2023 report publishes no transaction-cost medians. | supported with qualification | The 2023 column values in fee_series rows 12, 21 and 30 (6.0, 13.0, 8.0 and 2.0 percent) are the Q1 2025 report's tabulation of the 2023 study in its 'Median' rows and coincide exactly with the 2023 Complete Report's sequential waterfall averages, which that report describes as averages across participating advertisers; they should be labelled 'as tabulated in the Q1 2025 report' rather than as 2023 medians, and the ana_cost_waterfall_series rows for Q4 2025 should note that they are cited from the Q1 2026 report's restatement rather than from the Q4 2025 report. | scope, unit, date or wording narrowed to what the source carries | CUR-H data table fee_series, rows 12, 21 and 30 (ANA 2023 'median' column) and table ana_cost_waterfall_series |
| V-H-62 | ANA 2023 PMP tiers and OMP CPMs; benchmark total and median PMP/OMP CPMs for the 2023, Apr-Sep 2024 and Nov 2024-Mar 2025 columns; Q2 2025 7.15 vs 4.41 as medians; Q1 2026 cohorts 7.40 / 5.45; Q1 2026 market median CPM 4.41 from 5.55. | Refute-mode verifier for H.json re-opened the cited source | Rows 0-3 match the 2023 report p. 70 table. Rows 4-7 match the Q1 2025 report p. 8: total OMP CPM $2.75 / $3.35 / $3.03, total PMP CPM $5.83 / $7.46 / $10.43, median OMP $2.23 / $3.26 / $3.28, median PMP $6.68 / $5.67 / $9.37. Row 9 matches the Q1 2026 p. 4 table. Row 10: the Q1 2026 report's text says 'overall CPMs declined to a median of $4.41 from $5.55' while its p. 5 table gives 'CPM Total' $5.55 (Q4 2025) and $4.42 (Q1 2026). Row 8: the release says 'averaging'; the median label rests on the brief's fact-check. | supported with qualification | Row 8 should carry the Q2 2025 benchmark report, not only the release, as the source for 'medians' (the release says 'averaging'), and row 10 should note that the Q1 2026 report's text calls $4.41 a median while its retrospective table lists a 'CPM Total' of $4.42 for Q1 2026 and $5.55 for Q4 2025, so the two figures are not reconciled within the source. | scope, unit, date or wording narrowed to what the source carries | CUR-H data table pmp_vs_omp_cpm (rows 0-10) |
| V-I-63 | Six sell-side companies named in the brief sued Google after the liability opinion: OpenX (E.D. Va. No. 1:25-cv-1282, complaint filed 4 August 2025), PubMatic (E.D. Va., 8 September 2025), Magnite (E.D. Va., 16 September 2025), Equativ and Sharethrough (E.D. Va., 13 October 2025, E.D. Va. No. 1:25-cv-01755), Index Exchange (S.D.N.Y. No. 1:25-cv-10477, filed 17 December 2025) and Teads (S.D.N.Y., 3 August 2026); each of the opened pleadings or releases cites the April 2025 opinion, which is a liability finding, not a judgment. | Refute-mode verifier for I.json re-opened the cited source | OpenX complaint: header 'Case 1:25-cv-01282 Document 1 Filed 08/04/25 Page 1 of 87', 'JURY TRIAL DEMANDED', paragraph 176 on p. 73 ('On April 17, 2025, the Court issued a Memorandum Opinion holding that: ...'). PubMatic 8-K exhibit: 'REDWOOD CITY, Calif., Sep. 8, 2025', 'today filed a lawsuit in the U.S. District Court for the Eastern District of Virginia', 'The lawsuit follows the ruling from the U.S. District Court in April 2025'; no case number. Magnite exhibit: 'NEW YORK - September 16, 2025', 'had filed a lawsuit in the U.S. District Court for the Eastern District of Virginia', 'in light of the April 2025 U.S. District Court's ruling'; no case number. Teads exhibit: 'NEW YORK – August 3, 2026', 'filed a lawsuit in the U.S. District Court for the Southern District of New York against Google LLC and Alphabet Inc.'; it refers to 'the ruling by the U.S. District Court for the Eastern Di | supported with qualification | Six sell-side companies named in the brief sued Google after the liability opinion: OpenX (E.D. Va. No. 1:25-cv-01282, filed 4 August 2025), PubMatic (E.D. Va. No. 1:25-cv-01482, 8 September 2025), Magnite (E.D. Va. No. 1:25-cv-01541, 16 September 2025), Equativ and Sharethrough (E.D. Va. No. 1:25-cv-01755, 13 October 2025), Index Exchange (first in E.D. Va. No. 1:25-cv-02003 on 10 November 2025, voluntarily dismissed by notice of 17 December 2025, then S.D.N.Y. No. 1:25-cv-10477, filed 17 December 2025) and Teads (S.D.N.Y. No. 1:26-cv-06591, 3 August 2026); the OpenX complaint and the PubMatic and Magnite releases cite the April 2025 opinion by date, and the Teads release cites the E.D. Va. ruling without dating it; the opinion is a liability finding, not a judgment. | scope, unit, date or wording narrowed to what the source carries | CUR-I-F14 |
| V-I-64 | The CMA's Competition Act investigation into Google's ad-tech conduct (case 51145 per the brief; opened 25 May 2022; statement of objections 6 September 2024) remained 'Open' with no decision, its latest timetable phase being 'Considering next steps, whilst monitoring international developments' estimated for December 2025 to July 2026, on a page last updated 19 May 2026; the CMA's strategic market status designations of Google in general search and search advertising (10 October 2025) and in its mobile platform (22 October 2025) and every conduct requirement or commitment under them through 23 September 2026 concern search or mobile platforms and mention no ad-tech, ad exchange, display or deal mechanics. | Refute-mode verifier for I.json re-opened the cited source | Ad-tech page: 'Case state: Open'; Published 26 May 2022; Last updated 19 May 2026; opened 25 May 2022; timetable's latest phase 'December 2025 to July 2026 (estimate): Considering next steps, whilst monitoring international developments'; updates log '19 May 2026: Case timetable updated', '6 September 2024: Statement of objections issued to Google', '10 March 2023: Decision to combine this investigation with CMA's investigation into Google's conduct in relation to header bidding services'; '(Case 51134)' appears once; 51145 does not appear. Search SMS page: last updated 23 September 2026; log entries 10 October 2025 'Final decision published', 3 June 2026 'Publisher conduct requirement published', 17 June 2026 'Fair ranking and data portability conduct requirements published', 23 September 2026 'Further consultation on user choice conduct requirement published'; no mention of ad tech, ad | supported with qualification | The CMA's Competition Act investigation into Google's ad-tech conduct (case 51145 per the brief; opened 25 May 2022; statement of objections 6 September 2024) remained 'Open' with no decision, its latest timetable phase being 'Considering next steps, whilst monitoring international developments' estimated for December 2025 to July 2026, on a page last updated 19 May 2026; the CMA's strategic market status designations of Google in general search and search advertising (10 October 2025) and in its mobile platform (22 October 2025) concern search or mobile platforms, and the two case pages' update logs through 23 September 2026, which list the conduct requirements and commitments made under them, mention no ad-tech, ad exchange, display or deal mechanics (the conduct-requirement and commitment documents themselves were not opened). | scope, unit, date or wording narrowed to what the source carries | CUR-I-F20 |
| V-I-65 | By the cutoff, nineteen US states had signed comprehensive consumer privacy laws and all were in force, the latest effective dates being 1 January 2026 (Indiana, Kentucky, Rhode Island), per IAPP's chart of 24 November 2025; these laws, not a federal statute, are the US rules that govern the sale and processing of personal data used in in-flight enrichment. | Refute-mode verifier for I.json re-opened the cited source | The chart (text-extracted; dated '24 Nov. 2025') lists nineteen states under 'LAWS SIGNED (TO DATE)': California, Colorado, Connecticut, Delaware, Indiana, Iowa, Kentucky, Maryland, Minnesota, Montana, Nebraska, New Hampshire, New Jersey, Oregon, Rhode Island, Tennessee, Texas, Utah, Virginia, with the bill numbers and effective dates the stream's rows carry (the latest 1 Jan. 2026 for Indiana SB 5, Kentucky HB 15 and Rhode Island H 7787). That part is accurate as of the chart's date. But the chart predates the 2026 sessions: IAPP's news item of 17 June 2026 states 'Vermont became the 23rd state to enact a comprehensive state privacy law' when 'Gov. Phil Scott, R-Vt., signed Senate Bill 71, the Vermont Data Privacy and Online Surveillance Act, into law 16 June', effective '1 Jan. 2028', and that Vermont 'becomes the fourth state to pass a comprehensive privacy law this year, joining Alab | contradicted | By the cutoff, twenty-three US states had enacted comprehensive consumer privacy laws: the nineteen on IAPP's chart of 24 November 2025, all in force by 1 January 2026 (the latest effective dates being Indiana, Kentucky and Rhode Island), plus four enacted in 2026 (Oklahoma, Alabama, Louisiana and Vermont, per IAPP's report of 17 June 2026) that take effect in 2027 and 2028; so nineteen laws were in force at the cutoff. These state laws, not a federal statute, are the US rules that govern the sale and processing of personal data used in in-flight enrichment. | the source does not carry the claim as first written | CUR-I-F21 |
| V-I-66 | By IAB Europe's own account of 14 May 2025, the Belgian Market Court annulled the APD's February 2022 decision, holding IAB Europe a joint controller only for the creation and use of TC Strings and not for participants' own processing, with execution of the 2023 action plan remaining suspended; IAB Europe launched TCF v2.3 in April 2025 with an adoption deadline of 28 February 2026. | Refute-mode verifier for I.json re-opened the cited source | IAB Europe's statement (14 May 2025) says 'the Market Court has annulled the APD February 2022 decision' and, in the same statement, that the court 'upheld only part of the APD decision, namely that IAB Europe is a joint controller together with TCF participants solely in relation to the creation and use of TC Strings by publishers and vendors', with the 2023 action plan's execution remaining suspended; it does not mention the fine. Lewis Silkin (27 May 2025; dates the ruling 14/15 May 2025) says 'the Belgian Market Court upheld the Belgian Data Protection Authority's (DPA) decision to fine IAB Europe €250,000' and 'the Market Court followed the CJEU's reasoning, upholding the DPA's fine of €250,000', with IAB Europe a joint controller for TC Strings but not for subsequent advertising processing. The TCF page: 'In April 2025, TCF v2.3 was launched to resolve the legitimate interest ambig | supported with qualification | By IAB Europe's own account of 14 May 2025, the Belgian Market Court annulled the APD's February 2022 decision except the part finding IAB Europe a joint controller, which it upheld solely for the creation and use of TC Strings and not for participants' own processing (IAB Europe's words: 'upheld only part of the APD decision'), with execution of the 2023 action plan remaining suspended; a law-firm summary reports the EUR 250,000 fine as upheld; IAB Europe launched TCF v2.3 in April 2025 with an adoption deadline of 28 February 2026. | scope, unit, date or wording narrowed to what the source carries | CUR-I-F25 |
| V-I-67 | Inference: the US competition record addresses the auction mechanics of one ad server and one exchange (DFP and AdX) for open-web display 'Indirect Demand' (open and private auctions); none of the liability opinion, the remedies opinion or the proposed judgment mentions curation, curators or third-party deal IDs, so the record documents observed, court-found first-look, last-look and floor advantages for Google's exchange and a proposed restoration of publishers' per-source floor control, and establishes nothing about the default priority, floors or exclusivity of curated deals on any exchange. | Refute-mode verifier for I.json re-opened the cited source | A case-insensitive search of the three extracted texts: 'curator', 'curation', 'curated' and 'deal ID' have no hits in any of them; 'curat' as a substring hits 'accurate'/'accurately' (Dkt 1410 twice, Dkt 1858 once) and, once, the word 'curate' in Dkt 1858 p. 85, in quoted remedies-trial testimony: advertisers such as Omnicom 'expect [their demand-side platforms] to curate or prioritize one ad exchange over another' because of 'differences in quality, security, privacy, and fraud risk' (Lambert (Omnicom)), alongside The Trade Desk's testimony that it 'does not treat all ad exchanges the same'. The scope passages (Dkt 1861-1 II.B-C, definitions 23 and 35; Dkt 1858 p. 9 footnote and p. 9 'including private auction and open auction deals') read as the stream describes. | supported with qualification | Inference: the US competition record addresses the auction mechanics of one ad server and one exchange (DFP and AdX) for open-web display 'Indirect Demand' (open and private auctions); none of the liability opinion, the remedies opinion or the proposed judgment mentions curators, curation, curated deals or third-party deal IDs (the one occurrence of 'curate' is remedies-trial testimony, quoted at Dkt 1858 p. 85, that advertisers expect their DSPs to 'curate or prioritize one ad exchange over another'), so the record documents observed, court-found first-look, last-look and floor advantages for Google's exchange and a proposed restoration of publishers' per-source floor control, and establishes nothing about the default priority, floors or exclusivity of curated deals on any exchange. | scope, unit, date or wording narrowed to what the source carries | CUR-I-F26 |
| V-I-68 | Nine plaintiff rows with original and current courts, docket numbers, filing dates and coordination steps. | Refute-mode verifier for I.json re-opened the cited source | Every cell the stream sourced is carried by its source (OpenX caption; Equativ docket entries; the June 2026 case management plan's eight numbers; Teads' release and the brief's docket number, confirmed on the RECAP card as 1:26-cv-06591 filed 3 Aug 2026). Cells marked 'not opened' can now be filled from RECAP cards: PubMatic E.D. Va. 1:25-cv-01482 (filed 8 Sep 2025, terminated 18 Dec 2025); Magnite E.D. Va. 1:25-cv-01541 (filed 16 Sep 2025, terminated 18 Dec 2025); OpenX terminated 18 Dec 2025 (MDL order Doc 28 naming 'the ad exchange actions'); Index Exchange's original court was E.D. Va. (1:25-cv-02003, filed 10 Nov 2025, notice of voluntary dismissal 17 Dec 2025, terminated 18 Dec 2025), so the S.D.N.Y. action is a fresh filing; Sovrn 1:26-cv-01587 filed 25 Feb 2026; TRUSTX 1:26-cv-03135 filed 16 Apr 2026; Teads' docket carries a proposed case management plan of 14 Aug 2026 'This Doc | supported with qualification | Replace the Index Exchange row's original_court with 'E.D. Va. (1:25-cv-02003, filed 2025-11-10, voluntarily dismissed by notice of 2025-12-17, terminated 2025-12-18); S.D.N.Y. action 1:25-cv-10477 filed 2025-12-17 is a new filing, not a transfer'; fill PubMatic's original_number with 1:25-cv-01482 and Magnite's with 1:25-cv-01541; fill Sovrn's filing_date with 2026-02-25 and TRUSTX's with 2026-04-16; change Teads' coordination_step to 'coordinated in the MDL per docket entries of 14 August and 14 October 2026 (docket not opened in full)'. Source each from the RECAP search cards opened on 2026-11-03. | scope, unit, date or wording narrowed to what the source carries | data_tables: private_suits_against_google |
| V-I-69 | Thirty-nine instrument rows coded by status, including nineteen US state statute rows from the IAPP chart, the T-794/25 row ('pending') and the Belgian Market Court row. | Refute-mode verifier for I.json re-opened the cited source | All nineteen state rows match the chart's bill numbers, enactment years and effective dates (SB 1392, SB 190, SB 6, SB 227, HB 4, SB 619, SB 384, HB 154, SF 262, LB 1074, SB 255, SB 332, HB 1181, HF 4757, SB 541, SB 5, HB 15, H 7787, CCPA); the chart places SB 255 and SB 332 between Nebraska and Oregon in its alphabetical list, consistent with New Hampshire and New Jersey. The T-794/25 'pending' status is on InfoCuria. The Belgian row's 'annulled' is only part of IAB Europe's own account (see F25). The CPPA and FTC rows match their pages. The court and regulator rows match the sources checked under F06 to F20. | supported with qualification | Keep the nineteen state rows; add four rows coded 'statute (enacted, not in force)' for Oklahoma (SB 546, signed 2026-03-20, effective 2027-01-01), Alabama (Personal Data Protection Act, April 2026, effective 2027-05-01), Louisiana (SB 386, signed 2026-05-29, effective 2027-01-01) and Vermont (S.71, signed 2026-06-16, effective 2028-01-01), each sourced to an opened statute or IAPP record (only the Vermont item was opened here); amend the Belgian row per F25. | scope, unit, date or wording narrowed to what the source carries | data_tables: rules |
| V-J-70 | On 14 October 2026 LG Ad Solutions and Magnite announced a three-year extension under which LG makes Home Screen inventory available for agentic buying through Magnite's Seller Agent, with H/L the first buyer, for the Northern California Toyota Dealers Association, timed to the fall football season; the release names no protocol and does not say what the seller agent decides. | Refute-mode verifier for J.json re-opened the cited source | Release ('NEW YORK, Oct. 14, 2026'): 'LG Ad Solutions is making its Home Screen inventory available for agentic buying through Magnite's Seller Agent'; 'H/L is the first buyer to activate LG Home Screen inventory using Magnite's agentic buying capabilities'; Northern California Toyota Dealers Association; 'Timed to the fall football season'; Cobb and Laband quotes as claimed; 'a three-year extension of their partnership' across North America, APAC and EMEA; no AdCP or AAMP; the seller agent's decision is not described. AdExchanger (Alyssa Boyle, 14 October 2026, with interviews): H/L's buyer agent worked with Magnite's seller agent inside ClearLine using MCP ('the AI plumbing layer'); the agents 'set up and activated' H/L's brief targeting NFL viewers in market for a car; creative features 49ers quarterback Brock Purdy; Laband said the agent-led media plan was approved by LG and H/L befo | supported with qualification | On 14 October 2026 LG Ad Solutions and Magnite announced a three-year extension under which 'LG Ad Solutions is making its Home Screen inventory available for agentic buying through Magnite's Seller Agent', with independent agency H/L as 'the first buyer to activate LG Home Screen inventory' for the Northern California Toyota Dealers Association in a campaign timed to the fall football season; the release names no protocol and does not say what the seller agent decides, while AdExchanger's same-day report adds that H/L's buyer agent and Magnite's seller agent set up and activated the brief inside ClearLine over MCP and that LG and H/L approved the agent-led plan before it went live, and Magnite's own blog of 6 January 2026 describes the seller agent as supporting AdCP tests without stating a version. | scope, unit, date or wording narrowed to what the source carries | CUR-J-F15 |
| V-J-71 | Marketing Dive (17 March 2026) reported a Geloso Beverage Group campaign via Butler/Till 'planned and implemented completely by artificial intelligence agents using PubMatic's AgenticOS platform', run in December 2025 across CTV, online and mobile, with figures (5.5x, 40%, 98%, <1% DoubleVerify failures, 0% MFA) from press materials, none independently verified. | Refute-mode verifier for J.json re-opened the cited source | Marketing Dive, 'Published March 17, 2026': 'was planned and implemented completely by artificial intelligence agents' (AgenticOS named); 'The campaign, which ran in December 2025'; 'CTV, online and mobile'; 'geotargeted and aimed at those 21 and over'; 'reduced buy-side supply chain costs by 5.5x, according to press materials shared with Marketing Dive' (the only figure tied to that exact phrase); '40% more impressions than planned for', '98% video completion rate', 'Less than 1% of inventory failed DoubleVerify standards' are given 'per press details' / 'press materials claim'; '0% of the inventory comprised made-for-advertising sites' is attributed to PubMatic ('PubMatic says'); Butler/Till's 'enterprise Claude interface' named; 'early test' is not verbatim (the article says 'an early look at the technology's results'); AdCP, Activate and a January 2026 flight are not mentioned. The 1 | supported with qualification | Marketing Dive reported on 17 March 2026 that a Clubtails campaign for Geloso Beverage Group, run by agency Butler/Till, was 'planned and implemented completely by artificial intelligence agents' on PubMatic's platform (named AgenticOS by March 2026; the 15 December 2025 launch release, read in a syndicated copy, says the campaign 'is running through PubMatic's Activate platform' and calls the workflow 'AdCP-enabled', citing AdCP as 'an open protocol from AgenticAdvertising.org'), that it 'ran in December 2025' across CTV, online and mobile, geotargeted to adults 21 and over, with a brief submitted through Butler/Till's enterprise Claude interface; the figures, buy-side supply chain costs 'reduced ... by 5.5x' ('according to press materials shared with Marketing Dive'), '40% more impressions than planned for', 'a 98% video completion rate' and 'Less than 1% of inventory failed DoubleVeri | scope, unit, date or wording narrowed to what the source carries | CUR-J-F16 |
| V-J-72 | Across protocols and vendor interfaces, the highest state on non-author evidence is: AdCP, a publisher-asserted transaction (16 October 2025) plus vendor/agency-reported live campaigns on member platforms where the opened record names no protocol; AAMP, reference connectors and a sandbox pilot with OpenProposal a draft; Prebid Sales Agent, alpha connector; OpenX/Chalice, named customer without period; others announced, beta or unnamed; so H7 is at most supported with qualification on interested-party evidence, and unresolved if independent evidence is required. | Refute-mode verifier for J.json re-opened the cited source | The per-protocol records hold as the stream states them, with two upgrades on the AdCP side: the PubMatic/Butler-Till record is AdCP-named on an opened syndicated copy of the 15 December 2025 release (campaign 'running through PubMatic's Activate platform'); Magnite documents its seller agent as supporting AdCP tests, with a first test in December 2025 (Scope3 as buyer agent) and MiQ's claim 'In December we set live one of the first AdCP buys working with Magnite's seller agent' (vendor and partner assertions, no customer, spend or decision named); and the H/L transaction of October 2026 involved a buyer agent and seller agent over MCP with human approval per AdExchanger. Every production statement still comes from a vendor, its agency client, a partner quoted by the vendor or the protocol's publisher. | supported with qualification | Across the protocols and vendor interfaces screened, the highest state reached on evidence from a party other than the author is: AdCP, a demonstrated transaction asserted by the protocol's publisher (16 October 2025) plus vendor- and agency-reported live campaigns on vendor platforms that are AdCP members, of which the PubMatic/Butler-Till campaign (December 2025, on PubMatic's Activate platform) is named 'AdCP-enabled' in the vendors' 15 December 2025 release and the Magnite/LG/H-L transaction (October 2026) is linked to AdCP only through Magnite's January 2026 statement that its seller agent supports AdCP tests, while Magnite and MiQ assert December 2025 AdCP tests and 'one of the first AdCP buys' without a customer or period; AAMP, reference connectors (SDKs v2.4.2) and a sandbox pilot, with OpenProposal a draft; the Prebid Sales Agent, an alpha connector with no named deployment; Op | scope, unit, date or wording narrowed to what the source carries | CUR-J-F21 |
| V-J-73 | Row codes status 'beta' with status_stated 'Status: alpha'. | Refute-mode verifier for J.json re-opened the cited source | The README says alpha ('Status: alpha'; 'alpha software tracking a beta protocol'). The table's status vocabulary (live, beta, announced, retired, unconfirmed) has no 'alpha', so 'beta' overstates the source's own label. | supported with qualification | Add 'alpha' to the table's status vocabulary and code the Prebid Sales Agent row 'alpha', or keep 'beta' and state in the table notes that alpha is folded into beta. | scope, unit, date or wording narrowed to what the source carries | CUR-J data_tables.agentic row 'Prebid Sales Agent' (status column) |
| V-J-74 | Row dates AAMP 2.0 to '2026-04' with the standard page as source and the note 'day not confirmed on any page opened (brief says 23 April 2026)'. | Refute-mode verifier for J.json re-opened the cited source | The standard page carries no date for 2.0 (only 'AAMP 2.0 added 5 new transaction types' and 'announced in early 2026'), so the cited source does not carry the month. The SDK repositories' v2.0 annotated tags and release objects were created on 8 July 2026 (buyer-agent 11:22Z, seller-agent 10:44Z) and point to commits dated 13 April 2026 (buyer, cc9c4d77) and 17 April 2026 (seller, 0f544a54). The press index page 1 lists no AAMP 2.0 release. | supported with qualification | Replace the row's source basis: date '2026-04' rests on the brief and on the v2.0 tagged commits of 13 April (buyer-agent) and 17 April (seller-agent) 2026, while the v2.0 GitHub releases themselves are dated 8 July 2026 and the standard page gives no date; keep 'April 2026, day unconfirmed' and cite the repositories, not the standard page, for the month. | scope, unit, date or wording narrowed to what the source carries | CUR-J data_tables.agentic row 'AAMP 2.0' and data_tables.protocol_states (AAMP versions) |
| V-J-75 | Evidence cell: 'PubMatic AgenticOS campaign for Geloso via Butler/Till (December 2025) per press materials; AdCP wording only in unopened release'. | Refute-mode verifier for J.json re-opened the cited source | The 15 December 2025 release (syndicated copy, 16 December 2025) is headlined 'PubMatic and Butler/Till Launch AdCP-Enabled Agentic AI Campaign Across CTV', names AdCP as 'an open protocol from AgenticAdvertising.org' and says the campaign 'is running through PubMatic's Activate platform'; AgenticOS is not in it. | supported with qualification | Replace the evidence cell with: 'PubMatic campaign for Clubtails (Geloso Beverage Group) via Butler/Till, December 2025, on PubMatic's Activate platform, called AdCP-enabled in the vendors' 15 December 2025 release (syndicated copy opened; Business Wire original 403) and reported with unverified results by Marketing Dive on 17 March 2026 under the AgenticOS name'. | scope, unit, date or wording narrowed to what the source carries | CUR-J data_tables.protocol_states row 'AdCP / 4 / documented production use' |
| V-J-76 | Evidence cell: 'none opened; AdExchanger reports partial tests on 15 October 2025'. | Refute-mode verifier for J.json re-opened the cited source | Magnite's blog of 6 January 2026 ('Magnite Team') says Magnite embedded its seller agent in SpringServe, 'ran our first test in December with Scope3 serving as the buyer agent', had 'additional pilots lined up to test in early 2026', quotes MiQ's John Goulding 'In December we set live one of the first AdCP buys working with Magnite's seller agent', and quotes LG Ad Solutions and Warner Bros. Discovery as testing or exploring the seller agent; ppc.land (7 January 2026) relays the same and states no spend. No customer, period beyond the month, decision or transaction context is named. | supported with qualification | Add to the evidence cell: 'Magnite blog, 6 January 2026: first seller-agent test in December 2025 with Scope3 as buyer agent; MiQ asserts a live AdCP buy in December 2025; vendor and partner assertions, no customer or spend named' and set evidence_party to 'vendor and partner quoted by vendor'. | scope, unit, date or wording narrowed to what the source carries | CUR-J data_tables.protocol_states row 'AdCP / 3 / demonstrated interoperability test' |
| V-J-77 | Version 'n/a'; evidence 'H/L activation of LG Home Screen for the Northern California Toyota Dealers Association; decision not described'. | Refute-mode verifier for J.json re-opened the cited source | AdExchanger (14 October 2026): H/L's buyer agent worked with Magnite's seller agent inside ClearLine using MCP; the agents set up and activated the brief; the plan was approved by LG and H/L before going live; AdCP not named. Magnite's blog (6 January 2026) presents the seller agent as supporting AdCP tests, no version. | supported with qualification | Add to the evidence cell: 'per AdExchanger (14 October 2026), a buyer agent and the seller agent transacted over MCP inside ClearLine with human approval of the plan; Magnite describes the seller agent as AdCP-capable (blog, 6 January 2026) but no AdCP version is stated for this transaction'; keep version 'n/a'. | scope, unit, date or wording narrowed to what the source carries | CUR-J data_tables.protocol_states row 'Magnite Seller Agent (vendor interface) / 5' |
| V-U-78 | OpenXSelect is a curation and supply-side targeting platform across video, display and CTV, transacted across unnamed DSPs, with explicit US statements and unperiodised figures of 200k+ domains, 245 million+ US CTV users and 80+ data partners. | Refute-mode verifier for U.json re-opened the cited source | Hero: 'OpenXSelect™ is a next-generation curation and supply-side targeting (SST) platform for brands, agencies, and data partners'. 'In the U.S., double your reach against qualified cross-channel audiences' with a reach of 'over 245 million connected TV (CTV) users'. Footnote: 'The OpenX identity graph is available in the U.S. only.' '200k+' 'Direct publisher domains'. 'With more than 80 directly integrated data partners'. 'easily transacted across demand-side platforms (DSPs)'. No DSP named; no fee. The '2x higher match rate' statistic is sourced to 'a 2025 Merkury campaign for a U.S. grocery chain'. Lorem-ipsum copy sits in navigation menu descriptions (Company, Careers, Leadership, Knowledge Hub, News & Press, Blogs, Case Studies, Videos, Research). Copyright © 2026 OpenX. | supported with qualification | OpenXSelect is sold as a curation and supply-side targeting platform for brands, agencies and data partners across video, display and CTV, transacted across unnamed DSPs, with explicit US statements ('In the U.S.'; identity graph 'available in the U.S. only'), unperiodised figures of 200k+ direct publisher domains, 245 million+ US CTV users and 80+ data partners, and one dated US campaign reference (a 2025 Merkury campaign for an unnamed U.S. grocery chain) that is a candidate, not yet a documented, US production-use record for dimension 9. | scope, unit, date or wording narrowed to what the source carries | CUR-U-F02 |
| V-U-79 | Google Ad Manager curation settings remain in Beta; partners define the fee, described as 'a single fee that is fully transparent to the buyer', paid by Google to the partner 'taken out of the bid price'; twelve named curation partners listed. | Refute-mode verifier for U.json re-opened the cited source | Banner: 'This feature is in Beta' and 'Features in Beta phase might not be available in your network.' 'Curation partners define the fee of the data segment or curated package.' 'It is a single fee that is fully transparent to the buyer.' The fuller payment sentence reads 'Google then handles the fee payment to the curation partner (taken out of the bid price before it enters the auction).'; a shorter form '(taken out of the bid price)' also appears. 'A curation partner shares a Deal ID with the buyer, which can be subscribed to within the agency seat.' Partners in page order: Adelaide; Audigent/Experian (Audience); Audigent/Experian (Contextual); DoubleVerify Inc.; Integral Ad Science; LiveRamp, Inc.; Multilocal Media; Permutive; PrimeAudience; Proximic by Comscore; SWYM.ai; System1. Signals named: URL, App ID, Pub ID, GPID, metro-level geography, TCF consent string; identifiers: Device | supported with qualification | Google Ad Manager's curation settings remain in Beta on the help page read on 3 November 2026 (with the caveat that Beta features 'might not be available in your network'); curation partners define the fee of a segment or package, Google describes it as 'a single fee that is fully transparent to the buyer' and pays it to the partner 'taken out of the bid price before it enters the auction'; the page lists twelve named curation partners: Adelaide, Audigent/Experian (Audience), Audigent/Experian (Contextual), DoubleVerify Inc., Integral Ad Science, LiveRamp, Inc., Multilocal Media, Permutive, PrimeAudience, Proximic by Comscore, SWYM.ai and System1. | scope, unit, date or wording narrowed to what the source carries | CUR-U-F08 |
| V-U-80 | InMobi's Buyer Hub page describes 'An AI Agent for Media Buyers' that lets users 'Discover inventory, build deals, and optimize campaigns through natural language', records agency access dated 'August 31st, 2026', names no exchange, DSP or deal-ID mechanism, and mentions the US only in a sample query. | Refute-mode verifier for U.json re-opened the cited source | Title 'Buyer Hub — An AI Agent for Media Buyers \| InMobi Advertising'. 'Discover inventory, build deals, and optimize campaigns through natural language, with traders in control of every decision.' 'Buyer Hub Opens to Agencies.' 'Agency Access \| August 31st, 2026'. The US appears only in image alt text ('What CTV inventory is available in the US by publisher?'), not body text. A feature label reads 'Self Serve Premium Supply'. 'curators', 'exchange', 'SSP', 'DSP', 'deal ID', 'Scope3', 'Interchange', 'fee' absent from the first 100,000 characters; the remaining 97,836 were not read by this verifier either. | supported with qualification | InMobi's Buyer Hub page describes 'An AI Agent for Media Buyers' that lets users 'Discover inventory, build deals, and optimize campaigns through natural language, with traders in control of every decision', records 'Agency Access \| August 31st, 2026', names no exchange, DSP or deal-ID mechanism, labels a feature 'Self Serve Premium Supply', and mentions the US only in the alt text of a sample-query image; InMobi's Business Wire release of 13 August 2026 could not be opened (HTTP 403). | scope, unit, date or wording narrowed to what the source carries | CUR-U-F10 |
| V-U-81 | Magnite's blog of 9 September 2026 describes ClearLine Activate's self-serve audience layer and its named data partners, without any beta/GA word, DSP, fee, scale figure or channel statement. | Refute-mode verifier for U.json re-opened the cited source | Byline 'Magnite Team', 'September 9, 2026 \| 4 min read'. 'Magnite is simplifying that process through enhanced self-serve audience capabilities in ClearLine Activate (CLA).' 'Data onboarding and marketplace integrations: Experian, LiveRamp, TransUnion, Throtle, Cadent'. 'Direct 3P Data Partners: Acxiom, Comscore, Captify, Lotame, Nielsen, OpenAP, Samba, TruthSet, Genius Sports, TargetSmart'. 'Clean Rooms & CDPs: Snowflake, Databricks, Hightouch, Growthloop, Optable, Adobe'. No 'beta', 'generally available', 'DSP', 'fee', 'CTV' or 'audio' in the body. Mike Laband appears in the body, not the byline: 'Mike Laband, group SVP, US revenue at Magnite, put it plainly:' (a quote in the 'Why this matters for buyers' section). Tags: Addressability, Audience, Data. [108] | supported with qualification | Magnite's blog of 9 September 2026 (byline 'Magnite Team') describes ClearLine Activate (CLA) as the self-serve audience layer of ClearLine, naming its data onboarding partners (Experian, LiveRamp, TransUnion, Throtle, Cadent), direct third-party data partners (Acxiom, Comscore, Captify, Lotame, Nielsen, OpenAP, Samba, TruthSet, Genius Sports, TargetSmart) and clean rooms, quotes Mike Laband, group SVP, US revenue, in the body, and carries no beta or general-availability word, named DSP, fee, scale figure or channel statement. | scope, unit, date or wording narrowed to what the source carries | CUR-U-F11 |
| V-U-82 | Experian's curation page sells Curated Deals, Deal Engine and Deal Marketplaces, activatable 'on platforms like demand-side platforms (DSPs) and supply-side platforms (SSPs)', under Experian Information Solutions, Inc., with no channel, named platform, fee, figure, geography or Audigent mention. | Refute-mode verifier for U.json re-opened the cited source | 'Bespoke private marketplaces (PMPs) that combine premium data and inventory into a single Deal ID.' 'Self-service deal creation that puts the control of audiences, inventory, and deal strategy in the hands of buyers'. 'A library of deals built for speed and scale, giving buyers immediate access to pre-built curated deals.' The DSP/SSP sentence is an FAQ answer about Deal Marketplaces specifically: 'Experian Deal Marketplaces provides instant access to always-on, pre-built curated deals that are ready to activate on platforms like demand-side platforms (DSPs) and supply-side platforms (SSPs)'. 'no matter the channel' appears in an identity sentence. Footer '©2026 Experian Information Solutions, Inc.' 'Audigent' and 'Experian Marketing Services' absent (breadcrumb reads 'Experian Marketing'). Site-level cues: a country selector defaulting to 'United States' and a US toll-free number (877) | supported with qualification | Experian's Consumer View curation page sells three named services: Curated Deals ('Bespoke private marketplaces (PMPs) that combine premium data and inventory into a single Deal ID'), Deal Engine (self-service deal creation for buyers) and Deal Marketplaces (a library of pre-built curated deals), the last described as ready to activate 'on platforms like demand-side platforms (DSPs) and supply-side platforms (SSPs)', under the copyright of Experian Information Solutions, Inc., with no channel, named platform, fee or figure, no mention of Audigent, and only site-level US cues (a 'United States' country-selector default and a US toll-free number) rather than a product geography statement. | scope, unit, date or wording narrowed to what the source carries | CUR-U-F17 |
| V-U-83 | Adelaide's page states high-AU PMPs 'are available for digital display, OLV, and CTV across SSPs' including PubMatic, Microsoft Advertising (Xandr), TripleLift, Magnite and OpenX, that buyers use them 'across Google Display & Video 360, The Trade Desk, and Yahoo DSP', and the metric is 'Available in more than 20 of the world's largest advertising markets'. | Refute-mode verifier for U.json re-opened the cited source | 'Adelaide's high-AU PMPs are available for digital display, OLV, and CTV across SSPs, including PubMatic, Microsoft Advertising (Xandr), TripleLift, Magnite, and OpenX.' The DSP sentence is a client-results statement about one partner's clients: 'Teads clients have seen impressive results using high-AU PMPs across Google Display & Video 360, The Trade Desk, and Yahoo DSP.' Other case lines: 'a global credit card company leveraged Magnite's high-AU PMPs'; 'Colgate and The Trade Desk'. 'Available in more than 20 of the world's largest advertising markets'. 'United States'/'U.S.' absent. The page also names Adlook, Adobe DSP, Amazon Ads, Equativ, InMobi, Teads and Viant (Index Exchange as a logo only). No fee; the only price wording is 'a fair price'. © 2019–2026 Adelaide Metrics Inc. | supported with qualification | Adelaide's programmatic activation page states that 'Adelaide's high-AU PMPs are available for digital display, OLV, and CTV across SSPs' including PubMatic, Microsoft Advertising (Xandr), TripleLift, Magnite and OpenX; that 'Teads clients have seen impressive results using high-AU PMPs across Google Display & Video 360, The Trade Desk, and Yahoo DSP' (a client-results statement, not a general route statement); and that the metric is 'Available in more than 20 of the world's largest advertising markets', without naming the US or any fee. | scope, unit, date or wording narrowed to what the source carries | CUR-U-F19 |
| V-U-84 | Seedtag's NeuroX Curation is 'The Self-Serve Interface to the Neuro-Contextual Exchange', giving 'pre-bid control of Seedtag's Neuro-Contextual Exchange' to 'Build Web and CTV deals in a single flow' and push 'standard Deal IDs into existing DSPs', which reads as cohort A by the decision tree's first step. | Refute-mode verifier for U.json re-opened the cited source | 'Seedtag NeuroX Curation: The Self-Serve Interface to the Neuro-Contextual Exchange.' 'Give your team pre-bid control of Seedtag's Neuro-Contextual Exchange.' 'Seedtag NeuroX Curation transforms contextual into a self-serve workflow for activation teams.' 'Build Web and CTV deals in a single flow, with the same Liz targeting, margin logic, and premium supply.' 'Describe the campaign, review the suggested audiences, and push to your DSP.' 'One workspace, one strategy, standard Deal IDs into existing DSPs.' 'One owned supply path'. 'OLV' and 'Display' appear only as navigation labels; the body names Web and CTV deals. 'Coming Soon' labels sit beside menu items (Resources/Company), not the product. No DSP or SSP named; the footer's 'CTV Client Login' links to platform.beachfront.io. No geography, date or copyright. The page does not say whether the exchange carries non-Seedtag publishers. | supported with qualification | Seedtag's NeuroX Curation is 'The Self-Serve Interface to the Neuro-Contextual Exchange', giving activation teams 'pre-bid control of Seedtag's Neuro-Contextual Exchange' over 'One owned supply path' to 'Build Web and CTV deals in a single flow' and push 'standard Deal IDs into existing DSPs'; the page's body names only web and CTV deals (OLV and display are navigation labels), and the self-serve build tool on the operator's own exchange reads as cohort A by the decision tree's first step, subject to the author's decision. | scope, unit, date or wording narrowed to what the source carries | CUR-U-F24 |
| V-U-85 | Samba's Audiences page (2 November 2026, interest-marked) describes audience segments only and does not mention PMPs, curated deals or deal IDs, so 'Samba PMPs' is not documented as a purchased deal service on the page opened. | Refute-mode verifier for U.json re-opened the cited source | This verifier opened the samba.com homepage (page metadata 'Nov 2, 2026, 10:33 PM UTC'; Copyright © 2026 Samba) rather than the Audiences page. The homepage carries the sentence stream F relied on: 'Effective targeting with exclusive segments and PMPs built from Samba's deterministic cross-screen...' listed under the Audiences product. 'private marketplace', 'curated', 'deal', 'deal ID' and 'Semasio' do not appear; no platform is named where PMPs run; the page does not say PMPs are a separately purchased product. | supported with qualification | Samba's site documents PMPs in one homepage sentence ('exclusive segments and PMPs built from Samba's deterministic cross-screen' signals, page metadata 2 November 2026) under its Audiences product, naming no platform, deal ID, channel or fee, and the Audiences page itself (per U) describes segments only, so the register's 'Samba PMPs' row rests on a single marketing sentence and is not documented as a purchased deal service. | scope, unit, date or wording narrowed to what the source carries | CUR-U-F26 |
| V-U-86 | Dentsu Aegis Network's release of 15 September 2020 launched 'Dentsu Curate' on Xandr's platform in Hong Kong, South East Asia and Singapore, 'to be scaled to other APAC markets', as a 'Supply Path Optimisation & Curated Marketplace offering'; no later page documents status, channels or US availability. | Refute-mode verifier for U.json re-opened the cited source | 'Published on: 15th September 2020'. 'Dentsu Aegis Network and Xandr, AT&T's advanced advertising company, today announced a strategic partnership'. 'Dentsu Curate, the new programmatic supply solution from DAN'. 'It will be the first of its kind in APAC'. 'great success in Hong Kong and South East Asia ... driving upwards of 10% increase in viewability and 15% improvement in campaign performance.' 'This solution will now be scaled to other APAC markets.' 'Supply Path Optimisation & Curated Marketplace offering'. 'In Singapore for a BFSI client ... covering 30 markets'. Product written 'Dentsu Curate'. The US is mentioned only in a site-footer legal disclaimer ('...deemed, not applicable to the U.S.') and a 'USA (English)' site link, not in the product text. | supported with qualification | Dentsu Aegis Network's release of 15 September 2020 launched 'Dentsu Curate' as 'a new programmatic supply solution for clients' on Xandr's technology platform, deployed in Hong Kong, South East Asia and Singapore and 'to be scaled to other APAC markets', described as a 'Supply Path Optimisation & Curated Marketplace offering'; the product text names no US availability (the page's only US references are a site-footer legal disclaimer and a USA site link), and no later page documents its status or channels. | scope, unit, date or wording narrowed to what the source carries | CUR-U-F30 |
| V-U-87 | Onetag presents a 'Curation Platform Login' and 'Exchange Login', claims 2,000+ publishers, 5,000+ buyers, 35+ DSP partners and 50,000+ placements (unperiodised), lists offices in Pisa, Milan and London with no US statement, has acquired Aryel, and is a candidate cohort A product pending a product page. | Refute-mode verifier for U.json re-opened the cited source | Statistics: '50,000+ high-impact placements'; '200+ top brands'; '2,000+ global premium publishers'; '5,000+ buyers trading'; '35+ global DSP partners'. 'Curation Platform Login' (curation.onetag.com) and 'Exchange Login' are navigation items; no curation product page is linked. 'We've acquired Aryel to create the new Value Exchange for Programmatic Advertising'. 'Dynamically filter and optimize impressions before every bid. No more static pre-built deals'. 'Benefit from custom AI optimization loops delivering 20-50% benchmark outcomes improvement – with no additional fee'. Offices: Pisa (Porta a Mare 56121), Milan (Via Morimondo 26), London (1 Fetter Lane EC4A 1BR). 'US', 'United States', 'New York' absent. No copyright line, legal entity or visible date (upload paths 2025/12, 2026/10). | supported with qualification | Onetag, discovered in the sweep and absent from the register, presents a 'Curation Platform Login' and an 'Exchange Login' without a product page, claims '2,000+ global premium publishers', '5,000+ buyers trading', '35+ global DSP partners', '50,000+ high-impact placements' and '200+ top brands' (all unperiodised), advertises 'custom AI optimization loops' 'with no additional fee', lists offices in Pisa, Milan and London with no US statement, and has acquired Aryel; it is a candidate cohort A product pending a product page. | scope, unit, date or wording narrowed to what the source carries | CUR-U-F32 |
| V-U-88 | Permutive's documentation describes curation in which 'Permutive handles all demand generation, audience packaging, deal creation, campaign optimization, and buyer reporting', advertisers access 'Permutive-branded curated deals' through Monetize, Index Exchange Marketplaces, PubMatic, Google Curated Deals and other SSP curation seats including OpenX, with 'transparent fee structures capped at 40%'; a candidate cohort B product. | Refute-mode verifier for U.json re-opened the cited source | 'Permutive handles all demand generation, audience packaging, deal creation, campaign optimization, and buyer reporting.' '...made available to advertisers as part of Permutive-branded curated deals.' '...can access through Monetize, Index Exchange Marketplaces, Pubmatic, Google Curated Deals, and other SSP curation seats.' OpenX appears only as a bullet under Activation. 'Permutive maintains transparent fee structures capped at 40%.' (no base, no payer; no fee section). Dependencies table: 'Required for Monetize, Index Exchange, and Pubmatic curation activation.' Core Product table (Web / iOS / Android / CTV / API Direct) marks only Web 'Yes' for cohort signal activation, Standard Cohorts and Custom Cohorts; the other four are 'No'. Mentions a 'Curation Order Form' without naming the contracting entity. Changelog 'No changes listed yet'; no date or geography. | supported with qualification | Permutive's documentation describes a curation product in which publishers opt in first-party signals, 'Permutive handles all demand generation, audience packaging, deal creation, campaign optimization, and buyer reporting', advertisers access 'Permutive-branded curated deals' 'through Monetize, Index Exchange Marketplaces, Pubmatic, Google Curated Deals, and other SSP curation seats' (OpenX listed as an activation bullet), 'Permutive maintains transparent fee structures capped at 40%' with no base or payer stated, and the product's environment table marks only Web as supported (iOS, Android, CTV and API Direct 'No'); it is a candidate cohort B product absent from the register. | scope, unit, date or wording narrowed to what the source carries | CUR-U-F33 |
| V-U-89 | Across the 34 readable pages, US or North America is named as a market on six (OpenX, Sovrn, Verve, WPP Media 2022, InMobi, Roku), US office/dateline evidence on five more (TripleLift, Audigent, Comscore, Stagwell, IAS), documented non-availability on none, documented non-US scope on one (Dentsu), and US production use is documented on none. | Refute-mode verifier for U.json re-opened the cited source | Confirmed on the pages opened: OpenX ('In the U.S.'; identity graph 'available in the U.S. only'), Sovrn ('US / EU / CA / APAC'), WPP Media 2022 ('EMEA and North America'), Roku (site banner 'in the US'), Dentsu (APAC), TripleLift (New York dateline and address), Comscore (Reston, VA), Stagwell (New York City), IAS (New York dateline). Qualifications: InMobi's 'US' is image alt text, not copy; Dentsu's page also carries a US footer legal disclaimer; Experian's page has a site-level 'United States' country-selector default and a US toll-free number (not counted by U); Magnite's 'US revenue' attribution is to a quoted executive, not the author; Multilocal's 'US' is a site selector (U counts it as such). Candidate US-use records U did not record: OpenX's 2x match-rate statistic is 'from a 2025 Merkury campaign for a U.S. grocery chain' (US, dated, customer unnamed); Index's page carries a n | supported with qualification | Across the 34 readable vendor pages, statements naming the US or North America as a market appear on six (OpenX; Sovrn; Verve (not re-verified); WPP Media, 2022; InMobi, in image alt text only; Roku, in a site banner), US office or dateline evidence on five more (TripleLift, Audigent (not re-verified), Comscore, Stagwell, IAS), documented non-availability on none, and a documented non-US scope on one (Dentsu Curate, APAC, whose page's only US references are a footer legal disclaimer); the remaining pages carry no product geography statement (Experian's has only a site-level United States country default), and no page documents dimension 9's US production use for a product, the nearest candidates being OpenX's '2025 Merkury campaign for a U.S. grocery chain' (customer unnamed) and Index's Butler/Till case-study title (location not stated). | scope, unit, date or wording narrowed to what the source carries | CUR-U-F34 |
| V-U-90 | Vendor pages confirm the register's operating-entity names for PubMatic, Magnite, Nexxen, Verve, Experian Information Solutions, Lotame Solutions, Comscore (NASDAQ: SCOR), Adelaide Metrics Inc., Integral Ad Science, Inc., Roku, Inc. and Stagwell (NASDAQ:STGW); no page states an HQ beyond an office address or dateline; Sovrn's footer reads 'Sovrn, Inc.'. | Refute-mode verifier for U.json re-opened the cited source | Confirmed: '© Sovrn, Inc.'; TripleLift footer '53 W. 23rd St., 12th Floor, New York, NY 10010'; 'Stagwell (NASDAQ:STGW)'; 'Comscore (NASDAQ: SCOR)' with Reston, VA dateline; '©2026 Experian Information Solutions, Inc.'; '© 2019–2026 Adelaide Metrics Inc.'; '© 2026 Integral Ad Science, Inc.'; '© Copyright 2026 WPP Media Limited'; '© Multilocal Media' (no 'Ltd'); Onetag gives office addresses and no entity. Not re-verified by this verifier: the PubMatic (403), Magnite (copyright line not returned), Nexxen, Verve and Lotame copyright lines. | supported with qualification | Vendor pages opened by this verifier confirm the entity names for Experian Information Solutions, Inc., Comscore, Inc. (NASDAQ: SCOR), Adelaide Metrics Inc., Integral Ad Science, Inc., Stagwell (NASDAQ:STGW) and WPP Media Limited, and Sovrn's footer reads 'Sovrn, Inc.' rather than the register's 'Sovrn Holdings, Inc.'; the PubMatic, Magnite, Nexxen, Verve, Lotame and Roku copyright lines rest on U's read alone; no page states a headquarters beyond an office address (TripleLift, Audigent, Onetag) or a dateline. | scope, unit, date or wording narrowed to what the source carries | CUR-U-F36 |
| V-U-91 | environments_advertised: 'functionality table: web, iOS, Android, CTV (not a deal-scope statement)' | Refute-mode verifier for U.json re-opened the cited source | The Core Product table has columns Web, iOS, Android, CTV and API Direct and marks only Web 'Yes' for cohort signal activation, Standard Cohorts and Custom Cohorts; iOS, Android, CTV and API Direct are 'No'. The row's wording presents four environments where the documentation supports one. | contradicted | Replace the environments_advertised cell with: 'functionality table: Web only marked Yes; iOS, Android, CTV and API Direct marked No (not a deal-scope statement)'. | the source does not carry the claim as first written | CUR-U data_tables.channel_scope (Permutive Curation (candidate)) |
| V-U-92 | environments_advertised: 'web display; web video (OLV); CTV' | Refute-mode verifier for U.json re-opened the cited source | The body advertises 'Build Web and CTV deals in a single flow'; 'OLV' and 'Display' appear only as navigation/menu labels for the site's channels, not in the product description. | supported with qualification | Replace with: 'web and CTV (body: Build Web and CTV deals); OLV and display appear only as site navigation labels'. | scope, unit, date or wording narrowed to what the source carries | CUR-U data_tables.channel_scope (Seedtag NeuroX Curation) |
| V-U-93 | environments_documented_live: 'named SSP and DSP routes stated for the three environments (vendor page, undated)' | Refute-mode verifier for U.json re-opened the cited source | One sentence states availability 'for digital display, OLV, and CTV across SSPs, including PubMatic, Microsoft Advertising (Xandr), TripleLift, Magnite, and OpenX' (the three environments collectively, no per-environment execution evidence); the DSP sentence reports what 'Teads clients' used. No dated availability statement or console step is on the page. | supported with qualification | Replace with: 'none documented per environment; one collective availability sentence across five named SSPs; DSP names are a Teads-client results statement'. | scope, unit, date or wording narrowed to what the source carries | CUR-U data_tables.channel_scope (Adelaide High-Attention PMPs) |
| V-U-94 | geo: 'author is group SVP, US revenue; no availability statement' | Refute-mode verifier for U.json re-opened the cited source | The byline is 'Magnite Team'; 'Mike Laband, group SVP, US revenue at Magnite' is quoted in the body ('put it plainly:'). The page carries no geography statement. [108] | contradicted | Replace the geo cell with 'not stated (a Magnite US-revenue executive is quoted in the body; no availability statement)' and correct source CUR-U-S01's geography field the same way. | the source does not carry the claim as first written | CUR-U data_tables.universe (ClearLine, geo cell) |
| V-U-95 | routes: 'SSPs: PubMatic, Microsoft Advertising (Xandr), TripleLift, Magnite, OpenX; DSPs: Google DV360, The Trade Desk, Yahoo DSP' | Refute-mode verifier for U.json re-opened the cited source | SSP list verbatim. The DSP list comes from 'Teads clients have seen impressive results using high-AU PMPs across Google Display & Video 360, The Trade Desk, and Yahoo DSP.' The page also names Adlook, Adobe DSP, Amazon Ads, Equativ, InMobi, Teads and Viant in text or logos without a route statement. | supported with qualification | Replace the DSP part with: 'DSPs named only as where Teads clients bought (Google DV360, The Trade Desk, Yahoo DSP); other platforms named without a route statement: Adlook, Adobe DSP, Amazon Ads, Equativ, InMobi, Teads, Viant'. | scope, unit, date or wording narrowed to what the source carries | CUR-U data_tables.universe (Adelaide High-Attention PMPs, routes cell) |
| V-U-96 | routes: "'demand-side platforms (DSPs) and supply-side platforms (SSPs)'; none named" | Refute-mode verifier for U.json re-opened the cited source | The sentence is an FAQ answer about Deal Marketplaces ('Experian Deal Marketplaces provides instant access to always-on, pre-built curated deals that are ready to activate on platforms like demand-side platforms (DSPs) and supply-side platforms (SSPs)'); the Curated Deals and Deal Engine descriptions name no platform type. | supported with qualification | Replace with: "Deal Marketplaces: 'on platforms like demand-side platforms (DSPs) and supply-side platforms (SSPs)'; Curated Deals and Deal Engine: none stated; none named". | scope, unit, date or wording narrowed to what the source carries | CUR-U data_tables.universe (Experian Curated Deals, routes cell) |
| V-U-97 | note: 'natural-language inventory discovery and deal building; whether it sits on an InMobi exchange not stated in the part read'; register_changes: 'live for agencies from August 31st, 2026; US only in a sample query; exchange basis still not stated' | Refute-mode verifier for U.json re-opened the cited source | Confirmed, plus two items U omits: the hero sentence ends 'with traders in control of every decision' (an authority statement for H7) and a feature label reads 'Self Serve Premium Supply'; the US sample query is image alt text. | supported with qualification | Add to both cells: "page states 'with traders in control of every decision' (authority: human approval); feature label 'Self Serve Premium Supply'; the US sample query is image alt text". | scope, unit, date or wording narrowed to what the source carries | CUR-U data_tables.universe and register_changes (Buyer Hub) |
| V-U-98 | note: "single fee 'fully transparent to the buyer', paid via Google out of the bid price; twelve named partners" | Refute-mode verifier for U.json re-opened the cited source | The fuller sentence reads 'taken out of the bid price before it enters the auction'; the Beta banner adds 'Features in Beta phase might not be available in your network.' | supported with qualification | Amend to: "single fee 'fully transparent to the buyer', paid via Google 'taken out of the bid price before it enters the auction'; Beta 'might not be available in your network'; twelve named partners". | scope, unit, date or wording narrowed to what the source carries | CUR-U data_tables.universe (Google Ad Manager Curation, note cell) |
| V-U-99 | Every scale or price figure found on the pages opened is listed. | Refute-mode verifier for U.json re-opened the cited source | Two unperiodised context figures are missing: Equativ 'Analyzes 400K+ sites' and Onetag '200+ top brands'. Onetag's 'with no additional fee' (custom AI optimization loops) is a type-only fee statement, not a scale figure, and belongs in the H2b record if Onetag is admitted. All rows checked against the opened pages carry the figure, unit and absence of period correctly. | supported with qualification | Add rows: Maestro by Equativ, '400K+', sites analysed, Equativ platform, unperiodised; Onetag (candidate), '200+', top brands, Onetag, unperiodised. | scope, unit, date or wording narrowed to what the source carries | CUR-U data_tables.scale_figures (Maestro by Equativ; Onetag (candidate)) |
| V-U-100 | proposed status: 'unclear: no PMP, curated deal or deal ID on the Audiences page of 2 November 2026; segments only' | Refute-mode verifier for U.json re-opened the cited source | The samba.com homepage (metadata 2 November 2026) does carry one PMP sentence under the Audiences product ('exclusive segments and PMPs built from Samba's deterministic cross-screen' signals) but names no platform, deal ID, channel or fee and does not describe PMPs as a separately purchased product. | supported with qualification | Amend to: 'unclear: PMPs appear in one homepage sentence under the Audiences product (2 November 2026) with no platform, deal ID, channel or fee; the Audiences page describes segments only'. | scope, unit, date or wording narrowed to what the source carries | CUR-U data_tables.register_changes (Samba PMPs) |
| V-audigent-101 | Legal name not stated in the sources opened; Experian's release names the company only as 'Audigent' and Audigent's pages carry no copyright line or legal entity name. | Refute-mode verifier for audigent.json re-opened the cited source | The homepage carries no copyright line or entity name (confirmed). But the Platform Privacy Policy linked from the homepage footer at /platform-privacy-policy (the profile tried the guessed URL /privacy-policy) is readable and names the entity: 'Predictive Pop, Inc., doing business as Audigent'. It reads 'Last updated September 2, 2026'. | supported with qualification | Predictive Pop, Inc., doing business as Audigent, per Audigent's Platform Privacy Policy (last updated 2 September 2026). Neither Experian's acquisition release nor the product pages name the entity, and no company registry record was opened. | scope, unit, date or wording narrowed to what the source carries | AUDIGENT entity legal_name |
| V-audigent-102 | The Audigent brand stays live ('a part of Experian' in the logo, Index's 2025 list, Google's help page); Event class: acquisition; the SmartPMP product continues, and no retirement or rebrand of it was found. | Refute-mode verifier for audigent.json re-opened the cited source | Digiday, 11 August 2026, by Seb Joseph, headlined 'Experian retires the Audigent brand, folding it into Experian Marketing Services', says: 'Twenty months after it was acquired by Experian, the Audigent brand is no more.' The brand 'has been absorbed into Experian Marketing Services'. A spokesperson is quoted: 'Audigent's curation capabilities remain an important part of our platform' and 'We continue to invest in Audigent's technology, talent and innovation.' The article gives no effective date. It does not mention SmartPMP, ContextualPMP or CognitivePMP, or whether audigent.com will redirect. The audigent.com pages still loaded on 3 November 2026, with the logo alt text 'Audigent a part of Experian logo'. | contradicted | Replace the last sentences of the status note with: 'Digiday reported on 11 August 2026 that Experian has retired the Audigent brand and absorbed it into Experian Marketing Services (no effective date; an Experian spokesperson said Audigent's curation capabilities remain part of its platform). The audigent.com product pages were still live on 3 November 2026, and no source opened states whether SmartPMP continues under its name. Event classes: acquisition (4 December 2024), then brand retirement into the parent (reported August 2026, trade press, not an official release).' | the source does not carry the claim as first written | AUDIGENT entity status_note |
| V-audigent-103 | The SmartPMP page speaks to 'digital media buyers' and says a deal is 'dropped into a campaign line item' (CUR-AUDIGENT-S04); DV360 buyers 'add the deal ID to your line item'; on Google Ad Manager the deal ID 'can be subscribed to within the agency seat'. | Refute-mode verifier for audigent.json re-opened the cited source | 'digital media buyers' is on the SmartPMP page (S01), which does not say 'dropped into a campaign line item'. That phrase is on the ContextualPMP page (S04): 'ContextualPMPs™ are dropped into a campaign line item.' The ContextualPMP page says 'media buyers', not 'digital media buyers'. The DV360 and Google passages are verbatim. [194, 193, 194] | supported with qualification | The SmartPMP page speaks to 'digital media buyers' (CUR-AUDIGENT-S01), and the ContextualPMP page says 'ContextualPMPs™ are dropped into a campaign line item' (CUR-AUDIGENT-S04). | scope, unit, date or wording narrowed to what the source carries | AUDIGENT who_buys |
| V-audigent-104 | Audigent claims 'over 20' SSPs, naming none; exchange records name three (Google Ad Manager, Index Exchange, Nexxen); data from publisher partners, Experian, The Weather Company, Comscore, Hadron ID; Google route signals and identifiers. | Refute-mode verifier for audigent.json re-opened the cited source | Each stated item is verbatim. The release names 'its unique identity suite (Hadron ID™)'. One omission: Experian's post of 6 November 2025 says Audigent was named Microsoft Advertising's Curator of the Year. It also says 'We will continue to set the standard by expanding curated deals in Microsoft Curate' and refers to 'activating Audigent-curated deals in Microsoft Curate'. That is a fourth exchange route, Microsoft Curate, documented by the owner and not checked on a Microsoft record. | supported with qualification | Add: 'An Experian post of 6 November 2025 adds Microsoft Curate as a fourth exchange route for Audigent-curated deals (owner marketing; no Microsoft record opened).' | scope, unit, date or wording narrowed to what the source carries | AUDIGENT supply_and_data_dependencies |
| V-audigent-105 | Audigent standalone revenue, profit or consideration not disclosed; Audigent is a privately held subsidiary and was a private company before the acquisition; full-year report unreadable (S17 truncated, S18 HTTP 403). | Refute-mode verifier for audigent.json re-opened the cited source | The release states no terms. The Digiday article of 11 August 2026 discloses no revenue, deal, spend or staff figure. No cited source states Audigent's ownership before the acquisition. [315, 316] | supported with qualification | No standalone revenue, profit or consideration for Audigent appears in the sources opened (Experian's release, the FY25 results, Digiday's report of 11 August 2026). The FY25 full-year financial report could not be read (CUR-AUDIGENT-S17, CUR-AUDIGENT-S18). | scope, unit, date or wording narrowed to what the source carries | AUDIGENT financial standalone (not disclosed) |
| V-audigent-106 | The SmartPMP, ContextualPMP, CognitivePMP marketplace and homepage pages are live and undated; Experian ownership appears only in the logo. | Refute-mode verifier for audigent.json re-opened the cited source | All four pages loaded on 3 November 2026 with no copyright line; the logo alt text reads 'Audigent a part of Experian logo'. The CognitivePMP page's only date is '2022' in an award image's alt text. Digiday reported on 11 August 2026 that the Audigent brand had been retired and absorbed into Experian Marketing Services. | supported with qualification | On 3 November 2026 the SmartPMP, ContextualPMP, CognitivePMP marketplace and homepage pages were live and undated, with Experian ownership shown only in the logo, although Digiday had reported on 11 August 2026 that Experian retired the Audigent brand into Experian Marketing Services. | scope, unit, date or wording narrowed to what the source carries | AUDIGENT timeline 2026-11-03 (status observation) |
| V-audigent-107 | No seller list, domain list, seller count or seller categories for any deal; composition wording limited to 'premium curated publisher inventory', 'some of the biggest names in entertainment, sports, and lifestyle', 'hand-vetted inventory' from 'top-tier publishers'; availability unconfirmed; grade D; disclosure unknown. | Refute-mode verifier for audigent.json re-opened the cited source | No seller or domain list and no deal-level seller count were found. The record does carry product-wide category and count wording that the entry partly quotes. The SmartPMP page cites publishers from 'some of the biggest names in entertainment, sports, and lifestyle'. The DV360 post says the deals cover 'various sectors' and offers 'contextual sports content, including live sports and sports-related programming across CTV and OLV'. The ContextualPMP page says the product isolates 'hundreds, or even thousands of contextually relevant inventory sources'. The homepage gives '476,080' web/mobile sites and '810+ Million' web/mobile pages. | supported with qualification | No seller list, domain list or seller count for an identified deal was found. The only category wording is product-wide marketing: publisher verticals ('entertainment, sports, and lifestyle'), DV360 packages covering 'various sectors' and sports content, and an unquantified 'hundreds, or even thousands' of inventory sources for ContextualPMP. No identified deal's contents were observed; DSP and exchange interfaces are authenticated. | scope, unit, date or wording narrowed to what the source carries | AUDIGENT capability composition_disclosure |
| V-audigent-108 | Provenance documented at source type and named providers only; no per-segment collection method, intended use, permissions or legal basis; the Platform Privacy Policy was not read; grade D; public. | Refute-mode verifier for audigent.json re-opened the cited source | The product-page quotations are verbatim. The Platform Privacy Policy at /platform-privacy-policy (Predictive Pop, Inc. dba Audigent; last updated 2 September 2026) is readable and documents several fields at platform level. - Collection methods: 'cookies, beacons, tags and mobile SDKs', and ID syncing that matches 'publisher cookies, Audigent cookies, third party data provider cookies'. - Data collected: IP addresses, cookie identifiers, IDFA/AAID, hashed email and imprecise location, with user behaviour collected from 'the websites and mobile applications of our publisher partners'. - Purposes: segmentation, targeted advertising, reporting and analytics. - Legal bases (UK/GDPR section): consent for targeting ('We rely on your consent when we deliver targeted advertising to you') and legitimate interests for other processing. - US law: the CCPA/CPRA and state laws are referenced. The po | supported with qualification | Audigent documents source type and named providers on its product pages. Its Platform Privacy Policy (last updated 2 September 2026) documents collection methods, purposes and legal bases (consent for targeted advertising, legitimate interests) for the platform as a whole. No source gives these fields per segment or for SmartPMP specifically. | scope, unit, date or wording narrowed to what the source carries | AUDIGENT capability data_provenance |
| V-audigent-109 | Only quantified claims are vendor case pages on CPC, CTR, VCR and an advertiser KPI, none on a registered quality dimension; 'In anchor terms, these are claims without qualifying measurement or a vendor study without method'; grade D; live; public. | Refute-mode verifier for audigent.json re-opened the cited source | The case figures are verbatim (stream C verifier, CUR-C-F04, verified as stated). The Amazon post says 'prioritizing quality over quantity through premium web inventory', with no measurement. [190] | supported with qualification | The only quality statements are unmeasured assertions: 'quality over quantity through premium web inventory' and 'hand-vetted inventory'. The quantified case figures (CPC, CTR, VCR, the advertiser's KPI) concern outcome metrics, not a registered quality dimension. In anchor terms, these are claims without qualifying measurement (anchor 1). | scope, unit, date or wording narrowed to what the source carries | AUDIGENT capability quality_evidence |
| V-audigent-110 | SmartPMP page advertises no channel and U's scope is 'not stated', so the anchor percentage cannot be computed against U's frozen scope; on the DV360 post's own scope, three of three environments (CTV, display, OLV) announced as available; Alcon case documents display; live; grade D; date 2025-08-20. | Refute-mode verifier for audigent.json re-opened the cited source | The SmartPMP page names no channel (confirmed). The DV360 post reads 'across connected TV (CTV), display, and online video (OLV)' and gives 'over 250 ready-to-run deals'. The Amazon post names only 'premium web inventory'. The Alcon page's 'display creatives' was not re-opened by this verifier; the stream C verifier did not quote it. | supported with qualification | Under the frozen definition (execution within the product's advertised scope as captured in workstream U), U recorded SmartPMP's scope as 'not stated', so the cell is n/e unless the register's channel scope is amended, with a dated register_amendments entry, to adopt the DV360 post's scope (CTV, display, OLV), against which three of three environments are announced as available in one owner post of 20 August 2025. | scope, unit, date or wording narrowed to what the source carries | AUDIGENT capability channel_execution |
| V-audigent-111 | No certification, accreditation or audit naming Audigent, SmartPMP or its deals; no TAG seal on the homepage; TAG Registry not queried; unconfirmed; disclosure unknown. | Refute-mode verifier for audigent.json re-opened the cited source | The homepage shows award images ('Best Data Management Platform', 'Best Workplaces for Innovators 2022') and no TAG seal or certification (confirmed). The Platform Privacy Policy states 'Audigent adheres to the Digital Advertising Alliance (DAA) Self-Regulatory Principles', says Audigent is an NAI member that 'adheres to the NAI Framework', and gives its IAB TCF vendor ID as 561. The TAG Registry page returned only static text. Its search widget did not load, so Audigent's TAG status remains unresolved; no workaround was attempted. | supported with qualification | No certification, accreditation or audit covering SmartPMP or its deals was found in the sources screened. The only assurance-adjacent statements are company-level privacy self-regulation in Audigent's Platform Privacy Policy (DAA principles, NAI membership, IAB TCF vendor ID 561), self-stated and not checked on the bodies' own records. The TAG Registry search could not be loaded. | scope, unit, date or wording narrowed to what the source carries | AUDIGENT capability assurance |
| V-audigent-112 | No source lists sellers, domains, a seller count or seller categories for any deal; composition wording limited to adjectives; 476,080 sites company-wide. | Refute-mode verifier for audigent.json re-opened the cited source | The passages are verbatim. The record also carries product-wide category wording: publisher verticals ('entertainment, sports, and lifestyle'), DV360 deals 'covering various sectors' plus 'contextual sports content ... across CTV and OLV', and 'hundreds, or even thousands' of inventory sources for ContextualPMP. | supported with qualification | No source opened for this profile lists the sellers, domains or a seller count of any identified SmartPMP, ContextualPMP, CognitivePMP or Audigent package. The composition wording is product-wide marketing, made up of adjectives ('premium curated publisher inventory', 'hand-vetted inventory'), publisher verticals ('some of the biggest names in entertainment, sports, and lifestyle'; DV360 deals 'covering various sectors' and sports content) and an unquantified 'hundreds, or even thousands' of inventory sources. The homepage's 476,080 web/mobile sites is a company-wide, unperiodised figure. | scope, unit, date or wording narrowed to what the source carries | CUR-AUDIGENT-F05 |
| V-audigent-113 | Provenance at source type and named providers only, with no per-segment collection method, intended use, permissions or legal basis on the pages read; the Platform Privacy Policy was not read. | Refute-mode verifier for audigent.json re-opened the cited source | All five passages are verbatim. The Platform Privacy Policy (last updated 2 September 2026) documents platform-level collection methods, data categories, purposes and legal bases (consent for targeted advertising, legitimate interests), but nothing per segment and nothing for SmartPMP. | supported with qualification | Audigent documents the provenance of its deal data per segment only at the level of source type and named providers ('100% 1st party audience data, never 3rd party cookies'; 'verified, opt-in'; Condé Nast, TransUnion, Warner Music Group, Penske, a360 Media, Fandom; Experian deterministic signals and The Weather Company in the DV360 packages). Its Platform Privacy Policy (last updated 2 September 2026) adds platform-wide collection methods (cookies, beacons, tags, mobile SDKs, ID syncing), purposes and legal bases (consent for targeted advertising; legitimate interests), with no per-segment or product-specific documentation. | scope, unit, date or wording narrowed to what the source carries | CUR-AUDIGENT-F06 |
| V-audigent-114 | Three exchanges document Audigent as a curation or data partner (Google Ad Manager, Index Exchange, Nexxen); Audigent's 'over 20' SSPs names none; counterevidence none found. | Refute-mode verifier for audigent.json re-opened the cited source | The Google, Index and Nexxen passages are verbatim; the Index passages were verified by the CUR-F-F24 and CUR-E-F21 verifiers. Separately, the owner's post of 6 November 2025 names a fourth exchange: Audigent was 'named Microsoft Advertising's Curator of the Year', with 'expanding curated deals in Microsoft Curate' and 'activating Audigent-curated deals in Microsoft Curate'. | supported with qualification | Three exchanges document Audigent as a curation or data partner on their own records: Google Ad Manager, Index Exchange and Nexxen. Experian's post of 6 November 2025 adds a fourth exchange route, Microsoft Curate, where Audigent was named Microsoft Advertising's Curator of the Year and Experian says it is 'expanding curated deals'; no Microsoft record was opened. Audigent's own claim of 'over 20' SSPs names none. | scope, unit, date or wording narrowed to what the source carries | CUR-AUDIGENT-F09 |
| V-audigent-115 | No certification, accreditation or audit covering Audigent or its deals, no supply-chain sample and no IVT statement in the sources screened. | Refute-mode verifier for audigent.json re-opened the cited source | No seal, certification or audit is on the homepage. The Platform Privacy Policy self-states DAA adherence, NAI membership and IAB TCF vendor ID 561. The TAG Registry search widget did not load. | supported with qualification | In the sources screened by 3 November 2026, this review found no certification, accreditation or audit covering SmartPMP or its deals, no supply-chain sample (ads.txt, sellers.json, SupplyChain) for any Audigent deal, and no statement of IVT filtering method or measurer. Audigent's Platform Privacy Policy self-states company-level privacy self-regulation (DAA principles, NAI membership, IAB TCF vendor ID 561), not checked on those bodies' records, and the TAG Registry search could not be loaded. | scope, unit, date or wording narrowed to what the source carries | CUR-AUDIGENT-F14 |
| V-audigent-116 | 'never 3rd party cookies' and 'without cookies or device IDs' sit beside homepage '1.7+ billion mobile advertising IDs' and '13.3+ billion 1st-party device graph'; the record does not say which identifiers SmartPMP segments use; tension is an inference. | Refute-mode verifier for audigent.json re-opened the cited source | The homepage figures and the product-page phrases are verbatim. The Platform Privacy Policy adds that Audigent collects 'cookie identifiers' and IDFA/AAID, uses 'cookies, beacons, tags and mobile SDKs', and runs ID syncing that matches 'publisher cookies, Audigent cookies, third party data provider cookies'. | supported with qualification | Add: 'Audigent's Platform Privacy Policy (last updated 2 September 2026) documents cookie identifiers, IDFA/AAID and ID syncing with third-party data provider cookies at platform level; which of these feed SmartPMP segments is not stated.' | scope, unit, date or wording narrowed to what the source carries | AUDIGENT counterevidence 1 (cookies and device IDs) |
| V-audigent-117 | Rationalisation or rebranding is possible (inference, not a documented plan). | Refute-mode verifier for audigent.json re-opened the cited source | Digiday, 11 August 2026: 'Twenty months after it was acquired by Experian, the Audigent brand is no more'; the brand 'has been absorbed into Experian Marketing Services'. | contradicted | Replace the inference with: 'Digiday reported on 11 August 2026 that Experian retired the Audigent brand into Experian Marketing Services; whether SmartPMP and Experian Curated Deals remain separately sold services is not documented.' | the source does not carry the claim as first written | AUDIGENT risks 1 (channel overlap; rationalisation or rebranding 'possible') |
| V-clearline-118 | CLA blog names data by type and by provider (5 onboarding, 10 direct third-party, 6 clean-room/CDP partners), the release adds Magnite Access, the 10-K the curator's own data; no per-segment collection method, intended use, permissions or legal basis; the inputs of Cognitiv's models and Culture Hive's score are not described; availability live, grade D, disclosure public. | Refute-mode verifier for clearline.json re-opened the cited source | CLA blog (Magnite Team, 9 September 2026): 'Apply third-party, advertiser first-party and permissioned publisher first-party audience data'; partner lists verbatim as in the profile; 'CLA now enables buyers to' (present tense, supports 'live'); no collection method, consent, permission or legal basis for any segment ('permissioned' and 'approved' imply permissions without documenting them). The Cognitiv release names no data input of the models. PPC Land, however, does describe the Cultural Relevance Score's inputs at category level, citing Culture Hive's 22 April 2026 launch description: 'social signals, behavioural patterns, community identity markers and shared values'; weighting, validation and thresholds are not disclosed in the Magnite materials. | supported with qualification | The CLA blog (9 September 2026) names the data by type (third-party, advertiser first-party and 'permissioned publisher first-party data') and by provider: onboarding and marketplace partners Experian, LiveRamp, TransUnion, Throtle and Cadent; direct third-party partners Acxiom, Comscore, Captify, Lotame, Nielsen, OpenAP, Samba, TruthSet, Genius Sports and TargetSmart; clean rooms and CDPs Snowflake, Databricks, Hightouch, Growthloop, Optable and Adobe; the curation release adds Magnite Access audiences and the 10-K the curator's own first- or third-party data. No page documents per-segment collection method, intended use, permissions or legal basis. The inputs of Cognitiv's models are not described; Culture Hive's score is described only at category level ('social signals, behavioural patterns, community identity markers and shared values', per PPC Land relaying Culture Hive's April 202 | scope, unit, date or wording narrowed to what the source carries | CLEARLINE capability data_provenance |
| V-clearline-119 | No quality claim with a named metric and no measurement for ClearLine curated deals; only 'premium' and benefit statements; Adalytics observations predate ClearLine curation and are recorded under supply-path integrity; availability beta, grade D. | Refute-mode verifier for clearline.json re-opened the cited source | No metric or measurement in the curation release, the Cognitiv release ('optimize their omnichannel campaigns to the desired KPIs', no KPI named), the CLA blog or the 2023 release; PPC Land: no outcome data, results or lift published. The curation release's Magnite boilerplate adds an unmeasured suitability and quality assertion the profile does not record: agencies and brands trust the platform 'to access brand-safe, high-quality ad inventory'. | supported with qualification | No quality claim with a named metric (MFA share, viewability, IVT, suitability, attention or carbon) and no measurement was found for ClearLine curated deals: the curation release, the Cognitiv release, the CLA blog and the 2023 launch release carry only 'premium', benefit statements such as 'significantly increases spend going towards working media' (2023) and, in Magnite's company boilerplate in the curation release, the unmeasured assertion that agencies and brands use its platform to 'access brand-safe, high-quality ad inventory'; PPC Land reports no outcome data for the Culture Hive packages. The only independent observation opened that touches Magnite deals (Adalytics, examples to November 2024) predates ClearLine curation, names no ClearLine deal and is recorded under supply-path integrity, not as evidence behind a ClearLine claim. | scope, unit, date or wording narrowed to what the source carries | CLEARLINE capability quality_evidence |
| V-clearline-120 | Advertised scope CTV and video with an undefined omnichannel extension; documented live: CTV via the LG Home Screen buy for H/L 'via ClearLine (release and AdExchanger)' and premium video via the Cognitiv statement; one of one (CTV-only) or one or two of four; availability live, grade D. | Refute-mode verifier for clearline.json re-opened the cited source | 10-K: 'our ClearLine product provides buyers with direct access to premium CTV inventory and integrated curation capabilities'; 'Finally, our ClearLine product is a self-service solution that provides buyers with direct access to premium CTV'; 'For ClearLine, advertisers and agencies use a self-service programmatic solution to directly access premium inventory'. 2023 release: 'premium video inventory'. Curation release: 'to unify curation and activation across the company's premium omnichannel footprint' (phrase confirmed); boilerplate 'CTV, online video, display, and audio'. LG/Magnite release (14 October 2026): ClearLine appears only in infrastructure language and Laband's 'Combined with LG Ad Solutions' expanded use of ClearLine'; the release does not say the H/L buy ran through ClearLine. AdExchanger (per the J verifier): H/L's buyer agent worked with Magnite's seller agent 'within C | supported with qualification | Advertised scope: the FY2025 10-K describes ClearLine as giving buyers 'direct access to premium CTV inventory and integrated curation capabilities', the 2023 launch release as 'premium video inventory', and the curation release of 1 October 2025 as unifying curation and activation 'across the company's premium omnichannel footprint' (Magnite's boilerplate lists CTV, online video, display and audio for the company; the CLA blog speaks of an 'evolving omnichannel market'), so the product's scope is CTV and video with an undefined omnichannel extension. Documented live: CTV, in the 10-K's present-tense description ('For ClearLine, advertisers and agencies use a self-service programmatic solution to directly access premium inventory') and in AdExchanger's report of 14 October 2026 that H/L's buyer agent activated LG Home Screen inventory 'within ClearLine' (the LG/Magnite release of the sam | scope, unit, date or wording narrowed to what the source carries | CLEARLINE capability channel_execution |
| V-clearline-121 | No observed sample of a ClearLine curated deal; Magnite self-describes inventory-authorisation guidelines and real-time non-human-traffic elimination with unnamed technology; Adalytics documents Magnite-transacted PMP deal IDs (e.g. 'MGNI-MD-9-5198', 17 March 2024; 'MGNI-AP-5083', 4 October 2023 and 24 November 2024) serving declared data-centre bots, not ClearLine deals; availability unconfirmed, grade D. | Refute-mode verifier for clearline.json re-opened the cited source | Responsible Advertising page (Copyright 2026): 'Our inventory quality guidelines address inventory authorization, legal compliance'; 'We also utilize a tapestry of industry, proprietary, and third-party technology to identify and eliminate machine, bot, and other non-human generated traffic in real-time'; no ads.txt, sellers.json, SupplyChain, TAG, MRC or HUMAN. Adalytics (not re-opened here; the B verifier read it on 3 November 2026, CUR-B-F29): Magnite 'MGNI-MD-9-5198' (Dove), 'TRD-9262-00da4' (19 November 2024), 'MGNI-AP-5083' (24 November 2024); 'not all bot traffic is fraudulent'. The 17 March 2024 date and a 4 October 2023 instance of 'MGNI-AP-5083' are in no verifier record. The B verifier also confirmed (CUR-B-F27) Adalytics' Forbes report naming Magnite among 'four ad exchanges' that 'appear to incorrectly relay' 'www3.forbes.com' inventory as the 'www.' sub-domain, with 'Some o | supported with qualification | No observed sample of a ClearLine curated deal was found or taken, so authorisation (ads.txt, sellers.json, SupplyChain) and request duplication are unobserved for this product; Magnite self-describes inventory quality guidelines that 'address inventory authorization' and 'a tapestry of industry, proprietary, and third-party technology' to eliminate non-human traffic 'in real-time', naming no measurer. Adalytics' observational reports document Magnite-transacted PMP deal IDs (for example 'MGNI-MD-9-5198' and 'MGNI-AP-5083', the latter on 24 November 2024) delivering ads to declared data-centre bots, and Magnite among four exchanges that relayed mis-declared 'www3.forbes.com' inventory as the 'www.' domain, some of it via PMP deal IDs (2024); none of these deals is identified as a ClearLine deal, all predate ClearLine curation, and the authors note that 'not all bot traffic is fraudulent' | scope, unit, date or wording narrowed to what the source carries | CLEARLINE capability supply_path_integrity |
| V-clearline-122 | Adalytics documents ads served to declared data-centre bots via Magnite-transacted PMP deal IDs 'MGNI-MD-9-5198' (Dove, 17 March 2024), 'TRD-9262-00da4' (Dove, 19 November 2024), 'MGNI-AP-5083' (T-Mobile, 4 October 2023 and 24 November 2024); Magnite among the platforms claiming HUMAN pre-bid filtering; the examples predate ClearLine curation. | Refute-mode verifier for clearline.json re-opened the cited source | Not re-opened (budget); reusing the B verifier's reading of the same page on 3 November 2026 (CUR-B-F29, supported_as_stated): Magnite 'MGNI-MD-9-5198' (Dove, with 'charge-allDoubleVerifyBotAvoidance'), 'TRD-9262-00da4' (19 November 2024), 'MGNI-AP-5083' (24 November 2024); exchanges that 'have declared publicly that they filter bot traffic pre-bid through partnerships with HUMAN Security'; 'not all bot traffic is fraudulent'. No verifier record confirms the 17 March 2024 date of 'MGNI-MD-9-5198' or a 4 October 2023 instance of 'MGNI-AP-5083'. | supported with qualification | Adalytics' pre-bid bot report documents ads served to declared data-centre bots through Magnite-transacted PMP deal IDs, including 'MGNI-MD-9-5198' (Unilever Dove), 'TRD-9262-00da4' (Unilever Dove, 19 November 2024) and 'MGNI-AP-5083' (T-Mobile, 24 November 2024), and notes that Magnite is among the platforms that publicly claim pre-bid bot filtering with HUMAN Security; the examples predate ClearLine curation and none is identified as a ClearLine deal. | scope, unit, date or wording narrowed to what the source carries | CUR-CLEARLINE-F10 |
| V-clearline-123 | Magnite announced its next-generation SpringServe CTV platform, 'the infrastructure ClearLine curation is built on'. | Refute-mode verifier for clearline.json re-opened the cited source | 10-K: 'In April 2025, we announced the introduction of our next generation SpringServe CTV platform.' The curation release says only that 'ClearLine is built on the same underlying infrastructure as Magnite's SpringServe'; it does not name the April 2025 next-generation platform. | supported with qualification | In April 2025 Magnite announced 'the introduction of our next generation SpringServe CTV platform' (FY2025 10-K); the curation release of 1 October 2025 says ClearLine is built on 'the same underlying infrastructure as Magnite's SpringServe' without naming the next-generation platform. | scope, unit, date or wording narrowed to what the source carries | CUR-CLEARLINE timeline 2025-04 |
| V-clearline-124 | CTV advertised yes (10-K) and documented live 'yes (LG Home Screen via ClearLine, fall 2026)' sourced to CUR-CLEARLINE-S09; web video, display, in-app, audio and DOOH rows. | Refute-mode verifier for clearline.json re-opened the cited source | S09 (the LG/Magnite release) does not tie the H/L buy to ClearLine; AdExchanger (S08) does, and the 10-K (S04) describes ClearLine as a live CTV buying product. The 'premium omnichannel footprint' phrase in the other rows is verbatim. [172, 87, 212] | supported with qualification | Replace the CTV row's documented_live source with CUR-CLEARLINE-S08 (AdExchanger: agents 'within ClearLine') and CUR-CLEARLINE-S04 (the 10-K's present-tense description of ClearLine CTV buying); keep S09 only for the H/L campaign facts. | scope, unit, date or wording narrowed to what the source carries | CUR-CLEARLINE data_tables.clearline_channel_scope |
| V-index-marketplaces-125 | Index's knowledge base documents the selection logic a Marketplace Partner configures per package (publisher or deal targeting, domain and app-bundle lists up to 10,000, four include segment rule sets and one exclude, device, Web or In-App, format, geography, dayparting, minimum viewability probability, VCR and CTR (beta), video, streaming-TV content, IAB categories, at least one non-daypart dimension required); no update cadence; buyer view not stated. Live, grade C, authenticated. | Refute-mode verifier for index-marketplaces.json re-opened the cited source | 'Last Updated: October 26, 2026'. Step 13: 'You can target publishers or deals, not both.'; leaving the publisher list blank targets all eligible publishers and, per the page, automatically includes new publishers; deal targeting is labelled 'Targeting a deal in a Marketplace Package (Beta)'. Step 14: 'You can add up to 10,000 domains and app bundles.'; 'If values are added to both the Include and Exclude sections, the Exclude section takes priority.' Step 15: 'When including segments, you can add up to four segment rule sets.'; 'When excluding segments, you can add one segment rule set.' Step 16: device types Connected device (CTV), Connected TV (CTV), Personal computer, Phone, Set-top box (CTV), Tablet; In-App and Web; Banner, Native, Video. Step 18: Countries, States and provinces, Metro codes, 'Zip and postal codes (Beta)' (only zip and postal is labelled beta). Step 19: dayparting, | supported with qualification | Index's knowledge base documents the selection logic a Marketplace Partner configures per package: publisher targeting (a blank list targets all eligible publishers and automatically includes new ones) or deal targeting (beta), domain and app-bundle include and exclude lists of up to 10,000 entries with exclude taking priority, up to four include segment rule sets and one exclude, device type (three CTV device types among six), Web or In-App environment, Banner, Native or Video format, geography by country, state or province, metro code and zip or postal code (beta), dayparting (up to 400 schedules), a minimum viewability probability, minimum VCR and CTR (both beta), video placement types, streaming-TV content genres (networks and channels in beta), IAB Tier 1 and Tier 2 categories and custom key values (beta), with at least one non-daypart dimension required. Thresholds are chosen by th | scope, unit, date or wording narrowed to what the source carries | INDEXMARKETPLACES capability selection_transparency |
| V-index-marketplaces-126 | Advertised scope streaming TV, display, video and mobile app (launch release; streaming TV and CTV on the 2026 product page), mapped to web display, web video, mobile in-app and CTV; all four documented as available in the package builder; audio and DOOH neither advertised nor documented. Grade C. | Refute-mode verifier for index-marketplaces.json re-opened the cited source | S04 (Jan 23, 2024): 'across various channels including streaming TV, display, video, and mobile app'. S02 (read 3 November 2026): 'Streaming TV' in several headings; 'CTV' appears only in a linked release headline ('Bring Exclusive Creator Audiences to CTV and Beyond'); 'Online Video' appears as a product label; display, mobile, audio and DOOH not mentioned. CUR-U-F05 (verified) recorded the product page as advertising streaming TV and online video, with the verifier noting 'online video' is not a phrase on the page. S07: In-App and Web; Banner, Native, Video; three CTV device types; STV content genres. | supported with qualification | Advertised scope: the anchor takes the scope captured in workstream U, the 2026 product page (streaming TV, with 'Online Video' as a product label and 'CTV' only in a linked release headline), while the launch release of 23 January 2024 advertised the wider 'streaming TV, display, video, and mobile app'. The package builder documents Web and In-App environments, three CTV device types, Banner, Native and Video formats and streaming-TV content genres, so every environment in either scope is documented as available in the tool (100% under both); Index's case studies describe display (US, 2024) and streaming-TV executions without environment-level results. Audio and DOOH are neither advertised nor documented. | scope, unit, date or wording narrowed to what the source carries | INDEXMARKETPLACES capability channel_execution |
| V-index-marketplaces-127 | Package-creation page names four DSPs requiring a Buyer Seat ID (DV360, The Trade Desk, Amazon, Xandr); Microsoft said it would stop supporting Microsoft Invest media buying from 28 February 2026, so that route's status is unconfirmed; releases assert 'preferred DSPs' and 'all DSPs'. Grade C. | Refute-mode verifier for index-marketplaces.json re-opened the cited source | S07 Step 7: 'select the corresponding DSP for this package.' Step 8: 'If you work with any of the following DSPs, a Buyer Seat ID is required:' Display & Video 360, The Trade Desk, Amazon, Xandr; 'Index recommends to always select a Buyer Seat ID if available.' S04: Partners keep using their 'preferred DSPs'. Microsoft's post of 14 May 2025 (read by workstream E's verifier, CUR-E-F38, not re-opened here) retires Microsoft Invest media buying from 28 February 2026. | supported with qualification | The package-creation page (26 October 2026) names four DSPs for which 'a Buyer Seat ID is required': Display & Video 360, The Trade Desk, Amazon and Xandr. Microsoft said on 14 May 2025 it would stop supporting Microsoft Invest (Xandr) media buying from 28 February 2026, a documented withdrawal of that route before the cutoff, so three named endpoints (Display & Video 360, The Trade Desk and Amazon) carry documented activation that is current on the record; the knowledge base's continued listing of Xandr is not evidence that the route still operates. The releases assert 'preferred DSPs' (2024) and 'reach all DSPs' (2025) without names, and the full DSP list in the UI is not on the page. | scope, unit, date or wording narrowed to what the source carries | INDEXMARKETPLACES capability activation_routes |
| V-index-marketplaces-128 | US availability documented in the knowledge base (LiveRamp US-only segments; country, state, metro and postal targeting; 20 of 49 active vendors list the USA); US production use only in Index's own case study (SWYM.ai for Butler/Till's 'major US financial services brand', August to November 2024). Grade C. | Refute-mode verifier for index-marketplaces.json re-opened the cited source | S11: 21 of 49 rows list USA alone or among markets: Acxiom, Aeterna Labs, ANOKi AI, Attain, Audience Town, Audigent, Culture Hive Media Group, Digiseg (a 48-market list ending 'UK, USA'), Dstillery, Evertune, Experian Marketing Services, Givsly, IRIS.TV, Inscape by VIZIO, Key, LiveRamp, Plural, Resonate, Stirista, Subjective, True Relevance. S10: segments 'can only be used to target users in the United States'. S07 Step 18: Countries, States and provinces, Metro codes, Zip and postal codes (Beta). S13: 'a major US financial services brand', 'across multiple markets in the US', 'August through November 2024'. | supported with qualification | US availability is documented in the knowledge base: LiveRamp Data Marketplace segments 'can only be used to target users in the United States', geography targeting covers countries, states, metro codes and zip or postal codes (the last in beta), and 21 of the 49 active data vendors list the USA among their markets. US production use appears only in Index's own case study (grade D): SWYM.ai for Butler/Till's client, a 'major US financial services brand', 'across multiple markets in the US', August to November 2024, brand unnamed. | scope, unit, date or wording narrowed to what the source carries | INDEXMARKETPLACES capability us_availability |
| V-index-marketplaces-129 | Exchange-quality and about pages display TAG Brand Safety Certified, Certified Against Malware and Certified Against Fraud seals, 'TAG Platinum' status, an IAB Gold Standard logo and HUMAN, IAS and The Media Trust logos, all company level, undated and unscoped; TAG's registry not opened; no product audit. Grade D. | Refute-mode verifier for index-marketplaces.json re-opened the cited source | S16 logos (alt text): 'TAG Brand Safety Certified', 'TAG Certified Agains Malware' (sic), 'TAG certified against fraud', 'Logo for iab gold standard certified', 'HUMAN', 'IAS logo', 'The Media Trust', 'IAB Tech Lab'; no years and no scope; no IVT measurer or MRC named in the body. S17 history: heading 'Earned TAG Platinum status' under 2019 ('we achieved Platinum status with the Trustworthy Accountability Group'). TAG registry page: the search interface and entries did not render to the fetch tool. | supported with qualification | Index's exchange-quality page displays TAG Brand Safety Certified, TAG Certified Against Malware and TAG certified against fraud seals, an IAB Gold Standard logo and HUMAN, IAS, The Media Trust and IAB Tech Lab logos, undated and without scope, and its about page records 'TAG Platinum' status under 2019; all are company level with no scope covering Index Marketplaces. TAG's registry was opened but its entries did not render, so current seal status is vendor-stated; no audit of a Marketplace package's composition or fees was found. | scope, unit, date or wording narrowed to what the source carries | INDEXMARKETPLACES capability assurance |
| V-index-marketplaces-130 | Announced 23 January 2024 as an omnichannel platform in limited beta with broader rollout later in 2024; product page carries no status word; no GA announcement found, so the beta-to-general transition is undated. | Refute-mode verifier for index-marketplaces.json re-opened the cited source | S04: 'Toronto, Canada', 'Jan 23, 2024'; 'an omnichannel platform aimed at helping partners transact more efficiently'; 'tested with a select group of partners'; 'currently in limited beta testing'; 'broader rollout of Marketplaces later in the year'. S02: no beta, early-access, generally-available or launched wording. S17 history: heading 'Launched Index Marketplaces' under 2024 ('enabling businesses to build atop our platform to scale their solutions for curation, data activation, AI algorithms'). | supported with qualification | Index Exchange announced Marketplaces on 23 January 2024 as 'an omnichannel platform' then 'tested with a select group of partners' and 'currently in limited beta testing', with a broader rollout planned later in 2024; Index's about page lists 'Launched Index Marketplaces' under 2024 without a month or an availability term, the product page read on 3 November 2026 carries no status word, and no general-availability announcement was found, so the move from beta to general availability is dated by Index's own history only to the year 2024. | scope, unit, date or wording narrowed to what the source carries | CUR-INDEXMARKETPLACES-F01 |
| V-index-marketplaces-131 | The package builder (26 October 2026) documents the selection rule a partner sets, from publisher or deal targeting to IAB categories, with VCR and CTR in beta; limitations say VCR, CTR, metro or postal targeting and custom key values are beta. | Refute-mode verifier for index-marketplaces.json re-opened the cited source | Every quoted passage is on the page ('You can target publishers or deals, not both.'; 'You can add up to 10,000 domains and app bundles.'; 'If values are added to both the Include and Exclude sections, the Exclude section takes priority.'; 'you can add up to four segment rule sets'; 'Select the minimum viewability probability percentage that you want to target in the deal.'; 'you must include at least one of the targeting dimensions, other than dayparting.'). Beta labels on the page: deal targeting, zip and postal codes, VCR, CTR, STV networks and channels, custom key values. Metro codes are not labelled beta. | supported with qualification | The package builder (updated 26 October 2026) documents the selection rule a partner sets: publisher targeting (blank for all eligible publishers) or deal targeting (beta), domain and app-bundle include and exclude lists of up to 10,000 entries with exclude taking priority, up to four include segment rule sets and one exclude, Web or In-App environments, device types including three CTV types, Banner, Native and Video formats, geography (zip and postal codes in beta), dayparting, a minimum viewability probability, VCR and CTR (both beta), video placement settings, streaming-TV content (networks and channels in beta), IAB Tier 1 and Tier 2 categories and custom key values (beta), with at least one targeting dimension other than dayparting required. | scope, unit, date or wording narrowed to what the source carries | CUR-INDEXMARKETPLACES-F03 |
| V-index-marketplaces-132 | 49 active Data Vendors on 21 October 2026 (17 at early access); LiveRamp segments US only, with Index reserving the right to update targeting on deals using them; vendor permission required; no per-segment collection, use, consent or legal basis documented. | Refute-mode verifier for index-marketplaces.json re-opened the cited source | S11: 49 rows. S10: LiveRamp Data Marketplace segments 'can only be used to target users in the United States'; Index 'reserves the right to update targeting' for affected deals 'to include the United States and only the United States.'; no provenance, consent or legal-basis fields. S07: vendor permission sentence verbatim. The 17 launch vendors rest on CUR-E-F21. | supported with qualification | The knowledge base listed 49 active Data Vendors on 21 October 2026 (up from seventeen named at the Data Vendor Ecosystem's early access on 7 October 2025), each with data types, channels and markets, and a data-usage-conditions page restricts LiveRamp Data Marketplace segments to US users, with Index reserving the right to update targeting on deals that use them 'to include the United States and only the United States'; a partner must have the vendor's permission to target its segments, and no page opened documents collection method, intended use, consent or legal basis per segment. | scope, unit, date or wording narrowed to what the source carries | CUR-INDEXMARKETPLACES-F12 |
| V-index-marketplaces-133 | US availability documented in the KB (LiveRamp US-only, US-level geography, 20 of 49 vendors listing the USA); US production use only in Index's SWYM.ai case (August to November 2024). | Refute-mode verifier for index-marketplaces.json re-opened the cited source | 21 of 49 vendor rows list USA (the profile's list plus Digiseg, whose availability list ends 'UK, USA'); 'to include the United States and only the United States.'; 'a major US financial services brand'; 'across multiple markets in the US'; 'August through November 2024'; title verbatim. | supported with qualification | US availability of Index Marketplaces is documented in its knowledge base (US-only LiveRamp segment conditions; country, state, metro and zip or postal-code targeting; 21 of 49 active data vendors listing the USA), and US production use appears only in Index's own case study: SWYM.ai for Butler/Till's client, a 'major US financial services brand', 'across multiple markets in the US', August to November 2024. | scope, unit, date or wording narrowed to what the source carries | CUR-INDEXMARKETPLACES-F13 |
| V-index-marketplaces-134 | Every performance figure Index publishes for Marketplaces is its own claim without comparator, measurer or method (56%, 26%, 41%, 14%, 9x, 25% GPID); none measures a named quality dimension; the PMG passage does not name Index Marketplaces. | Refute-mode verifier for index-marketplaces.json re-opened the cited source | The listed figures and titles are verbatim. The PMG article body says 'Using Index Marketplaces, PMG activated IRIS-enabled contextual and emotional data directly within Index's premium streaming supply' and 'Through Index Marketplaces and sell-side decisioning, the team was able to simplify activation across fragmented supply'. A fourth Index Marketplaces case, dentsu X and Dos Equis (27 April 2026), reports '84% Of impressions aired during live College Football Playoff games (versus shoulder content)', '7% Lower CPMs than benchmark' and '0% Of impressions aired outside of college football content', with no defined benchmark, measurer or method. | supported with qualification | Every performance figure found for Index Marketplaces is Index's own case-study or release claim without a named measurer or method: conversion rate up 56% and cost per conversion down 26% for a US financial-services brand (August to November 2024), cost per booking down 41% and bookings up 14% for a hotel brand in Europe over two months, 9x brand preference on streaming TV for an outdoor lifestyle brand (a case whose body names Index Marketplaces), 84% of impressions in live College Football Playoff games, CPMs 7% below an undefined benchmark and 0% of impressions outside college-football content for dentsu X and Dos Equis (27 April 2026), and a 25% GPID accuracy improvement; none measures MFA share, viewability, invalid traffic, suitability or attention for a Marketplace deal with a stated method. | scope, unit, date or wording narrowed to what the source carries | CUR-INDEXMARKETPLACES-F15 |
| V-index-marketplaces-135 | dentsu, Chalice AI and a hotel brand in Europe over two months: Chalice's model in Index Cloud judged each impression using 'more than 35 page signals' and supplied a targeting key to Index Marketplaces, which created the deal ID; the exchange-quality page names Chalice's CurateAI 'through Index Marketplaces'. | Refute-mode verifier for index-marketplaces.json re-opened the cited source | All three quotations verbatim (Richardson, Turner, Thankamony). 'the two-month campaign in Europe'; 'The model scored every ad opportunity across more than 35 signals'; 'Chalice supplied one of those keys: a real-time, campaign-specific decision'; 'Within the Marketplaces platform, deals are curated and built against targeting keys'; 'when an impression matches, Marketplaces creates the Deal ID that makes it buyable.' S16 teaser (an article card in the page's 'Learn more' section): 'dentsu activated Chalice's CurateAI technology through Index Marketplaces.' S17: '2026: Launched Index Cloud in closed beta'. | supported with qualification | Index's 6 October 2026 case describes a partner model in production inside Index infrastructure: for dentsu and a hotel brand in Europe over two months, Chalice AI ran a custom decisioning model in Index Cloud (launched in closed beta in 2026, per Index's about page) that scored every ad opportunity 'across more than 35 signals' and supplied a targeting key to Index Marketplaces, which created the deal ID that made matching impressions buyable; an article card on Index's exchange-quality page names it Chalice's CurateAI 'through Index Marketplaces'. | scope, unit, date or wording narrowed to what the source carries | CUR-INDEXMARKETPLACES-F16 |
| V-index-marketplaces-136 | Assurance is company-level and self-presented: TAG seals with HUMAN, IAS and The Media Trust logos, 'every app and site' screening, MFA prohibition, 'TAG Platinum' on the about page, none dated or scoped to Index Marketplaces. | Refute-mode verifier for index-marketplaces.json re-opened the cited source | S16: 'we screen and regularly audit every app and site on the exchange'; 'We prohibit made-for-advertising (MFA) sites.'; logos 'TAG Brand Safety Certified', 'TAG certified against fraud', 'TAG Certified Agains Malware' (sic), IAB Gold Standard, HUMAN, IAS, The Media Trust, IAB Tech Lab; no years. S17: 'Earned TAG Platinum status' under 2019. | supported with qualification | Index's assurance record is company-level and self-presented: its exchange-quality page shows TAG Brand Safety Certified, TAG Certified Against Malware and TAG certified against fraud seals with HUMAN, IAS, The Media Trust and IAB Tech Lab logos and claims it screens and audits 'every app and site on the exchange' and prohibits MFA sites, and its about page records 'TAG Platinum' status under 2019; the seals are undated and none is scoped to Index Marketplaces. | scope, unit, date or wording narrowed to what the source carries | CUR-INDEXMARKETPLACES-F18 |
| V-index-marketplaces-137 | Post-purchase control sits with partner, media owner and Index, not the buyer: partner fee-disclosure choice, custom fee lock-in, media-owner opt-out, and Index's right to update targeting on LiveRamp-segment deals; no buyer notice, audit, export or remedy documented. | Refute-mode verifier for index-marketplaces.json re-opened the cited source | S07: 'If you choose this option, you can't change it to your default fee later.' S10: Index 'reserves the right to update targeting' for deals using LiveRamp Data Marketplace segments 'to include the United States and only the United States.' S12: 'You can opt out of Marketplaces in the Index UI.' S01 passage per CUR-E-F20. | supported with qualification | Post-purchase control on Index Marketplaces sits with the partner, the media owner and Index, not the buyer, on the pages opened: the partner chooses the fee disclosure model and a custom deal fee cannot later revert to the default, media owners can opt out of Marketplaces demand, and Index 'reserves the right to update targeting' on deals that use LiveRamp Data Marketplace segments to restrict them to the United States (a right scoped to enforcing that vendor's US-only condition); no page documents notice to the buyer, an audit right, data export or a remedy for misdescribed inventory or charges. | scope, unit, date or wording narrowed to what the source carries | CUR-INDEXMARKETPLACES-F19 |
| V-index-marketplaces-138 | 9x brand preference, 10x favourability, 3x consideration for an outdoor lifestyle brand on streaming TV; limits: the article does not name Index Marketplaces or a deal ID in the passage captured ('within Index's premium streaming supply'). | Refute-mode verifier for index-marketplaces.json re-opened the cited source | Figures verbatim: '9X Higher brand preference', '10X Higher favorability', '3X Higher consideration'; 'an outdoor lifestyle brand'. The article body names the product as the route: 'Using Index Marketplaces, PMG activated IRIS-enabled contextual and emotional data directly within Index's premium streaming supply' (Solution section) and 'Through Index Marketplaces and sell-side decisioning, the team was able to simplify activation across fragmented supply' (Results section); an IRIS.TV pull quote says 'within Index Marketplaces'; the close reads 'Explore Index Marketplaces today.' No deal ID, baseline ('exceeded benchmarks' undefined), lift vendor, method, period or geography. | contradicted | PMG, IRIS.TV and Index on streaming TV for an outdoor lifestyle brand, activated 'Using Index Marketplaces' (article body, 12 November 2025): 9x brand preference, 10x favourability and 3x consideration, with no baseline, period, geography, method or named lift vendor and no deal ID. | the source does not carry the claim as first written | INDEXMARKETPLACES performance_evidence 3 (PMG, IRIS.TV) |
| V-index-marketplaces-139 | Index sues Google in E.D. Va. (1:25-cv-02003) on 10 November 2025, voluntarily dismissed 17 December 2025 (terminated 18 December), and files Index Exchange Inc. v. Google LLC in S.D.N.Y. (1:25-cv-10477) on 17 December 2025, 'coordinated in the MDL before Judge Castel'. | Refute-mode verifier for index-marketplaces.json re-opened the cited source | Not opened here (the profile records an access error; no workaround attempted). Workstream I's research record of the RECAP card reads 'Index Exchange Inc. v. Google LLC (S.D.N.Y. 2025) Docket Number: 1:25-cv-10477 Date Filed: December 17th, 2025 ... Assigned To: P. Kevin Castel'; I's verifier (CUR-I-F14) confirmed the E.D. Va. case 1:25-cv-02003 filed 10 Nov 2025, 'NOTICE of Voluntary Dismissal by Index Exchange Inc.' filed 17 Dec 2025, terminated 18 Dec 2025, and that the S.D.N.Y. action is a new filing. | supported with qualification | Index Exchange sues Google in E.D. Va. (1:25-cv-02003) on 10 November 2025, voluntarily dismissed by notice of 17 December 2025 (terminated 18 December 2025), and files Index Exchange Inc. v. Google LLC in S.D.N.Y. (1:25-cv-10477) on 17 December 2025, assigned to Judge P. Kevin Castel, who presides over the ad-tech MDL (dates as corrected by workstream I's verifier, CUR-I-F14; the docket itself was not opened). | scope, unit, date or wording narrowed to what the source carries | INDEXMARKETPLACES timeline litigation row (2025-11-10) |
| V-index-marketplaces-140 | Summary: 'every performance figure is Index's own case-study claim without comparator or method'. who_buys: 'Index's case studies name Butler/Till, PMG and dentsu as agencies; the brands are unnamed.' | Refute-mode verifier for index-marketplaces.json re-opened the cited source | The Index Marketplaces product page links 'A First-to-Market Win: dentsu X and Dos Equis Tap Gracenote To Reach College Football Viewers' (27 April 2026): 'Dentsu X worked with Index Exchange to activate Gracenote's first-to-market content-level metadata across premium streaming supply via Index Marketplaces.'; brand Dos Equis named; DIRECTV named as a media owner; results stated against shoulder content and an undefined benchmark. | supported with qualification | who_buys: Index's case studies name Butler/Till, PMG, dentsu and dentsu X as agencies; the brands are unnamed except Dos Equis (dentsu X, 27 April 2026). Summary: every performance figure is Index's own case-study claim without a named measurer or method, and the only comparators stated are undefined benchmarks. | scope, unit, date or wording narrowed to what the source carries | INDEXMARKETPLACES summary and who_buys |
| V-index-marketplaces-141 | Two first-price rows ($5.00, 10%, $5.56, $4.00, $5.56; $2.00, 15%, $2.35, $5.00, $5.00); note: gross-up formula publisher floor / (1 - fee percentage); S08 reliability note 'two worked floor examples'. | Refute-mode verifier for index-marketplaces.json re-opened the cited source | Both rows match. The page shows six first-price and six fixed-price examples: percentage fees gross up as floor / (1 - fee %); fixed-rate CPM fees are added ($5.00 + $1.50 = $6.50); vendor fees are added before the gross-up (($5.00 + $1.00) / (1 - 10%) = $6.67; ($2.00 + $1.00) / (1 - 15%) = $3.53). | supported with qualification | Notes: for a percentage fee the page grosses up as publisher floor / (1 - fee percentage); a fixed-rate CPM fee is added to the publisher floor without gross-up, and a vendor fee is added before the gross-up; the page carries six first-price and six fixed-price examples, of which the table reproduces two. | scope, unit, date or wording narrowed to what the source carries | INDEXMARKETPLACES data_table index_marketplaces_floor_worked_examples |
| V-index-marketplaces-142 | 20 active Data Vendors list the USA among their markets. | Refute-mode verifier for index-marketplaces.json re-opened the cited source | 21 rows list USA, including Digiseg; 49 rows in total. | supported with qualification | 2026-10-21: 21 active Data Vendors list the USA among their markets (derived; see F13). | scope, unit, date or wording narrowed to what the source carries | INDEXMARKETPLACES data_table index_marketplaces_data_vendor_counts (row 2026-10-21 USA) |
| V-index-marketplaces-143 | Six rows: every quantified performance claim found for Index Marketplaces. | Refute-mode verifier for index-marketplaces.json re-opened the cited source | The six rows match their sources; three quantified claims from the dentsu X and Dos Equis case are missing: 84% of impressions in live College Football Playoff games (versus shoulder content), 7% lower CPMs than benchmark, 0% of impressions outside college-football content (comparator: shoulder content or undefined benchmark; measurer none named; geography not stated; period not stated; published 2026-04-27). | supported with qualification | Add three rows from the dentsu X and Dos Equis case (27 April 2026): 84% of impressions in live College Football Playoff games (comparator: shoulder content); CPMs 7% below an undefined benchmark; 0% of impressions outside college-football content; measurer none named; geography and period not stated. | scope, unit, date or wording narrowed to what the source carries | INDEXMARKETPLACES data_table index_marketplaces_performance_claims |
| V-index-marketplaces-144 | Xandr route 'unconfirmed after 28 February 2026'; risk 3: Index reserves the right to update targeting on LiveRamp-segment deals, a post-purchase change right with no documented buyer notification. | Refute-mode verifier for index-marketplaces.json re-opened the cited source | S07 still lists Xandr (26 October 2026); Microsoft's notice (CUR-E-F38) withdraws Microsoft Invest media buying from 28 February 2026. S10's change right is 'to include the United States and only the United States'. | supported with qualification | Xandr row status: 'listed in the knowledge base; route withdrawn by Microsoft from 28 February 2026 (CUR-E-F38)'. Risk 3: Index reserves the right to change targeting on deals using LiveRamp Data Marketplace segments to restrict them to the United States, a narrow exchange-held change right with no documented buyer notification. | scope, unit, date or wording narrowed to what the source carries | INDEXMARKETPLACES data_table index_marketplaces_activation_routes (Xandr row) and risk 3 (LiveRamp change right) |
| V-lotame-145 | Lotame's Services Privacy Notice documents company-wide collection through pixels, tags, cookies and SDKs from 'Business customers, data partners, and graphing partners', licensed TV viewership data for CTV, consent as EU legal basis with TCF vendor ID 95, US-only sensitive behaviour categories, and that Lotame 'is unable to capture any UOOM signal'. | Refute-mode verifier for lotame.json re-opened the cited source | Identifier table sources: Cookie ID (Lotame PID) 'Lotame'; Lotame Panorama ID 'Lotame'; Device ID 'Business customers, data partners, and graphing partners'; IP Address, Third Party ID and Email Address 'Business customers, data partners'. Methods: 'such as pixels and tags to collect information from you'; 'we use cookies in conjunction with our tags and pixels'; 'embed our software development kit into their software'; 'via our secure application programming interfaces and secure file transfer mechanisms'; 'These cookies contain our Lotame PID and Lotame Panorama ID.' 'Lotame licenses television viewership data from third parties that offer smart TVs and video streaming devices.' GDPR supplement: 'Lotame’s legal basis for processing your Personal Data is consent, which is given by you to our business customers and data partners'; 'Lotame’s identification number within the TCF is 95.' Be | supported with qualification | Lotame's Services Privacy Notice (last modified 19 November 2025) documents company-wide, not per curated deal, that Lotame collects data through pixels, tags, cookies (Lotame PID, Panorama ID), SDKs, APIs and file transfers, with identifiers sourced from Lotame itself (cookie IDs), from 'Business customers, data partners' (IP address, third-party ID, email) and, for device IDs, also from 'graphing partners'; licenses television viewership data from smart-TV and streaming-device companies for CTV targeting; states in its GDPR supplement that its legal basis is consent given to its business customers and data partners, with TCF vendor ID 95; lists 'Racial and ethnic data (U.S. only)' and 'Political, religious, and philosophical beliefs (U.S. only)' among behaviour categories; and says Lotame 'is unable to capture any UOOM signal'. | scope, unit, date or wording narrowed to what the source carries | CUR-LOTAME-F09 |
| V-lotame-146 | The only record naming buying endpoints for Lotame's curated deals is Google's 4 November 2024 Ad Manager curation launch (Lotame among eight partners; packages 'currently available for Preferred Deals and Private Auctions via Display & Video 360 and Yahoo'; 'Google Ad Manager will handle billing'); the help page lets the curator enter its fee 'as either percentage of media or CPM' and mark fees 'visible or hidden from buyer', without saying who pays. | Refute-mode verifier for lotame.json re-opened the cited source | Blog (per verifier F): 'Curation already supports Audigent, IAS, Liveramp, Lotame, Multilocal, Permutive, PrimeAudience and Scope3.'; 'currently available for Preferred Deals and Private Auctions via Display & Video 360 and Yahoo' 'with support for Programmatic Guaranteed and other DSPs coming in the months ahead.'; 'Google Ad Manager will handle billing'. Help page: title 'Curation for data and curation partners (Beta) - Authorized Buyers Help'; 'This feature is in Beta'; 'Curators can bundle Private Auctions and Preferred Deals into curated packages.'; 'Enter the "Fee" as either percentage of media or CPM.'; '(Optional) Mark your package fees as visible or hidden from buyer.'; names no DSP and no curation partner; no payer, billing, composition-visibility or reporting statement; '©2026 Google'. | supported with qualification | The only record naming buying endpoints that could carry Lotame's packages is Google's 4 November 2024 Ad Manager curation launch, which lists Lotame among eight supported curation partners whose packages, as a group, were 'currently available for Preferred Deals and Private Auctions via Display & Video 360 and Yahoo', 'with support for Programmatic Guaranteed and other DSPs coming in the months ahead', and says 'Google Ad Manager will handle billing'; Google's current curation help page (Beta, ©2026) names no DSP, lets the curator enter its fee 'as either percentage of media or CPM' and optionally mark package fees 'visible or hidden from buyer', without saying who pays. | scope, unit, date or wording narrowed to what the source carries | CUR-LOTAME-F10 |
| V-lotame-147 | Every quality or performance claim found for Lotame's curated product (three early-trial figures, the PHD view-through and projected-saving figures, and the unmeasured 'No MFAs. No waste.', match-rate and 'carbon efficient' promises) is a vendor assertion without a published method; no independent or commissioned measurement against a pre-curation comparator was found. Derivation: 8 claims, 7 vendor assertions and 1 forecast. | Refute-mode verifier for lotame.json re-opened the cited source | All quoted figures verbatim. The same sources carry further unmeasured quality claims the count omits: capability page 'bespoke audiences matched to trusted, brand-safe media supply', heading 'Refined brand-safe inventory at scale', 'safe, scalable, high-performing ad placements'; PDF page 3 'Continuous supply-led optimizations drove campaign lift' and 'resulting in performance that surpassed established benchmarks'; PDF page 2 'brand-safe environments'. None has a method, measurer or period. | supported with qualification | Every quality or performance claim found for Lotame's curated product (three early-trial figures; the PHD view-through, projected-saving, 'campaign lift' and 'surpassed established benchmarks' statements; and the unmeasured 'No MFAs. No waste.', 'brand-safe' supply, match-rate and 'carbon efficient' promises) is a vendor assertion or vendor forecast without a published method, and this review found no independent or commissioned measurement of a Lotame curated deal against a pre-curation comparator in the sources screened by 3 November 2026. | scope, unit, date or wording narrowed to what the source carries | CUR-LOTAME-F13 |
| V-lotame-148 | No seller list, domain list, seller count, categories or authenticated composition interface documented; the only composition language is the PHD case's 'transparently optimized based on performance'. live, D, public, 2026. | Refute-mode verifier for lotame.json re-opened the cited source | Capability page also asserts transparency: 'Brands gain greater audience reach and scale with control and transparency'. PDF page 3: the 'Better Control & Transparency' passage. Google's help page does not address buyer visibility of package composition. No seller, domain or SSP list anywhere. | supported with qualification | No seller list, domain list, seller count, seller categories or authenticated composition interface is documented for any Lotame curated marketplace; the page names no SSP and the PHD case names neither the SSPs ('multiple supply-side platforms (SSPs)') nor the publishers. The only composition language is general assertion: the capability page says 'Brands gain greater audience reach and scale with control and transparency', and the case's 'Better Control & Transparency' section says the marketplaces were customized using the agency's allow lists and 'transparently optimized based on performance'. Google's curation help page does not address whether buyers can see a package's composition. Searched: the capability page, the launch release, the case page and PDF, Google's launch post and curation help page, and the Onetag, Index and Magnite records, on 3 November 2026. | scope, unit, date or wording narrowed to what the source carries | LOTAME capability composition_disclosure |
| V-lotame-149 | Product records name a source type only; one case segment has its collection route and 'No modeling'; the Services Privacy Notice documents company-wide methods, source categories ('Business customers, data partners, and graphing partners'), licensed TV data and the legal basis (consent; TCF 95); no provider, intended use or permission per deal or segment. live, C, public, 2025-11-19. | Refute-mode verifier for lotame.json re-opened the cited source | 'cookieless publisher data' is on the release; 'AI-powered consumer insights' on the capability page ('Lotame's earned expertise in custom audience creation combined with AI-powered consumer insights'); 'market-level consumer data and unique audience insights from its Analytics suite' on the PDF's page 2 only (absent from the page and the release). Notice identifier sources and legal basis as in the F09 check; the consent basis is stated in the GDPR supplement. | supported with qualification | Product records name source types, with Lotame as the provider of its own inputs ('cookieless publisher data', release; 'AI-powered consumer insights', capability page; 'market-level consumer data and unique audience insights from its Analytics suite', PHD case); one case segment is defined with its collection route ('Data is collected via loyalty / credit cards and online actions across publisher sites'), with the loyalty and card data provider unnamed, and 'No modeling was done'. The Services Privacy Notice (vendor documentation, last modified 19 November 2025) documents, company-wide rather than per segment, the collection methods (pixels, tags, cookies, SDKs, APIs, file transfers), the identifier sources (Lotame for cookie IDs; 'Business customers, data partners' for IP address, third-party ID and email; 'graphing partners' in addition for device IDs), licensed TV viewership data for | scope, unit, date or wording narrowed to what the source carries | LOTAME capability data_provenance |
| V-lotame-150 | Every quality claim is a vendor assertion without qualifying measurement (early-trial 20%, 15%, 21%; PHD view-through above 70% against an unsourced 60%; projected savings up to 33%; 'No MFAs. No waste.'; match rates). No study with method and comparator, and no independent study, found. live, D, public, 2026. | Refute-mode verifier for lotame.json re-opened the cited source | All quoted claims verbatim. The case reports a measured outcome against a stated (unsourced) comparator, and the release reports 'early trials' results, both without method; the page adds unmeasured 'brand-safe' suitability claims ('trusted, brand-safe media supply', 'Refined brand-safe inventory at scale'). | supported with qualification | Every quality claim is a vendor assertion without published method: early-trial results of '20% better campaign efficiencies, improved video completion rates by 15%, and 21% higher viewability' (no comparator, sample, period, geography or measurer); a PHD case 'View-Through Rate consistently above 70%, beating industry benchmarks of 60%' (benchmark unsourced, period and measurer unstated), 'projected cost savings of up to 33%' and 'performance that surpassed established benchmarks'; and the unmeasured promises 'No MFAs. No waste.', 'trusted, brand-safe media supply' and 'Higher-than-your-DSP match rates'. The trial and case figures are vendor-reported study outcomes without method (the case with an unsourced comparator), which matches anchor 2 ('a vendor or commissioned study without a comparator or without method'); the page promises are claims without qualifying measurement (anchor 1). | scope, unit, date or wording narrowed to what the source carries | LOTAME capability quality_evidence |
| V-lotame-151 | Lotame's page names no endpoint (n/e); Google's 4 November 2024 post names DV360 and Yahoo for curation partners' packages on Ad Manager Curation (Beta); Onetag names no DSP; Index and Magnite describe Lotame as a data supplier. beta, D, public, 2026. | Refute-mode verifier for lotame.json re-opened the cited source | Capability page: 'available to DSPs' only. Google blog (per verifier F): DV360 and Yahoo at launch, 'with support for Programmatic Guaranteed and other DSPs coming in the months ahead'. Help page (Beta, ©2026): names no DSP. No Lotame page mentions Google Ad Manager. | supported with qualification | Lotame's own page names no buying endpoint ('available to DSPs'), which is n/e by the anchor. Two endpoints are named only by a partner platform, for curation partners as a group and at launch: Google's 4 November 2024 post lists Lotame among the supported curation partners whose packages were 'currently available for Preferred Deals and Private Auctions via Display & Video 360 and Yahoo', 'with support for Programmatic Guaranteed and other DSPs coming in the months ahead', on Ad Manager Curation (Beta). Google's current help page (©2026) names no DSP and no Lotame page mentions Google Ad Manager, so neither route is documented for a Lotame deal or confirmed after November 2024. Onetag's release names no DSP ('for DSPs'), and the Index and Magnite records describe Lotame as a data supplier to those platforms' tools, not as a route for Lotame-built deals. | scope, unit, date or wording narrowed to what the source carries | LOTAME capability activation_routes |
| V-lotame-152 | No Lotame charge disclosed with amount, rate, base, payer, collector or period; only the 2024 'data/tech fees' assertion; on Ad Manager the curator's fee field and visibility setting exist but Lotame's setting is undocumented; 'no Deal Sync API curfeetype declaration for Lotame was found'. live, D, unknown, 2026. | Refute-mode verifier for lotame.json re-opened the cited source | 'Cuts down operational workload and data/tech fees'; capability page carries no price; help page 'Enter the "Fee" as either percentage of media or CPM.' and '(Optional) Mark your package fees as visible or hidden from buyer.' The profile's searches list IAB Tech Lab Deal Sync API adoption lists as not screened. | supported with qualification | No charge Lotame controls or collects is disclosed with an amount, rate, base, payer, collector or period; the only pricing statement is the 2024 general assertion that the offering 'Cuts down operational workload and data/tech fees', and the current page carries no price. On the Google Ad Manager route the curator's fee is entered 'as either percentage of media or CPM' and can optionally be marked 'visible or hidden from buyer', but Lotame's setting is not documented, and no Deal Sync API curfeetype declaration for Lotame was searched for (adoption lists not screened). Third-party charges (SSP and DSP take rates; Index's statement that it 'does not charge data fees') are attributes, not Lotame's disclosure. | scope, unit, date or wording narrowed to what the source carries | LOTAME capability pricing_transparency |
| V-lotame-153 | Undated security summary says the ISMS is 'certified to meet the ISO 27001 standard', covering the Spherical platform and corporate systems; the Schellman entry did not render; company certification does not cover the product; the CDAP review is not an independent audit; no TAG seal or product audit found. unconfirmed, C, public, 2026. | Refute-mode verifier for lotame.json re-opened the cited source | The summary does not say the certification covers 'corporate systems'; it says risk analysis extends 'not only on the Spherical platform, but on all of the data assets that we utilize to service our clients' data needs' and gives no certified scope. Footer names Lotame Solutions, Inc.; undated. Schellman entry did not render (access error). CDAP passage verbatim on PDF page 1. | supported with qualification | Lotame's undated security summary (Public V2.1.1, in the name of Lotame Solutions, Inc.) says it implements an Information Security Management System 'certified to meet the ISO 27001 standard'; it names no certifier, certificate number, scope statement or validity dates, and says only that its risk analysis extends 'not only on the Spherical platform, but on all of the data assets that we utilize to service our clients' data needs'. The certification announcement URL now redirects to Schellman's certificate directory entry 1571242-15, which did not render (access error), so the certifier's statement of holder, scope, standard version and validity was not read, and whether a certificate held by Lotame Solutions, Inc. continued after that entity's merger into Epsilon Data Management, LLC on 1 October 2025 is undocumented. Information-security certification of the operating company does not | scope, unit, date or wording narrowed to what the source carries | LOTAME capability assurance |
| V-lotame-154 | Summary: 'Google's 4 November 2024 Ad Manager curation launch lists Lotame among eight partners whose packages ran through Display & Video 360 and Yahoo'. | Refute-mode verifier for lotame.json re-opened the cited source | The post says partners' packages were 'currently available for Preferred Deals and Private Auctions via Display & Video 360 and Yahoo'; it reports availability, not that any Lotame package ran. | supported with qualification | Replace 'lists Lotame among eight partners whose packages ran through Display & Video 360 and Yahoo' with 'lists Lotame among eight supported curation partners whose packages were "currently available" via Display & Video 360 and Yahoo at launch'. | scope, unit, date or wording narrowed to what the source carries | LOTAME summary |
| V-lotame-155 | Exchange and platform records that name Lotame: Google Ad Manager Curation, Onetag, Index Marketplaces, Magnite ClearLine Activate; only Google names Lotame as a curator. | Refute-mode verifier for lotame.json re-opened the cited source | PubMatic's Connect for Buyers page (©2026) shows a Lotame Panorama ID campaign with Xaxis, Xandr and PubMatic ('Source: Lotame Internal Data, 4/11/22 – 5/9/22'; 94% video completion, 83% viewability in Safari/Firefox against a '34%+ viewability benchmark'), as recorded by the PubMatic Connect profile. [246, 246] | supported with qualification | Add a row: PubMatic Connect for Buyers page (CUR-PUBMATICCONNECT-S01): Lotame as identity and data source in a 2022 Panorama ID campaign with Xaxis, Xandr and PubMatic; no curated-deal role; predates Lotame Curation (August 2024); figures more than 24 months old at the cutoff. The note 'Only the Google record names Lotame as a curator' stands. | scope, unit, date or wording narrowed to what the source carries | LOTAME data_table lotame_platform_records |
| V-microsoft-curate-156 | Vendor documentation describes the curator's selection toolset (Microsoft universal categories applied from inventory review, seller categories, seller and exposed-publisher include or exclude lists, inventory type App & Web by default, allowlists and blocklists with the AppNexus Blocklist always applied, buy-side targeting with segments and geography, viewability or completion-rate thresholds as a hard limit on predicted rates); prediction models, category review cadence and deal update cadence not documented; ads.txt not supported. Live, grade C, public, 2026-02-18. | Refute-mode verifier for microsoft-curate.json re-opened the cited source | Create page (ms.date 2026-02-18, updated_at 2026-02-18T04:56Z, git 18c9ba0a, 2,410 words): 'Universal categories are defined by Microsoft Advertising. When Microsoft Advertising reviews inventory, we apply these categories based on the inventory's content.'; 'Seller Categories are defined by sellers.'; 'By default, your curated line item will target all sellers. However, you can narrow targeting to include or exclude specific sellers.'; 'By default, your line items will target all inventory types (i.e., App & Web)'; 'Select one or more blocklists (by default, the AppNexus Blocklist will always be applied).'; 'Ads.txt Do not use (not supported for curated deal line items).'; 'Thresholds (Viewability Threshold or Completion Rate Threshold) create a hard limit for curated deal line item targeting. Any impressions with a predicted view or completion rate below the threshold won't be matched | supported with qualification | Vendor documentation describes the curator's selection toolset in operational terms: Microsoft-defined universal categories applied 'based on the inventory's content' when Microsoft reviews inventory, seller-defined categories, include or exclude lists of sellers and exposed publishers, inventory type (App & Web by default, App Only, Web Only), allowlists and blocklists (the AppNexus Blocklist always applied), buy-side targeting with segments and geography, and curator-set viewability or completion-rate thresholds that act as a 'hard limit' on Microsoft's predicted rates. Since April 2026 Microsoft also offers Max Reach, which expands a curator's first-party segments 'to similar users' and adds 'contextual expansion', with no method documented. On the pages opened (the linked Buy-Side Targeting and Inventory Performance Targeting pages were not opened), Microsoft's prediction and expansi | scope, unit, date or wording narrowed to what the source carries | MICROSOFTCURATE capability selection_transparency |
| V-microsoft-curate-157 | Data inputs named by type and owner but not per segment: curator audience data and segments, Microsoft universal categories (non-personal contextual), seller categories, 'Cross-device enhanced by Microsoft' on automatically and at no cost when a segment is targeted, opt-out only through an account representative; no segment source, method, permissions, consent route or legal basis, nor the cross-device graph's source. Live, grade C, public. | Refute-mode verifier for microsoft-curate.json re-opened the cited source | Create page: 'Cross-device enhanced by Microsoft will be automatically enabled for your curated deal at no cost if you target a segment. If you prefer not to benefit from cross-device on your curated deal, please contact your account representative to opt out of this feature.' Universal and seller category passages as quoted in the selection entry. No segment-level source, method, permission or legal basis on the page. Q1 2026 roundup (14 April 2026): 'Curators can now use the Max Reach option to expand their first-party segments to similar users for targeting.' and 'Max Reach introduces contextual expansion, allowing curators to extend beyond addressable user IDs.', with no model, data source or consent statement. | supported with qualification | Data inputs are named by type and owner, but not per segment. Curators bring 'audience data' and target segments. Microsoft applies universal categories from its review of inventory content, a non-personal contextual input with Microsoft as provider. Sellers define seller categories. 'Cross-device enhanced by Microsoft' is enabled 'automatically' and 'at no cost' whenever a segment is targeted, and the documented opt-out route is the account representative. Since April 2026 Microsoft's Max Reach can also expand a curator's first-party segments 'to similar users' and add 'contextual expansion', a Microsoft-modelled input whose method and data source are not described. No page opened gives a segment's source, collection method, intended use, permissions, consent route or legal basis, or the source of Microsoft's cross-device graph or Max Reach model. In anchor terms the record names source | scope, unit, date or wording narrowed to what the source carries | MICROSOFTCURATE capability data_provenance |
| V-microsoft-curate-158 | Auction type always Open Auction for curators; Dynamic CPM competes in first or second price depending on the seller; no priority by default, not guaranteed, seller floors and ad quality never overridden, bids compete net of fees unless agreed otherwise; floors Fixed, Standard or Market, above, at or below a seller floor, with second-price clearing below a fixed price; exclusivity one buyer per line item, fixed after saving, supply still biddable on the open exchange; Microsoft-controlled Curation Optimization sampling with no threshold, rate or opt-out. Live, grade C, public. | Refute-mode verifier for microsoft-curate.json re-opened the cited source | Create page: 'Auction Type is always set to "Open Auction" for curators.'; 'Dynamic CPM ("standard auction") - The curated deal line item competes in either a first or second-price auction at the price bid by the buyer, whenever the buyer bids above the ask price.'; 'You cannot change the selected buyer after the curated deal line item is created and saved.' Curated Deals page: 'Curated transactions are not guaranteed deals and will not override any seller account configurations, including ad quality settings or floor price settings.'; bids 'submitted to your final auction net of such fees or charges unless you agree otherwise with the applicable curator'; 'With curation, buyers can still bid on seller inventory via open exchange'. Floors page (ms.date 2025-10-21): Fixed, Standard and Market types; 'The ability to set curator floor prices above, at, or even below a seller floor gives cur | supported with qualification | Each mechanic is documented with its default. Auction type: 'always set to "Open Auction" for curators'; a Dynamic CPM deal 'competes in either a first or second-price auction at the price bid by the buyer'. Priority: none by default; curated transactions 'are not guaranteed deals', never override seller floor or ad-quality settings, and compete 'in the seller's final auction with all other deal and open exchange bids, net of all applicable fees and charges', net-of-fee submission being a default 'unless you agree otherwise with the applicable curator'. Floors: Fixed, Standard or Market, set 'above, at, or even below a seller floor'; a fixed-price deal on a second-price seller may clear below the configured price. Exclusivity: one buyer selection per line item, fixed after saving; for DV360 buyers that selection can be a Central Partner ID, which makes one deal 'automatically eligible fo | scope, unit, date or wording narrowed to what the source carries | MICROSOFTCURATE capability auction_mechanics |
| V-microsoft-curate-159 | The documentation names one buying route, Microsoft Invest, plus 'all major DSPs'; the product page and the 2026 post name no DSP; Microsoft said on 14 May 2025 it would no longer support media buying through Invest from 28 February 2026; so the only named route has been withdrawn and no third-party DSP is named with documented activation functionality on any Microsoft page opened. Unconfirmed, grade C, public, 2026-09-10. | Refute-mode verifier for microsoft-curate.json re-opened the cited source | The passages the profile quotes are verbatim: Curated Deals page 'Curated deals can be bought via Microsoft's Invest and all major DSPs.'; May 2025 post 'we will no longer support media buying through our DSP, Microsoft Invest, starting on February 28, 2026.' and 'we will continue to support access to Microsoft and partner inventory through third-party DSPs' ('Amazon' does not appear in that post); product page 'available via all major demand-side platforms (DSPs).'; 2026 post 'access to buyers through integrations with major third-party DSP partners'. But Microsoft Curate's own release notes, which the profile listed as unopened, document a named third-party DSP route. Release Notes index (ms.date 2026-01-21): 'January 21, 2026: DV360 Central Partner ID Support for Multi-Buyer Deals'. That note (ms.date 2026-01-26; 'Release date: January 21, 2026', 'Product: Microsoft Monetize and Curat | contradicted | The About and Curated Deals pages (ms.date 2025-10-21) name Microsoft Invest and 'all major DSPs' as buying routes, and Microsoft said on 14 May 2025 that it 'will no longer support media buying through our DSP, Microsoft Invest, starting on February 28, 2026' (execution after that date not confirmed on a Microsoft page opened). Microsoft Curate's release note of 21 January 2026 (Product: 'Microsoft Monetize and Curate'), however, documents deal creation for one named third-party DSP: curators choose the DV360 Partner ID or Central Partner ID 'provided by the buyer' in Buyer Selection, 'Deal permissions are managed directly within Display & Video 360 (DV360)', and buyer-seat reporting is available. At the cutoff the documentation therefore names one live third-party endpoint with documented activation functionality (Google Display & Video 360) besides the withdrawn Invest route. The Trad | the source does not carry the claim as first written | MICROSOFTCURATE capability activation_routes |
| V-microsoft-curate-160 | No observed sample of Curate deals checked against ads.txt, sellers.json or SupplyChain; documentation says ads.txt targeting not supported, AppNexus Blocklist always applied, and auctions now carry 'industry IDs, schains, and, in recent years, curated deals'; Adalytics 2024 recorded as counterevidence. Grade C, unknown. | Refute-mode verifier for microsoft-curate.json re-opened the cited source | Create page: 'Ads.txt Do not use (not supported for curated deal line items).'; 'by default, the AppNexus Blocklist will always be applied'. Curation Optimization page (ms.date 2025-10-21, 294 words), opening paragraph: 'With the growth of the programmatic industry, systems must process a greater volume of auctions with more and more information. For example, there are now industry IDs, schains, and, in recent years, curated deals incorporated into the auction.' The sentence explains auction processing load as the reason for sampling; it says nothing about whether curated bid requests carry a complete SupplyChain object or how a curator appears in it. | supported with qualification | No observed sample of Microsoft Curate deals checked against ads.txt, sellers.json or SupplyChain was found or made for this profile. The documentation says that ads.txt targeting is 'not supported for curated deal line items', so a curator cannot add an ads.txt filter, and that the AppNexus Blocklist is always applied. The Curation Optimization page mentions schains only as an example of the growing information in auctions ('industry IDs, schains, and, in recent years, curated deals'), which is not a statement about schain completeness for curated deals. No page opened describes Monetize's ads.txt enforcement, IVT filtering for curated deals, or how a curator appears in sellers.json or the schain. One independent observation exists: Adalytics, 2024, funding not stated, found Microsoft Xandr among four exchanges that appeared to relay www3.forbes.com inventory as www, some of it through | scope, unit, date or wording narrowed to what the source carries | MICROSOFTCURATE capability supply_path_integrity |
| V-microsoft-curate-161 | A Microsoft Curate curated deal line item is set up for one selected buyer that cannot be changed after saving, while the same seller inventory remains biddable on the open exchange; the product page's 'ability to assign deals to multiple buyers on any DSP' is not documented as a mechanism on the Learn pages opened. | Refute-mode verifier for microsoft-curate.json re-opened the cited source | Create page: 'You cannot change the selected buyer after the curated deal line item is created and saved.' Curated Deals page: 'With curation, buyers can still bid on seller inventory via open exchange'. Product page: 'the ability to assign deals to multiple buyers on any DSP.' Curate release note of 21 January 2026 ('Product: Microsoft Monetize and Curate'): 'This enhancement restores multi-buyer deal functionality with DV360 buyers'; 'You can now create a single deal that is automatically eligible for all DV360 Partner IDs associated with a selected Central Partner ID.'; 'Confirm whether you are targeting: A single Partner ID, or A Central Partner ID (multi-buyer scope).'; 'Each deal can target only one Central Partner ID.' | supported with qualification | A Microsoft Curate curated deal line item is set up for one buyer selection that cannot be changed after saving, while the same seller inventory remains biddable on the open exchange. Microsoft Curate's release note of 21 January 2026 documents one multi-buyer mechanism: for DV360 buyers the selection can be a Central Partner ID, which makes a single deal 'automatically eligible for all DV360 Partner IDs' associated with it (one Central Partner ID per deal). The product page's broader 'ability to assign deals to multiple buyers on any DSP' is documented only for DV360 on the Learn pages opened. | scope, unit, date or wording narrowed to what the source carries | CUR-MICROSOFTCURATE-F08 |
| V-microsoft-curate-162 | Microsoft Curate's documentation names Microsoft Invest and 'all major DSPs' as buying routes without naming a third-party DSP, and Microsoft announced on 14 May 2025 that it would no longer support media buying through Invest from 28 February 2026, so at the cutoff the only named route is withdrawn and no third-party DSP with documented activation is named on any Microsoft page opened. | Refute-mode verifier for microsoft-curate.json re-opened the cited source | The seven support passages are verbatim (About page wording rests on CUR-E-F05 and CUR-U-F03; the rest re-opened here). Microsoft Curate's release note 'Microsoft Curate (January 21, 2026) DV360 Central Partner ID Support for Multi-Buyer Deals' (ms.date 2026-01-26; 'Product: Microsoft Monetize and Curate') names a third-party DSP with documented functionality: 'Microsoft Monetize and Curate now fully support targeting DV360 Central Partner IDs'; 'Deal permissions are managed directly within Display & Video 360 (DV360)'; 'When creating deals for DV360 buyers: In Buyer Selection, choose the Central Partner ID provided by the buyer.' | contradicted | Microsoft Curate's About and Curated Deals pages (ms.date 2025-10-21) name Microsoft Invest and 'all major DSPs' as buying routes, and Microsoft announced on 14 May 2025 that it would no longer support media buying through Invest from 28 February 2026. Curate's release note of 21 January 2026, however, documents deal creation for DV360 buyers by Partner ID or Central Partner ID, with deal permissions managed in Display & Video 360. At the cutoff the Curate documentation therefore names one third-party DSP with documented activation functionality (DV360) besides the withdrawn Invest route. The Trade Desk is named only for Monetize sellers' deal-sync status (April 2026), and no other DSP is named on the Microsoft pages opened. | the source does not carry the claim as first written | CUR-MICROSOFTCURATE-F11 |
| V-microsoft-curate-163 | Microsoft's 10 September 2026 post announces 'agentic workflows' for Microsoft Curate through Copilot's MCP-backed capabilities, but describes no deployment, customer, period or executed decision, so on this record agentic selection on Curate is at the announcement stage. | Refute-mode verifier for microsoft-curate.json re-opened the cited source | Post (Nick Bolt, Senior Product Marketer, September 10, 2026): section heading 'Bringing agentic workflows to Microsoft Curate'; 'Curators can use Copilot’s MCP-backed capabilities to bring account-specific intelligence directly into Microsoft Curate.'; 'Model Context Protocol (MCP) is an open standard that gives AI agents a consistent, controlled way' (the rest of that sentence was returned only as a paraphrase); 'Copilot in Microsoft Curate adds contextual guidance directly within the platform'; reference [5] 'Microsoft Advertising, Publisher Release Roundup: Copilot Enhancements, July 2026' (not opened). Curate release note 'July 28, 2025 - Copilot in Microsoft Curate' (ms.date 2025-10-21): 'Copilot in Microsoft Curate provides real-time, contextual assistance, helping users navigate the platform efficiently. Built on a Large Language Model (LLM), Copilot enhances the user experience | supported with qualification | Copilot in Microsoft Curate, an LLM assistant that answers from Microsoft Learn help documentation, has been released since 28 July 2025 per Curate's release notes. Microsoft's 10 September 2026 post announces 'agentic workflows' for Microsoft Curate through 'Copilot's MCP-backed capabilities' but describes no deployment, customer, period or executed decision for them. On this record the Copilot help assistant is live, and MCP-backed agentic selection on Curate is at the announcement stage. | scope, unit, date or wording narrowed to what the source carries | CUR-MICROSOFTCURATE-F18 (and differentiation 5) |
| V-microsoft-curate-164 | The 'Create a Curated Deal Line Item' page was updated 2026-02-18; it still offers Invest-era buy-side objects and makes no mention of the Invest change. | Refute-mode verifier for microsoft-curate.json re-opened the cited source | ms.date 2026-02-18, updated_at 2026-02-18T04:56Z. Prerequisites: 'You will need to create or have access to the following buy-side objects: Advertiser ... Insertion order (seamless)'. The word 'Invest' does not appear anywhere on the page. | supported with qualification | The 'Create a Curated Deal Line Item' page was updated (ms.date and updated_at 2026-02-18). It requires buy-side objects (an advertiser and a seamless insertion order), does not name Microsoft Invest at all, and makes no mention of the Invest change. | scope, unit, date or wording narrowed to what the source carries | MICROSOFTCURATE timeline 2026-02-18 |
| V-microsoft-curate-165 | The announced start of the end of media-buying support in Microsoft Invest, the only DSP Microsoft Curate's documentation names as a buying route; execution unconfirmed. | Refute-mode verifier for microsoft-curate.json re-opened the cited source | The May 2025 post sets the date ('starting on February 28, 2026'). Curate's release note of 21 January 2026 documents deals for DV360 buyers ('When creating deals for DV360 buyers: In Buyer Selection, choose the Central Partner ID provided by the buyer.'), so Invest is not the only DSP the Curate documentation names. | contradicted | This is the announced start of the end of media-buying support in Microsoft Invest, one of two DSPs Microsoft Curate's documentation names as a buying route (the other is Google Display & Video 360, documented in Curate's release note of 21 January 2026). No Microsoft page opened after this date confirms or contradicts the execution of the Invest change. | the source does not carry the claim as first written | MICROSOFTCURATE timeline 2026-02-28 |
| V-microsoft-curate-166 | The record 'names no live buying route: the only named DSP, Microsoft Invest, lost media-buying support from 28 February 2026 ... and "all major DSPs" is unsupported'; 'No third-party DSP is named as a route on any Microsoft page opened'; the documentation 'names no third-party DSP'; poor fit for buyers needing 'a documented activation path in a named third-party DSP (none named; Invest withdrawn)'. | Refute-mode verifier for microsoft-curate.json re-opened the cited source | Microsoft Curate's release note of 21 January 2026 documents deals for DV360 buyers by Partner ID or Central Partner ID, with deal permissions managed in Display & Video 360 (passages quoted in the activation_routes entry). | contradicted | Summary: replace the route sentence with 'It names one live third-party buying route: Curate's release note of 21 January 2026 documents deals for Google Display & Video 360 buyers by Partner ID or Central Partner ID. The other named DSP, Microsoft Invest, lost media-buying support from 28 February 2026 by Microsoft's own announcement, and "all major DSPs" is otherwise unsupported.' Limits 3: 'The only third-party DSP documented as a route is DV360 (release note, 21 January 2026); the other named route, Microsoft Invest, lost media-buying support from 28 February 2026 by Microsoft's announcement.' Risks 1: 'Route risk: the About and Curated Deals pages (ms.date 2025-10-21) still name Microsoft Invest, whose media buying Microsoft stopped supporting from 28 February 2026; the only third-party DSP documented is DV360, so a buyer on any other DSP cannot establish from public records that it | the source does not carry the claim as first written | MICROSOFTCURATE summary, limits 3, risks 1, poor_fit 2 and open_questions 1 (activation routes) |
| V-microsoft-curate-167 | One buyer per curated deal line item, fixed after saving; the product page's multiple-buyer phrase is not documented as a mechanism. | Refute-mode verifier for microsoft-curate.json re-opened the cited source | As in F08: one buyer selection fixed after saving; for DV360 the selection can be a Central Partner ID covering all associated Partner IDs ('multi-buyer scope'), one Central Partner ID per deal. | supported with qualification | Replace 'one buyer per line item' with 'one buyer selection per line item, fixed after saving; for DV360 buyers the selection can be a Central Partner ID that makes the deal eligible for all associated DV360 Partner IDs (release note, 21 January 2026)'. In limits 1, replace 'is not documented as a mechanism' with 'is documented only for DV360 Central Partner IDs'. | scope, unit, date or wording narrowed to what the source carries | MICROSOFTCURATE limits 1, who_buys, data_tables microsoft_curate_access_register (exclusivity_buyer) and microsoft_curate_who_decides (buyer eligibility) |
| V-microsoft-curate-168 | Eleven rows across curation fee, exchange charge to curator, SSP take rate, buyer-side fees, data fee, win condition and DSP take rate, with payer, collector, base, amount, default and buyer visibility. | Refute-mode verifier for microsoft-curate.json re-opened the cited source | Rows for BrASC, MASC, SASC, the fixed-floor curator take, cross-device and the win condition match the pages (MASC 'is an alternative charge which can be applied when a curator takes a margin'; 'Only BrASC can be charged as a fee'; SASC 'the charge for selling inventory to third-party buyers'). For the buyer-side fees row, Charges for Sellers mentions Microsoft's buyer charge once: 'Resold Revenue is net earnings to the network after Microsoft Advertising Seller and Buyer Auction Service Charges have been applied.' It does not tie that charge to curated deals or give its base. | supported with qualification | In the buyer-side fees row, replace the collector 'not stated on the pages read' with 'not stated for curated deals; Charges for Sellers refers to Microsoft Advertising "Buyer Auction Service Charges" without base, rate or curated-deal scope'. Other rows stand. | scope, unit, date or wording narrowed to what the source carries | MICROSOFTCURATE data_table microsoft_curate_fee_stack |
| V-pubmatic-connect-169 | Disclosure asserted in marketing for the authenticated Media Buyer Console (Inventory Discovery Tool, Log-Level Transaction Verification, Two-Way Fee Transparency), publisher lists accepted by the Deal Management agent and Adloox's console steps; no identified example deal with a seller or domain list; the US CTV package page could not be read; availability live, grade D, disclosure authenticated, date 2026-07-17. | Refute-mode verifier for pubmatic-connect.json re-opened the cited source | Console page: 'Inventory Discovery Tool: View and analyze available inventory from the past 30 days, helping to make informed decisions'; 'Log-Level Transaction Verification: Easily access transaction logs for verification'; 'Two-Way Fee Transparency: Allowing buyers and sellers to know exactly where the inventory is coming from and delivering visibility into the fee structures they’ve agreed upon'; also 'Brand Safety: With publisher block lists, buyers have precise control over brand safety'. Blog: 'specifying pricing, flight dates, targeting parameters, and publisher lists conversationally'. Adloox steps: 'Select Maximize Reach, Upload or Select publishers'; the page also names the console's 'IAB Categories', 'Viewability' and 'Domains' tabs. The US CTV page opened on this pass (article element): 'CTV Marketplaces / Scaled CTV audiences and inventory across a variety of premium publish | supported with qualification | Replace 'and the US CTV package page could not be read (access error)' with 'and the US CTV package page (read in its article element on 3 November 2026) names only 'CTV Marketplaces' with 'Scaled CTV audiences and inventory across a variety of premium publishers', no seller list'; add that the console offers 'publisher block lists' and that Adloox's steps let the builder 'Select Maximize Reach, Upload or Select publishers' (so a deal need not carry a publisher list). | scope, unit, date or wording narrowed to what the source carries | PUBMATICCONNECT capability composition_disclosure |
| V-pubmatic-connect-170 | Every quantified quality claim found is a vendor assertion: Agence 79 (5x CTR vs the open market; VCR +16.7% and viewability +12.5% vs targets; 'Agence 79 internal data'; no period, volume, measurer or method), the March 2025 yield and price figures (25%+ eCPM, CPM 'improve by 37%'), and a 2022 Lotame case; no independent study against a pre-curation comparator; live, grade D, public, 2026-06-11. | Refute-mode verifier for pubmatic-connect.json re-opened the cited source | Release: 'A remarkable 25%+ increase in eCPMs compared to non-curated deals'; 'Historically, campaigns with data applied via Connect have seen CPM improve by 37% when compared to campaigns where the same data is applied on the buy-side.' Connect for Buyers: '94% VIDEO COMPLETION RATE IN SAFARI/ FIREFOX / 83% VIEWABILITY RATE IN SAFARI/FIREFOX / 34%+ VIEWABILITY BENCHMARK / Source: Lotame Internal Data, 4/11/22 – 5/9/22'. Agence 79 passages as verified verbatim by stream C's verifier (CUR-C-F01). Not in the cell: the Goodway Group case study linked from the storefront ('VIEW CASE STUDY') reports, for 'PubMatic’s Consumer Experience Auction Package with Jounce Media', '180%+ HIGHER WIN RATES', '24% INCREASE IN VIEWABILITY OVER BENCHMARK', '7-14% MORE COST-EFFICIENT CPMs OVER OTHER DEALS', increased audible rates 'as reported by Moat', 'Source: Goodway Group Internal Data, 2/1/22 – 5/31/22' | supported with qualification | Add to the observed text: 'a 2022 Goodway Group case for PubMatic's Consumer Experience Auction Package built with Jounce Media (180%+ higher win rates, 24% viewability over an undefined benchmark, 7-14% more cost-efficient CPMs than other deals, audible rates 'as reported by Moat'; Goodway Group internal data, 1 February to 31 May 2022; no method; more than 24 months old at the cutoff)'. Keep the conclusion that every quantified claim is a vendor or agency assertion and that no independent study against a pre-curation comparator was found. | scope, unit, date or wording narrowed to what the source carries | PUBMATICCONNECT capability quality_evidence |
| V-pubmatic-connect-171 | US availability shown by vendor marketing only (region pages under /auction-packages/us/; US-sports packages; Comscore's Reston, VA dateline); no opened record names a US customer with a period for a Connect deal (IPG Mediabrands, Goodway Group and Minute Media quotes give no geography or period); live, grade D, public, 2026-11-03. | Refute-mode verifier for pubmatic-connect.json re-opened the cited source | Storefront: 'Explore Auction Packages available in your region:' with 'Global' as the region shown; the MLB and NFL + NCAA Football packages appear in that default view, not on a US region page. Two /us/ region pages open (Video & VCR Refresh; CTV), but their article text never mentions the United States. The Goodway Group case study linked from the storefront quote gives a period: 'Source: Goodway Group Internal Data, 2/1/22 – 5/31/22', for 'PubMatic’s Consumer Experience Auction Package', with no geography stated. IPG Mediabrands quote: no geography or period. | supported with qualification | US availability is shown by vendor marketing only: region pages under pubmatic.com/auction-packages/us/ (Video & VCR Refresh and CTV pages read; their text does not mention the US) and US-sports packages (MLB; 'curated NFL + NCAA Football auction packages') shown in the storefront's default 'Global' view, plus Comscore's Reston, VA dateline. One opened record names a customer with a period, the Goodway Group case for PubMatic's Consumer Experience Auction Package (Goodway Group internal data, 1 February to 31 May 2022), but it states no geography, so US production use rests on the inference that Goodway Group is a US agency. | scope, unit, date or wording narrowed to what the source carries | PUBMATICCONNECT capability us_availability |
| V-pubmatic-connect-172 | No observed sample of a Connect deal's ads.txt, sellers.json or SupplyChain; console self-description ('adhering to industry standards such as sellers.json and ads.txt'); 10-K fraud detection without a named accredited measurer; blog says curators 'don’t sit in the media payment chain'; live, grade D, public, date 2026-03-19. | Refute-mode verifier for pubmatic-connect.json re-opened the cited source | Console: 'Compliance: MBC ensures brand safety and quality control by adhering to industry standards such as sellers.json and ads.txt'. Blog, dated 'March 19, 2025': 'curators do select inventory and data, but unlike traditional ad networks, they don’t sit in the media payment chain and have no financial incentives to favor certain media over others.' The cell's other sources are the 10-K (filed 2026-02-26 per EDGAR XBRL) and the undated 2026 console page. | supported with qualification | claim kept with the verifier’s qualification in the ledger | scope, unit, date or wording narrowed to what the source carries | PUBMATICCONNECT capability supply_path_integrity |
| V-pubmatic-connect-173 | Every quantified quality or performance claim found for Connect curation is a vendor or agency assertion without published method: Agence 79, the March 2025 25%+ eCPM and 37% CPM claims, and a 2022 Lotame identity case; H3 unresolved. | Refute-mode verifier for pubmatic-connect.json re-opened the cited source | The S04 and S01 passages verbatim. The list is incomplete: the Goodway Group case linked from the auction-package storefront reports '180%+ HIGHER WIN RATES', '24% INCREASE IN VIEWABILITY OVER BENCHMARK' and '7-14% MORE COST-EFFICIENT CPMs OVER OTHER DEALS' for 'PubMatic’s Consumer Experience Auction Package with Jounce Media', audible rates 'as reported by Moat', 'Source: Goodway Group Internal Data, 2/1/22 – 5/31/22'. | supported with qualification | Every quantified quality or performance claim found for PubMatic curation is a vendor or agency assertion without published method: the Agence 79 campaign (CTR five times 'the same campaign on the open market', VCR +16.7% and viewability +12.5% against targets, 'Source: Agence 79 internal data'), the March 2025 release's 25%+ eCPM and 37% CPM claims, a 2022 Goodway Group case for the Consumer Experience Auction Package built with Jounce Media (180%+ win rates, 24% viewability over an undefined benchmark, 7-14% more cost-efficient CPMs than other deals, audibility 'as reported by Moat', Goodway Group internal data, 1 February to 31 May 2022), and a 2022 Lotame identity case on the Connect for Buyers page (interest mark); none measures a registered quality dimension against a pre-curation comparator with a published design, so H3 stays unresolved on this product's record. | scope, unit, date or wording narrowed to what the source carries | CUR-PUBMATICCONNECT-F13 |
| V-pubmatic-connect-174 | US availability of Connect auction packages is shown only by vendor marketing (region pages under /auction-packages/us/, US-sports packages); no opened record names a US customer with a period for a Connect curated deal. | Refute-mode verifier for pubmatic-connect.json re-opened the cited source | As under the us_availability check. The Video & VCR article lead is verbatim: 'PubMatic’s Auction Packages deliver contextually relevant and targeted omnichannel reach at scale.' | supported with qualification | US availability of Connect auction packages is shown only by vendor marketing (region pages under pubmatic.com/auction-packages/us/, whose text does not mention the US, and US-sports packages shown in the storefront's default 'Global' view); the one opened record naming a customer with a period (Goodway Group, Consumer Experience Auction Package, 1 February to 31 May 2022) states no geography, so no opened record documents a US customer with a period for a Connect curated deal except by inference from Goodway Group's location. | scope, unit, date or wording narrowed to what the source carries | CUR-PUBMATICCONNECT-F16 |
| V-pubmatic-connect-175 | The Deal Management agent, introduced for publishers in August 2025 with 'more than 1,000 AI-powered deals', was extended on 28 April 2026 to buyers and curators to create auction packages in natural language inside AgenticOS or over MCP, with the agent producing 'a structured summary for review' activated 'With a single approval' and humans 'fully in control of commercial decisions'; decision = deal configuration under human approval. | Refute-mode verifier for pubmatic-connect.json re-opened the cited source | 'Last August, we introduced our Deal Management agent ... bringing AI-powered deal setup to publishers.' 'Since launch, the agent has enabled more than 1,000 AI-powered deals, transacting millions in platform spend.' 'Buyers and curators can now use the Deal Management agent to create auction packages through a natural language interface — specifying pricing, flight dates, targeting parameters, and publisher lists conversationally.' 'The agent is accessible within AgenticOS via PubMatic Assistant. Buyers can also connect ... via MCP connectivity.' The review step belongs to the publisher PG example: 'the agent generates a structured summary for review. With a single approval, those deals can be activated simultaneously.' For buyers the blog says 'buyers simply describe what they want and the agent handles the rest, keeping every decision auditable'. Close: 'keeping buyers and publishers | supported with qualification | PubMatic's blog of 28 April 2026 says its Deal Management agent, introduced for publishers in August 2025 ('Last August') and credited with 'more than 1,000 AI-powered deals' (publisher PMP and PG deals), can 'now' be used by buyers and curators to 'create auction packages through a natural language interface', inside AgenticOS via PubMatic Assistant or over MCP, though its closing sentence speaks of 'extending those benefits to buyers and curators next'. The review-and-single-approval step ('a structured summary for review', 'With a single approval') is described in the publisher PG example; for buyers the blog says the agent 'handles the rest, keeping every decision auditable' and keeps 'buyers and publishers fully in control of commercial decisions'. No named curator deployment, period or executed curated deal is documented. | scope, unit, date or wording narrowed to what the source carries | CUR-PUBMATICCONNECT-F19 |
| V-pubmatic-connect-176 | FY2025 revenue $282,926k (FY2024 $291,256k; FY2023 $267,014k), FY2025 net loss $14,462k; Q2 2026 revenue $78,593k (Q2 2025 $71,095k), Q2 2026 net loss $1,202k; all net basis ('gross billings to buyers, net of amounts we pay publishers'); no Connect or curation revenue. | Refute-mode verifier for pubmatic-connect.json re-opened the cited source | XBRL (10-K filed 2026-02-26): revenue 282,926,000 (2025), 291,256,000 (2024), 267,014,000 (2023); net income (loss) -14,462,000, 12,504,000, 8,881,000. XBRL (10-Q filed 2026-08-06): revenue 78,593,000 (Q2 2026), 71,095,000 (Q2 2025), 141,160,000 (H1 2026), 134,920,000 (H1 2025); net loss -1,202,000 (Q2 2026), -5,208,000 (Q2 2025), -13,712,000 (H1 2026), -14,694,000 (H1 2025). 10-K text table matches. 10-K: 'We report revenue on a net basis.' 'This represents gross billings to buyers, net of amounts we pay publishers and rebates associated with SPO agreements'. 10-Q: 'net of amounts we pay publishers and rebates associated with SPO agreements with buyers.' | supported with qualification | Replace the definition quotation with 'gross billings to buyers, net of amounts we pay publishers and rebates associated with SPO agreements' (both filings), and note the 10-K filing date (26 February 2026). All figures stand. | scope, unit, date or wording narrowed to what the source carries | CUR-PUBMATICCONNECT-F20 |
| V-pubmatic-connect-177 | PubMatic says it launched Connect in 2022; 'The March 2025 release says PubMatic began building Connect 'five years ago'.' | Refute-mode verifier for pubmatic-connect.json re-opened the cited source | Blog: 'launching Connect in 2022'. Release: 'Recognizing this shift five years ago, PubMatic invested in building Connect'. | supported with qualification | Replace the second sentence with: 'The March 2025 release says PubMatic recognised the shift to curated deals 'five years ago' and invested in building Connect; it gives no build start date.' | scope, unit, date or wording narrowed to what the source carries | CUR-PUBMATICCONNECT timeline 2022 |
| V-pubmatic-connect-178 | Lotame, Xaxis, Xandr and PubMatic Panorama ID campaign shown on the Connect for Buyers page. | Refute-mode verifier for pubmatic-connect.json re-opened the cited source | 'This collaborative campaign — with Xaxis, Xandr, and PubMatic — reflects the power to both combat cookie deprecation and outperform existing third-party cookies. Our Panorama ID continues to provide the data connectivity and addressability ...', attributed to 'SEBASTIAN YOFF, MANAGING DIRECTOR, LATIN AMERICA & US HISPANICS, LOTAME'; 'Source: Lotame Internal Data, 4/11/22 – 5/9/22'. | supported with qualification | Campaign with Xaxis, Xandr and PubMatic using Lotame's Panorama ID, shown on the Connect for Buyers page ('Source: Lotame Internal Data, 4/11/22 – 5/9/22'); interest mark (Lotame). | scope, unit, date or wording narrowed to what the source carries | CUR-PUBMATICCONNECT timeline 2022-04-11/2022-05-09 |
| V-pubmatic-connect-179 | FY2025 Form 10-K share counts dated 19 February 2026; filing date not captured. | Refute-mode verifier for pubmatic-connect.json re-opened the cited source | XBRL facts for accession 0001422930-26-000010 (form 10-K, fy 2025) carry filed date 2026-02-26. | supported with qualification | Add an event: '2026-02-26, filing: Form 10-K for the year ended 31 December 2025 filed (accession 0001422930-26-000010)', and drop 'the filing date itself was not captured'. | scope, unit, date or wording narrowed to what the source carries | CUR-PUBMATICCONNECT timeline 2026-02-19 |
| V-pubmatic-connect-180 | PubMatic extends the Deal Management agent to buyers and curators to create auction packages through natural language, inside AgenticOS via PubMatic Assistant or over MCP (event type 'feature extension (buy side)'). | Refute-mode verifier for pubmatic-connect.json re-opened the cited source | 'Buyers and curators can now use the Deal Management agent to create auction packages through a natural language interface'; closing: 'extending those benefits to buyers and curators next'. | supported with qualification | Set event_type to 'vendor announcement of availability (buy side)' and add: 'the blog's closing sentence describes the buyer and curator extension as coming next; no general-availability record or curator deployment was found'. | scope, unit, date or wording narrowed to what the source carries | CUR-PUBMATICCONNECT timeline 2026-04-28 |
| V-pubmatic-connect-181 | Publishers opted into PubMatic's curation tools saw 'up to 5% incremental daily deal revenue uplift' (some '10%+'), 'Up to 14 percentage point increase in spending from a broader range of buyers' and 'A remarkable 25%+ increase in eCPMs compared to non-curated deals'. | Refute-mode verifier for pubmatic-connect.json re-opened the cited source | 'Revenue Uplift: Many publishers opted into PubMatic’s curation tools have seen up to 5% incremental daily deal revenue uplift, with some publishers seeing as much as 10%+ increase'; 'Diverse Demand: Up to 14 percentage point increase in spending from a broader range of buyers'; 'Higher eCPMs: A remarkable 25%+ increase in eCPMs compared to non-curated deals'. | supported with qualification | 'Many publishers opted into PubMatic’s curation tools have seen up to 5% incremental daily deal revenue uplift' (some 'as much as 10%+'), 'Up to 14 percentage point increase in spending from a broader range of buyers' and 'A remarkable 25%+ increase in eCPMs compared to non-curated deals'. | scope, unit, date or wording narrowed to what the source carries | CUR-PUBMATICCONNECT performance_evidence 2 (March 2025 publisher figures) |
| V-pubmatic-connect-182 | 94% video completion rate and 83% viewability rate in Safari/Firefox against a '34%+ viewability benchmark'. | Refute-mode verifier for pubmatic-connect.json re-opened the cited source | Three stat tiles: '94% VIDEO COMPLETION RATE IN SAFARI/ FIREFOX', '83% VIEWABILITY RATE IN SAFARI/FIREFOX', '34%+ VIEWABILITY BENCHMARK'; 'Source: Lotame Internal Data, 4/11/22 – 5/9/22'. | supported with qualification | 94% video completion rate and 83% viewability rate in Safari/Firefox, shown beside a tile reading '34%+ viewability benchmark' (whether 34% is the benchmark level or the margin over it is not stated), with Lotame's Panorama ID. | scope, unit, date or wording narrowed to what the source carries | CUR-PUBMATICCONNECT performance_evidence 4 (Lotame Panorama ID case) |
| V-pubmatic-connect-183 | 'My sense is that when we apply it on the sell side, the fees are lower than in the DSP.' | Refute-mode verifier for pubmatic-connect.json re-opened the cited source | 'My sense is that when we apply it on the sell side, the fees are lower than in the DSP.' followed by 'I'm hesitant to say that's the case for all DSPs.' | supported with qualification | Add the CEO's hedge: 'I'm hesitant to say that's the case for all DSPs.' | scope, unit, date or wording narrowed to what the source carries | CUR-PUBMATICCONNECT performance_evidence 7 (CEO on sell-side fees) |
| V-pubmatic-connect-184 | Storefront testimonials (Publicis Groupe India, OMG Germany, Goodway Group) report increased viewability, win rates, efficient CPMs and exceeding an 'Audible Rates benchmark' from auction packages; 'testimonials; no figures, periods or method'. | Refute-mode verifier for pubmatic-connect.json re-opened the cited source | Storefront quotes verbatim (Publicis Groupe India on the 'Consumer Experience Auction Package'; OMG Germany: 'Bringing our activation closer to the sell-side has driven substantial benefits in efficiency, transparency and control', which does not mention auction packages; Goodway Group on 'Audible Rates benchmark'). Each quote has a 'VIEW CASE STUDY' link; the Goodway case study gives '180%+ HIGHER WIN RATES', '24% INCREASE IN VIEWABILITY OVER BENCHMARK', '7-14% MORE COST-EFFICIENT CPMs OVER OTHER DEALS', Moat-reported audibility and 'Source: Goodway Group Internal Data, 2/1/22 – 5/31/22'. | supported with qualification | Storefront testimonials (Publicis Groupe India and Goodway Group on the Consumer Experience Auction Package; OMG Germany on sell-side activation generally) report increased viewability, win rates, efficient CPMs and exceeding an 'Audible Rates benchmark'; the quotes carry no figures, but the linked Goodway Group case study does (see F13 correction; agency internal data, February to May 2022, no method). | scope, unit, date or wording narrowed to what the source carries | CUR-PUBMATICCONNECT performance_evidence 8 (storefront testimonials) |
| V-stagwell-curate-185 | Not disclosed for Curate; company-wide the FY2025 10-K describes revenue from 'agency arrangements in the form of fees for services performed, commissions, and from performance incentives or bonuses'; the filings do not break Curate out. | Refute-mode verifier for stagwell-curate.json re-opened the cited source | 10-K, Sources of Revenue: 'The primary source of revenue is from agency arrangements in the form of fees for services performed, commissions, and from performance incentives or bonuses.' The release states no fee; the Q3 release does not mention Curate. | supported with qualification | Not disclosed. No source opened states a curation fee, platform fee, data fee or margin for Curate, or says who collects one. Company-wide, the FY2025 10-K says 'The primary source of revenue is from agency arrangements in the form of fees for services performed, commissions, and from performance incentives or bonuses'. Whether Curate is charged as a separate line, folded into agency fees or commissions, or earned through negotiated pricing ('exclusive pricing') is not stated. The filings do not break Curate out. | scope, unit, date or wording narrowed to what the source carries | STAGWELLCURATE how_it_earns |
| V-stagwell-curate-186 | Supply: FreeWheel and Magnite 'will provide' publisher relationships (July); 'FreeWheel will power Stagwell Curate' and Curation Hub and Buyer Cloud 'will integrate' (April); Digiday paraphrase 'small set of preferred SSPs'. Data: 'log‑level data' and Sincera supply-quality models; Digiday: Claude agents and OpenSincera data for a 'domain health check'; 'OpenSincera is The Trade Desk's free ad-experience metadata application, live from 6 June 2025'. Buying: existing DSP, none named. | Refute-mode verifier for stagwell-curate.json re-opened the cited source | Release: 'Partners including Freewheel and Magnite will provide access to deep, direct publisher relationships' 'across streaming, video, web and audio'; 'supply‑quality models powered by The Trade Desk's Sincera'; inventory 'continuously evaluated using log‑level data'; no DSP named. Digiday narration: 'small set of preferred SSPs', 'AI agents built in Claude', OpenSincera, 'domain health check'. April passages verified by CUR-G-F03. OpenSincera release (verified by CUR-G-F19, dated 13 May 2025): it 'will be available at open.sincera.io, starting on June 6, 2025'. | supported with qualification | Replace 'live from 6 June 2025' with 'announced on 13 May 2025 as available at open.sincera.io from 6 June 2025'. | scope, unit, date or wording narrowed to what the source carries | STAGWELLCURATE supply_and_data_dependencies |
| V-stagwell-curate-187 | The Trade Desk's OpenSincera became available at open.sincera.io (announced 13 May 2025). | Refute-mode verifier for stagwell-curate.json re-opened the cited source | Per stream G's verifier (CUR-G-F19, supported as stated), the 13 May 2025 release says OpenSincera 'will be available at open.sincera.io, starting on June 6, 2025'. That is a stated future date in an announcement; no source in the record observes availability on that date. | supported with qualification | 2025-06-06 (announced date): The Trade Desk announced on 13 May 2025 that OpenSincera 'will be available at open.sincera.io, starting on June 6, 2025'; availability on that date is the vendor's stated plan, not an observed launch. Stagwell later names Sincera as the source of its supply-quality models. | scope, unit, date or wording narrowed to what the source carries | STAGWELLCURATE timeline 2025-06-06 |
| V-stagwell-curate-188 | Provenance documented only for the supply-quality input at source type and provider level ('log‑level data' unsourced; Sincera models; Digiday: OpenSincera data); OpenSincera's release describes non-personal ad-experience metadata; audience, contextual and performance signals have no provider, method, use, permission or legal basis. Live, grade D, public. | Refute-mode verifier for stagwell-curate.json re-opened the cited source | Release: 'log‑level data' (no source named) and 'supply‑quality models powered by The Trade Desk's Sincera'; bespoke packages on audiences, contextual signals, KPIs (no provider). Digiday narration: 'data pulled from The Trade Desk's OpenSincera'. OpenSincera release (per CUR-G-F19): 'metrics on the ad experience such as ads-to-content ratio, page weight, average ads-in-view, and ad refresh rate'. | supported with qualification | Replace 'OpenSincera's release describes non-personal ad-experience metadata such as' with 'OpenSincera's release describes ad-experience metrics (coded here as non-personal contextual inputs, an inference) such as', and add that the release names Sincera models while Digiday names OpenSincera data, so which product's data Stagwell applies is not settled. | scope, unit, date or wording narrowed to what the source carries | STAGWELLCURATE capability data_provenance |
| V-stagwell-curate-189 | Four advertised environments (CTV, web video, web display, audio); April release also names streaming and linear TV; execution claimed only in the release's present tense; no deal, case, partner page or filing documents execution; FreeWheel's Curation Hub page covers CTV only and does not name Stagwell; mobile in-app not stated. Live, grade D, public. | Refute-mode verifier for stagwell-curate.json re-opened the cited source | Release: marketplace paragraph 'across CTV, online video, display and audio'; opening paragraph 'across streaming TV, video, display and audio'; partner sentence (future tense) 'across streaming, video, web and audio'. FreeWheel Curation Hub page: 'Elevate your CTV media strategy with FreeWheel Curation Hub'; it refers to CTV, streaming, linear TV and video, and mentions no display, web, audio or mobile; Stagwell not mentioned. April release 'Unified cross-screen activation across streaming and linear TV' (per CUR-G-F03). | supported with qualification | Replace 'FreeWheel's Curation Hub page covers CTV only' with 'FreeWheel's Curation Hub page covers CTV and TV video only (CTV, streaming and linear TV; no display, web, audio or mobile)', and add that the release's opening paragraph words the scope as 'streaming TV, video, display and audio'. | scope, unit, date or wording narrowed to what the source carries | STAGWELLCURATE capability channel_execution |
| V-stagwell-curate-190 | No charge disclosed in any of the four series; no curfeetype observed; general statements only ('pre‑negotiated PMP deals and exclusive pricing', future tense via the Media Machine; 'material reduction in ad tech fees', April; Adams's aim to 'reduce tech fees'); 10-K describes revenue company-wide as agency fees, commissions and incentives. A general pricing assertion without a schedule. | Refute-mode verifier for stagwell-curate.json re-opened the cited source | Release: no fee, rate or figure; 'pre‑negotiated PMP deals and exclusive pricing' sits in the Media Machine sentence ('will connect'). Digiday: 'reduce tech fees' is Digiday narration reporting Adams's claim. April 'More working media for clients enabled by material reduction in ad tech fees' per CUR-G-F04. 10-K: 'The primary source of revenue is from agency arrangements in the form of fees for services performed, commissions, and from performance incentives or bonuses.' | supported with qualification | Replace 'The FY2025 10-K describes revenue company-wide as' with 'The FY2025 10-K says the company's primary source of revenue is'. | scope, unit, date or wording narrowed to what the source carries | STAGWELLCURATE capability pricing_transparency |
| V-stagwell-curate-191 | No certification, accreditation or audit naming Stagwell Curate, its deals or Stagwell Inc. as the assured entity; TAG and MRC registries not searched within budget; Sincera is an input; Stagwell Control is Stagwell's own product (at most self-attestation). Unconfirmed, grade D, mode unknown. | Refute-mode verifier for stagwell-curate.json re-opened the cited source | The opened Curate sources name no certification or audit. A bounded web search by this verifier (3 November 2026) found ABC's release of 6 March 2026 listing 'Assembly Group' among ABC clients recognised in TAG's latest certification release; the page names three TAG programmes but not which seal each client holds, and does not mention Stagwell, MDC, the Stagwell Media Platform or Curate. A search snippet reported a TAG release listing 'Stagwell Global "MDC - Assembly"' among certified companies; that page was not opened. Stagwell's Assembly launched its own 'Assembly Control' brand-safety product in July 2025, built with an unnamed 'leading ad verification and measurement platform'. | supported with qualification | Add: A bounded search on 3 November 2026 found that ABC, a TAG certification auditor, listed 'Assembly Group' among clients recognised in TAG's certification release (6 March 2026), without naming the seal or its scope. That this is Stagwell's Assembly agency is an inference; no source names Stagwell Inc., the Stagwell Media Platform or Curate as an assured entity, and no seal's scope is documented to cover Curate. TAG's and MRC's own registries were not opened. | scope, unit, date or wording narrowed to what the source carries | STAGWELLCURATE capability assurance |
| V-stagwell-curate-192 | Adams told Digiday that Curate should 'reduce tech fees', that clients 'can see our buy data', and that the process is the agency's own rather than a third-party 'black box'. | Refute-mode verifier for stagwell-curate.json re-opened the cited source | '[Clients] can see our buy data' is a direct Adams quote. 'reduce tech fees' is Digiday's narration reporting Adams's claim, and 'black box' is Digiday's narration framing Adams's view of third-party processes; neither is in quotation marks attributed to Adams. | supported with qualification | Digiday reports Adams's aim to 'reduce tech fees' and frames his view against a third-party 'black box' (both Digiday's wording), and quotes Adams directly that clients 'can see our buy data'. | scope, unit, date or wording narrowed to what the source carries | STAGWELLCURATE performance 3 (Adams to Digiday) |
| V-stagwell-curate-193 | No charge disclosed in the deal path; general statements only ('pre‑negotiated PMP deals and exclusive pricing'; 'material reduction in ad tech fees'; 'reduce tech fees'); company-wide the 10-K describes revenue from agency fees, commissions and incentives. | Refute-mode verifier for stagwell-curate.json re-opened the cited source | All three general statements confirmed ('reduce tech fees' is Digiday narration of Adams's claim). The release contains no fee figure. 10-K: 'The primary source of revenue is from agency arrangements in the form of fees for services performed, commissions, and from performance incentives or bonuses.' | supported with qualification | No source opened discloses any charge in Stagwell Curate's deal path (curation or platform fee, data fee, SSP take rate, DSP take rate). The operator's statements are general: 'pre‑negotiated PMP deals and exclusive pricing' (July 2026), 'material reduction in ad tech fees' (April 2026) and, in Digiday's paraphrase, Adams's aim to 'reduce tech fees'. Company-wide, Stagwell's FY2025 10-K says its primary source of revenue is 'agency arrangements in the form of fees for services performed, commissions, and from performance incentives or bonuses'. | scope, unit, date or wording narrowed to what the source carries | CUR-STAGWELLCURATE-F09 |
| V-stagwell-curate-194 | Curate advertised across CTV, online video, display and audio (July 2026); April release names 'streaming and linear TV'; no source documents execution in any environment; FreeWheel's Curation Hub page describes CTV only. | Refute-mode verifier for stagwell-curate.json re-opened the cited source | Release: 'across CTV, online video, display and audio' (marketplace paragraph) and 'across streaming TV, video, display and audio' (opening paragraph). FreeWheel page: CTV, streaming, linear TV and video; no display, web, audio or mobile. | supported with qualification | Stagwell advertises Curate across 'CTV, online video, display and audio' (marketplace paragraph) and 'streaming TV, video, display and audio' (opening paragraph) in July 2026, and its April 2026 release also names 'streaming and linear TV'. No source documents execution in any single environment, and FreeWheel's Curation Hub page, the partner technology named for the platform, describes CTV and TV video only (no display, web, audio or mobile). | scope, unit, date or wording narrowed to what the source carries | CUR-STAGWELLCURATE-F13 |
| V-stagwell-curate-195 | Supply runs through FreeWheel and Magnite and the quality models come from Sincera; 'the April-to-July change in partner wording shows the architecture was still moving at launch'. | Refute-mode verifier for stagwell-curate.json re-opened the cited source | The two releases do word the partners differently (FreeWheel 'will power' in April; FreeWheel and Magnite 'will provide' publisher relationships in July). Neither says the architecture changed. | supported with qualification | Partner dependence: supply runs through FreeWheel and Magnite, and the quality models come from The Trade Desk's Sincera. A change in any of these relationships would change what the product is, and the change in partner wording between the April and July releases suggests (inference) that the architecture was still moving at launch. | scope, unit, date or wording narrowed to what the source carries | STAGWELLCURATE risks 2 (partner dependence) |
| V-stagwell-curate-196 | Not documented for Curate; the 10-K mentions 'in some cases acting as an agent for a disclosed principal' in a company-wide risk factor on media commitments. | Refute-mode verifier for stagwell-curate.json re-opened the cited source | The phrase appears in the risk factor 'We are exposed to the risk of client defaults', among precautions against client payment default; the factor notes that certain Brands 'enter into contractual commitments with media providers'. | supported with qualification | not documented for Curate; the 10-K lists 'in some cases acting as an agent for a disclosed principal' among precautions in the company-wide risk factor 'We are exposed to the risk of client defaults', which concerns commitments to media providers made for clients | scope, unit, date or wording narrowed to what the source carries | STAGWELLCURATE data_table stagwell_curate_q1_who_decides agency_or_principal_role |
| G-01 | YMax: cohort = "exchange" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | Yieldmo's page describes operator-built YMax deal packages activated through the buyer's DSP as standard deal IDs and a managed service, not a tool for third parties to build deals on Yieldmo's exchange (decision-tree step 2) | structural | cohort -> "curator" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-02 | Seedtag NeuroX Curation: cohort = "curator" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | Seedtag's page describes a self-serve interface giving buyers pre-bid control of Seedtag's Neuro-Contextual Exchange to build deals (decision-tree step 1) | structural | cohort -> "exchange" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-03 | Adloox attention deals: cohort = "curator" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | Adloox sells attention segments at $0.15 CPM that buyers select while building their own deals in PubMatic's Media Console; a data supplier into cohort A tools, not a deal product (decision-tree step 4) | structural | cohort -> "listed" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-04 | Samba PMPs: cohort = "curator" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | samba.com's Audiences page of 2 November 2026 describes segments and names no PMP, curated deal or deal ID product; interest mark kept | structural | cohort -> "listed" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-05 | Sovrn Curated Deals: operating_entity = "Sovrn Holdings, Inc." | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | page footer reads Sovrn, Inc. | structural | operating_entity -> "Sovrn, Inc. (page footer); Sovrn Holdings, Inc. pending a filing" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-06 | Sovrn Curated Deals: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | verbatim coverage statement | structural | us_available -> "yes: sales coverage 'US / EU / CA / APAC' on the buyers page (CUR-U-S12)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-07 | WPP Media Premium Marketplace: operating_entity = "WPP Media (GroupM)" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | footer and launch article | structural | operating_entity -> "WPP Media Limited (site footer); GroupM at the February 2022 launch" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-08 | WPP Media Premium Marketplace: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | no later page found | structural | us_available -> "documented in 2022 only: 'currently available in EMEA and North America' (article of 24 February 2022, stream U); current status unclear" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-09 | dentsu Curate: aliases = [] | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | names as written in the 2020 release | structural | aliases -> ["Dentsu Curate", "Curate (dentsu)", "Supply Path Optimisation & Curated Marketplace offering"] | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-10 | dentsu Curate: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | release body | structural | us_available -> "no on the record opened: Hong Kong, South East Asia, Singapore, 'scaled to other APAC markets' (2020 release, stream U)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-11 | OpenXSelect: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | verbatim page statements | structural | us_available -> "yes: 'In the U.S., double your reach against qualified cross-channel audiences'; identity graph 'available in the U.S. only' (CUR-U-S03)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-12 | TL Direct: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | release of 10 June 2026 | structural | us_available -> "US nexus documented (New York dateline, New York address); explicit availability not stated (CUR-U-S07)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-13 | Brand+ Marketplace: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | page statements | structural | us_available -> "yes, regionally: 'across NAM, APAC and EU'; '150+ countries' (CUR-U-S14; interest mark)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-14 | Buyer Hub: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | page text | structural | us_available -> "unconfirmed: live for agencies from 31 August 2026; the US appears only in a sample query; exchange basis not established (stream U)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-15 | Roku Curate: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | site-level statement | structural | us_available -> "US implied by the site banner '125M+ daily active users in the US' (CUR-U-S30); page of 27 April 2026" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-16 | Stagwell Curate: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | release of 16 July 2026 | structural | us_available -> "yes: New York dateline, 'available immediately to clients across the Stagwell network' (CUR-U-S27)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-17 | Comscore-Certified Deal IDs: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | release | structural | us_available -> "US nexus (Reston, VA dateline); no availability statement (release of 10 June 2025, stream U)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-18 | SmartPMP: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | page address | structural | us_available -> "US nexus (New York office on the page); no availability statement (CUR-U-S15)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-19 | Adelaide High-Attention PMPs: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | page body | structural | us_available -> "'Available in more than 20 of the world's largest advertising markets'; US not named (CUR-U-S18); the attention paper records US availability of AU" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-20 | Multilocal curation: operating_entity = "Multilocal Media Ltd" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | footer | structural | operating_entity -> "Multilocal Media (legal form not confirmed on the page opened)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-21 | Multilocal curation: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | page | structural | us_available -> "site country selector lists UK and US; 'Always-On Global Coverage' (CUR-U-S21)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-22 | Experian Curated Deals: operating_entity = "Experian Marketing Services (Experian Information Solutions, Inc.)" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | footer | structural | operating_entity -> "Experian Information Solutions, Inc. (copyright line)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-23 | PubMatic Connect: aliases = ["Connect"] | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | product navigation names both | structural | aliases -> ["Connect", "Connect for Buyers", "Connect for Publishers"] | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-24 | PubMatic Connect: purchased_service = "Packaging and transaction of curated deals on PubMatic's exchange inventory via a single deal ID" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | browser read | structural | purchased_service -> "Inventory 'curated through our SSP and enriched by Connect's data signals', transacted via a single deal ID and activated through PubMatic's Activate or the buyer's DSP (browser read of the buyers page)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-25 | Microsoft Curate: purchased_service = "Building and selling curated deals on Microsoft Monetize's exchange inventory" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | route note | structural | purchased_service -> "Building and selling curated deals on Microsoft Monetize's exchange inventory; documentation of 21 October 2025 says the deals can be bought via Microsoft Invest and all major DSPs, a route statement that predates the announced end of Invest media buying on 28 February 2026" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-26 | ClearLine: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | no availability statement found | structural | us_available -> "not stated on the vendor pages opened by workstream U (3 November 2026)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-27 | Index Marketplaces: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | no availability statement found | structural | us_available -> "not stated on the vendor pages opened by workstream U (3 November 2026)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-28 | Maestro by Equativ: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | no availability statement found | structural | us_available -> "not stated on the vendor pages opened by workstream U (3 November 2026)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-29 | Nexxen Curated Marketplace: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | no availability statement found | structural | us_available -> "not stated on the vendor pages opened by workstream U (3 November 2026)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-30 | FreeWheel Curation Hub: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | no availability statement found | structural | us_available -> "not stated on the vendor pages opened by workstream U (3 November 2026)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-31 | Google Ad Manager Curation: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | no availability statement found | structural | us_available -> "not stated on the vendor pages opened by workstream U (3 November 2026)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-32 | Commerce Grid: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | no availability statement found | structural | us_available -> "not stated on the vendor pages opened by workstream U (3 November 2026)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-33 | Experian Curated Deals: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | no availability statement found | structural | us_available -> "not stated on the vendor pages opened by workstream U (3 November 2026)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-34 | Lotame Curated Marketplaces: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | no availability statement found | structural | us_available -> "not stated on the vendor pages opened by workstream U (3 November 2026)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-35 | Seedtag NeuroX Curation: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | no availability statement found | structural | us_available -> "not stated on the vendor pages opened by workstream U (3 November 2026)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-36 | IAS Curation: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | no availability statement found | structural | us_available -> "not stated on the vendor pages opened by workstream U (3 November 2026)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-37 | Microsoft Curate: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | no availability statement found | structural | us_available -> "not stated on the vendor pages opened by workstream U (3 November 2026)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-38 | YMax: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | no availability statement found | structural | us_available -> "not stated on the vendor pages opened by workstream U (3 November 2026)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-39 | Dstillery ID-free Deal IDs: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | no availability statement found | structural | us_available -> "not stated on the vendor pages opened by workstream U (3 November 2026)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-40 | Semasio curated deals: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | no availability statement found | structural | us_available -> "not stated on the vendor pages opened by workstream U (3 November 2026)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-41 | Samba PMPs: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | no availability statement found | structural | us_available -> "not stated on the vendor pages opened by workstream U (3 November 2026)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-42 | Adloox attention deals: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | no availability statement found | structural | us_available -> "not stated on the vendor pages opened by workstream U (3 November 2026)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-43 | PubMatic Connect: us_available = "unconfirmed; workstream U to establish from the vendor's documentation" | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | no availability statement found | structural | us_available -> "not stated on the vendor pages opened by workstream U (3 November 2026)" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-44 | Onetag Curation Platform: record = null | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | exchange candidate discovered late by workstream U on 3 November 2026; only the homepage was opened, so it is listed and not scored; a candidate for a later edition | structural | record -> "listed" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-45 | Permutive Curation: record = null | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | cohort B candidate discovered late by workstream U on 3 November 2026 from Google Ad Manager's curation partner list and Permutive's documentation; listed and not scored; a candidate for a later edition | structural | record -> "listed" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| G-46 | ClearLine: aliases = null | Workstream U (vendor-universe sweep) and its verifier, 3 November 2026 | Magnite's FY2025 10-K names the curation feature inside ClearLine 'the Magnite Curator Marketplace' (ClearLine profile, verified); an alias, not a new row | structural | aliases -> "add 'Magnite Curator Marketplace'" | register frozen after workstream U and before scoring (brief: cohorts and scope, sequencing) | data/product-register.json |
| S-001 | {"v": 1, "grade": "D", "verification_status": "verified"} | cell verifier: qualified | OpenXSelect page: headline "Curate, Control, and Perform. With Complete Transparency", a testimonial calling it "curation that is totally transparent", and only the company-wide "200k+ Direct publisher domains". No deal-level seller or domain list. Curated Packages page (page 1 of 7, re-opened 3 Nov 2026) names publishers package by package. Holiday Inventory: "such as Wurl, Plex, Scripps, Frndly TV, Hallmark Media, Xumo, Dish, AMC, and more". Reality TV and Fall Movie name Wurl, Plex, Sling/Dish, Tubi, Scripps and wedotv "and more". Three packages are Tubi-only (Tubi Drama, Tubi Comedy, Tubi Action, inventory "across Tubi's expansive library"). No domain or app list, no seller count; custom packages by contact. [17, 10] | supported with qualification | Curated Packages, an alias of the scored product in the register, names example publishers per package: partial composition, anchor 3 (grade D cap 3). | Curated Packages, an alias of the scored product in the register, names example publishers per package: partial composition, anchor 3 (grade D cap 3). | data/scores.json or data/matrix-extra.json: openxselect/composition_disclosure |
| S-002 | {"v": 1, "grade": "D", "verification_status": "verified"} | cell verifier: qualified | Text: buyers "can apply first-, second-, and third-party data"; "With more than 80 directly integrated data partners, customers can enhance their curation, SST, and measurement strategies"; "The OpenX identity graph is available in the U.S. only". The "Our Partners" section also shows a strip of 28 partner logos as images with no alt text. Their file names include Experian, Eyeota, Dstillery, Samba, 33Across, Captify, Resonate, Attain, AudienceTown, Affinity Answers, InMarket, Ibotta, ShareThis, Start.io, Adelaide, Scope3, Sightly and Chalice. None is tied to a segment, source type, collection method or permission. [17, 17] | supported with qualification | Rationale corrected: partners appear as logos, not 'none named'; the score of 1 stands because no provider is tied to any segment. | Rationale corrected: partners appear as logos, not 'none named'; the score of 1 stands because no provider is tied to any segment. | data/scores.json or data/matrix-extra.json: openxselect/data_provenance |
| S-003 | {"v": 3, "grade": "D", "verification_status": "verified"} | cell verifier: qualified | Curation page: AI bullets "Cuts planning time by up to 40%", "Analyzes 400K+ sites to surface premium inventory" and "Auto-generates optimal PMP/Deal configurations". A separate "Unparalleled Control" list has "Extensive data marketplace for all your targeting needs" and "Flexible pricing and custom business rules". CUR-E-S18 says deal creation is "initiated after you have set targeting criteria in the Media Planning module" but does not describe the criteria. Equativ's Create meta-deals article (updated 5 Aug 2026) names strategies: "semantic, contextual, data, player size, viewability targeting". No thresholds or cadence on any page opened. [114, 317, 15] | supported with qualification | The 3 rebased on the help centre's named targeting strategies; the cited marketing line alone was anchor 1. | The 3 rebased on the help centre's named targeting strategies; the cited marketing line alone was anchor 1. | data/scores.json or data/matrix-extra.json: maestro-by-equativ/selection_transparency |
| S-004 | {"v": "ne", "grade": null, "verification_status": null} | cell verifier: consistency review | [3] | structural | New cell from the help-centre article the verifier opened: a documented publisher list for small deals, no domains. | New cell from the help-centre article the verifier opened: a documented publisher list for small deals, no domains. | data/scores.json or data/matrix-extra.json: maestro-by-equativ/composition_disclosure |
| S-005 | {"v": 3, "grade": "C", "verification_status": "verified"} | cell verifier: qualified | CUR-E-S18 (updated 23 Apr 2026) gives the pricing model as "Market Floor, Custom Floor, or Fixed Price" and the deal settings as including "buyer seats or DSPs". It names the three deal types only in the meta-deals paragraph and does not tie them to pricing models. The Create meta-deals article (updated 5 Aug 2026) says "Private auction deals: first-price auctions with multiple competing buyers" and "Preferred deals: direct one-to-one deals between a publisher and a buyer", and lists the pricing models each deal type allows. The Create deals article (updated 15 Oct 2026) says market floor sets no floor, a custom floor is "the starting point for your bids in the auction", and the guaranteed model has "higher priority in DSPs compared to other models". None of the three pages documents request sampling. [114, 3, 317] | supported with qualification | Rationale corrected: the help centre documents the auction type, which the scorer had recorded as undocumented; the score of 3 stands. | Rationale corrected: the help centre documents the auction type, which the scorer had recorded as undocumented; the score of 3 stands. | data/scores.json or data/matrix-extra.json: maestro-by-equativ/auction_mechanics |
| S-006 | {"v": "ne", "grade": null, "verification_status": null} | cell verifier: consistency review | [3] | structural | New cell from the help-centre article the verifier opened: three named buying endpoints. | New cell from the help-centre article the verifier opened: three named buying endpoints. | data/scores.json or data/matrix-extra.json: maestro-by-equativ/activation_routes |
| S-007 | {"v": 2, "grade": "C", "verification_status": "verified"} | cell verifier: qualified | CUR-E-S18 matches the why: "Based on your access level (with or without margin), you can also apply additional costs or curator margins that influence the final media price paid to publishers". The Create deals article (updated 15 Oct 2026) goes further. Under market floor, "the deal price is calculated by taking the publisher's recommended bid price and adding the margin on top". Under custom floor and fixed price, the media price comes from "first subtracting the fixed costs from the deal floor price, and then subtracting the curator margin", with a worked example (deal floor EUR 0.59, 10% margin, media price EUR 0.531). An optional carbon-compensation contribution of "0.05 in your selected deal's currency" is offered (unit not stated). The page does not state the amount of the "fixed costs" or "applicable fees", Equativ's own charge, or whether the buyer sees the margin. The meta-deal [114, 3] | supported with qualification | The help centre documents the base and calculation of the curator margin: anchor 3, not 2. | The help centre documents the base and calculation of the curator margin: anchor 3, not 2. | data/scores.json or data/matrix-extra.json: maestro-by-equativ/pricing_transparency |
| S-008 | {"v": 1, "grade": "D", "verification_status": "verified"} | cell verifier: contradicted | The cited release (CUR-C-S09, 1 Oct 2025, opened 3 Nov 2026) says partners can onboard "their first-party data" or use data "through Nexxen's data management platform". That data includes insights from "Nexxen's proprietary audience discovery tool", "Nexxen's proprietary TV Audience targeting segments", which "blend data assets inclusive of automatic content recognition ('ACR') data", and third-party sources. Experian's quote says the product makes it easy to use "Audigent's first-party audiences and Experian's syndicated audiences". 33Across names its "Glossary" proprietary data. No per-segment collection method, permissions or legal basis. [254, 189] | contradicted | The cited release names source types and providers; the scorer's 'no provider' was contradicted by the verifier. | The cited release names source types and providers; the scorer's 'no provider' was contradicted by the verifier. | data/scores.json or data/matrix-extra.json: nexxen-curated-marketplace/data_provenance |
| S-009 | {"v": 3, "grade": "D", "verification_status": "verified"} | cell verifier: qualified | Product page (read 2026-11-03): 'Leverage TripleLift Audiences 3k+ off-shelf behavioral and contextual segments at no extra cost', 'Free 1P data onboarding and lookalike modeling', 'AI-powered custom audience creation', and 'TL Spark agentic intelligence layer analyzes demand patterns and market data to shape traffic, adjust floors'. No thresholds or update cadence. Facts and score 3 (grade D cap) hold. The page is a public marketing page, not an authenticated interface. [117, 117] | supported with qualification | Disclosure mode corrected from authenticated to public. | Disclosure mode corrected from authenticated to public. | data/scores.json or data/matrix-extra.json: tl-direct/selection_transparency |
| S-010 | {"v": 3, "grade": "D", "verification_status": "qualified"} | cell verifier: qualified | Curation Hub page: 'Standardized content metadata and audience signals integrated for scalable, differentiated packaging and targeting'; no thresholds or cadence. The phrase names input classes (content metadata, audience signals) rather than a selection rule, which puts anchor 3 at its lower edge; it is comparable in granularity to TL Direct's 'behavioral and contextual segments', so 3 is defensible under the grade D cap. The page is public marketing. [16] | supported with qualification | Disclosure mode corrected; score kept at the lower edge of anchor 3. | Disclosure mode corrected; score kept at the lower edge of anchor 3. | data/scores.json or data/matrix-extra.json: freewheel-curation-hub/selection_transparency |
| S-011 | {"v": 1, "grade": "D", "verification_status": "verified"} | cell verifier: qualified | Curation Hub page: 'Direct access to FreeWheel publisher partners and the marketplace chosen by major publishers' and 'reconciled reporting to maximize performance and safeguard profitability'. The page contains no seller or domain list, count or category and no explicit transparency statement. The quotes are accurate, but neither is an assertion of transparency about a deal's composition. [16] | supported with qualification | Anchor 1 requires a transparency assertion; none is on the page. n/e, as for TL Direct. | Anchor 1 requires a transparency assertion; none is on the page. n/e, as for TL Direct. | data/scores.json or data/matrix-extra.json: freewheel-curation-hub/composition_disclosure |
| S-012 | {"v": 1, "grade": "C", "verification_status": "verified"} | cell verifier: qualified | Agency page: 'Enter the Deal ID to see deal metadata including name and pricing' and the deal 'shows as "created by" by a curator'; no seller, publisher or domain list, count or category. The Ad Manager page's reporting statement ('To report on how curation was used by buyers to target your inventory, create an Interactive report') is reporting to publishers, and 'fully transparent to the buyer' concerns the fee. The why's facts are accurate, but none is an assertion of composition transparency to the buyer. [24] | supported with qualification | The only transparency statement concerns publisher reporting; anchor 1 is not met. n/e. | The only transparency statement concerns publisher reporting; anchor 1 is not met. n/e. | data/scores.json or data/matrix-extra.json: google-ad-manager-curation/composition_disclosure |
| S-013 | {"v": 3, "grade": "C", "verification_status": "verified"} | cell verifier: qualified | Ad Manager page lists the contextual signals, user identifiers (Device ID, Ad Manager third-party cookie, Hosted Match data) and secure signals, and a partner table of twelve entries for eleven companies: Adelaide; Audigent/Experian (Audience); Audigent/Experian (Contextual); DoubleVerify Inc.; Integral Ad Science; LiveRamp, Inc.; Multilocal Media; Permutive; PrimeAudience; Proximic by Comscore; SWYM.ai; System1. No collection method, intended use, permissions or legal basis per segment. Anchor 3 holds. [59, 24, 59] | supported with qualification | Partner count corrected; score kept. | Partner count corrected; score kept. | data/scores.json or data/matrix-extra.json: google-ad-manager-curation/data_provenance |
| S-014 | {"v": 3, "grade": "C", "verification_status": "qualified"} | cell verifier: Scored 3 although the anchor 3 text ('auction type documented, the rest not') names auction type as its requirement, and the why says auction type and priority are not documented. The 3 is an interpolation ('closest fit': sampling, floor, seat and net-of-fee rules documented) that the anchors do not provide. Anchor 2 ('mechanics documented only for one transaction type') arguably fits literally, since everything documented concerns the single curated demand-deal type. Grade C caps at 4, so the cap is not breached; the issue is the anchor text alone. | Scored 3 although the anchor 3 text ('auction type documented, the rest not') names auction type as its requirement, and the why says auction type and priority are not documented. The 3 is an interpolation ('closest fit': sampling, floor, seat and net-of-fee rules documented) that the anchors do not provide. Anchor 2 ('mechanics documented only for one transaction type') arguably fits literally, since everything documented concerns the single curated demand-deal type. Grade C caps at 4, so the cap is not breached; the issue is the anchor text alone. [22, 23, 266] | structural | Anchor 3 requires the auction type, which is not documented; mechanics are documented for one transaction type only: anchor 2. | Anchor 3 requires the auction type, which is not documented; mechanics are documented for one transaction type only: anchor 2. | data/scores.json or data/matrix-extra.json: commerce-grid/auction_mechanics |
| S-015 | {"v": 3, "grade": "D", "verification_status": "unverified"} | cell verifier: qualified | The publisher view is an image caption on a marketing screenshot: 'BuyerHub Assistant answering: What CTV inventory is available in the US by publisher?' ... 'with top publishers, eCPM, and ad requests'. Separately: 'Surface campaign performance and deal health directly at the trading desk.' Gating is implied by 'Role-based access', 'access controls' and 'Sales-assisted onboarding. No commitment required.' The page has no deal-level seller or domain list, no seller count and no identified example deal. [174] | supported with qualification | Rationale rebased on the page's own caption; score kept. | Rationale rebased on the page's own caption; score kept. | data/scores.json or data/matrix-extra.json: buyer-hub/composition_disclosure |
| S-016 | {"v": 2, "grade": "D", "verification_status": "unverified"} | cell verifier: qualified | Source types are named: 'deterministic (SDK-powered) and contextual signals' (heading 'Precise signals'), plus 'device, audience, or content metadata signals' and 'contextual classified brand-safe content'. The engine applying them is attributed to Verve in the first person: 'Our audience intelligence engine allows for better deal construction' (heading 'Curate smarter deals'). The page does not say whose SDK supplies the deterministic signals; it mentions Verve's 'SDK and OpenRTB connection' for publishers but does not tie it to the signals. It names no third-party data provider; GeoEdge, IAB OMSDK, Human, Pixalate, MMPs and Apple SKAN are listed as measurement partners, not data sources. Collection method, intended use, permissions and legal basis are absent, and nothing marks the contextual inputs as non-personal. [21] | supported with qualification | Same-standard correction (an under-score): peers whose own engine is the named provider score 3. | Same-standard correction (an under-score): peers whose own engine is the named provider score 3. | data/scores.json or data/matrix-extra.json: brand-plus-marketplace/data_provenance |
| S-017 | {"v": 3, "grade": "D", "verification_status": "unverified"} | cell verifier: qualified | The quoted line 'Latency-free ad delivery and scalability across NAM, APAC and EU.' exists, but it is the description of the 'Global data centers' feature under 'Technical features'. It is an infrastructure and delivery statement, not a market-availability statement, and the US is never named as a market. '150+ countries' is present ('Powering brands, agencies, and publishers in 150+ countries') above a logo wall with US-headquartered brands (Amazon, Microsoft, Target, Starbucks, McDonald's, Eli Lilly and others). The logos have no stated role, geography, period or tie to Brand+ deals, so they do not document US production use. 'U.S.A.' appears only in the footer trademark notice. [21, 21] | supported with qualification | A data-centre statement does not document US availability; anchor 1, consistent with TL Direct's dateline-only score. | A data-centre statement does not document US availability; anchor 1, consistent with TL Direct's dateline-only score. | data/scores.json or data/matrix-extra.json: brand-plus-marketplace/us_availability |
| S-018 | {"v": 3, "grade": "D", "verification_status": "verified"} | cell verifier: contradicted | June release: the deal IDs are enabled across PubMatic's premium supply 'including many of the top CTV publishers', a description of PubMatic's supply, not a per-environment statement of live execution. The 5 May 2025 release (CUR-F-S06, cited elsewhere in this row but not opened by workstream U per CUR-U-F18 limitations) names no channel and has Magnite promising 'Comscore-vetted inventory across various formats at scale' and 'omnichannel supply', and the product is built on MMX (web) and Video Metrix (video) rankings. So the advertised scope is not 'CTV only'; it is unspecified and broader, and the '1 of 1 advertised environment' denominator does not hold. [227, 244] | contradicted | The CTV phrase describes the SSP's supply, not documented live execution; n/e, as for Adelaide. | The CTV phrase describes the SSP's supply, not documented live execution; n/e, as for Adelaide. | data/scores.json or data/matrix-extra.json: comscore-certified-deal-ids/channel_execution |
| S-019 | {"v": 3, "grade": "D", "verification_status": "verified"} | cell verifier: qualified | Facts hold: May release 'initially available through Magnite's (NASDAQ: MGNI) supply-side platform (SSP)'; June release PubMatic quote 'By enabling Comscore-Certified Deal IDs in our SSP'. Neither release names a DSP or buying platform. Both named platforms are SSPs (sell-side hosts), not 'buying endpoints (DSPs and buying platforms)' as the dimension defines them. [227, 244] | supported with qualification | SSPs are not buying endpoints under the dimension definition; n/e, as for Dstillery. | SSPs are not buying endpoints under the dimension definition; n/e, as for Dstillery. | data/scores.json or data/matrix-extra.json: comscore-certified-deal-ids/activation_routes |
| S-020 | {"v": 3, "grade": "D", "verification_status": "qualified"} | cell verifier: qualified | The five names in the availability sentence (PubMatic, Microsoft Advertising (Xandr), TripleLift, Magnite, OpenX) are SSPs; Index Marketplaces is an SSP-side curation platform whose 7 January 2026 release presents AU as a signal for buyers and curators 'at the point of deal creation', names no deal ID or Adelaide PMP and no DSP ('through the DSP of choice'); Index's 2024 launch release lists Adelaide among launch partners only; Google Ad Manager's help page lists Adelaide for 'contextual information and inventory packaging' and says generically 'A curation partner shares a Deal ID with the buyer', naming no DSP. None of these is a buying endpoint. The DSPs are named on the same Adelaide page in the high-AU PMP paragraph: Teads clients saw results 'using high-AU PMPs across Google Display & Video 360, The Trade Desk, and Yahoo DSP' (the page presents these as the same high-AU PMP category [60, 187, 204, 59] | supported with qualification | Score kept on the correct basis: three DSPs, not the seven sell-side hosts. | Score kept on the correct basis: three DSPs, not the seven sell-side hosts. | data/scores.json or data/matrix-extra.json: adelaide-high-attention-pmps/activation_routes |
| S-021 | {"v": 3, "grade": "D", "verification_status": "verified"} | cell verifier: qualified | Google's post of 4 November 2024 (Peentoo Patel) names IAS among eight curation partners under the heading 'Data and inventory curation made easy for agencies'; no DSP is named under that heading. The sentence 'currently available for Preferred Deals and Private Auctions via Display & Video 360 and Yahoo, with support for Programmatic Guaranteed and other DSPs coming in the months ahead' sits under the separate heading 'Programmatic buying streamlined for agencies' and describes the single deal ID agencies 'negotiate directly with publishers', not curated partner deals. The link to curated deals comes from Google's Authorized Buyers page 'Curation for agencies (Beta)' (https://support.google.com/authorizedbuyers/answer/15282785): the agency subscribes to the curated auction package's Deal ID in its Agency Seat, and curation applies targeting 'to single deal ID for all DSPs integrated wit [196, 59, 187] | supported with qualification | Endpoints rest on partner documentation and an inference chain, on a Beta feature: anchor 2, availability beta. | Endpoints rest on partner documentation and an inference chain, on a Beta feature: anchor 2, availability beta. | data/scores.json or data/matrix-extra.json: ias-curation/activation_routes |
| S-022 | {"v": 2, "grade": "D", "verification_status": "qualified"} | cell verifier: qualified | Homepage: 'proprietary database, powered by our Media Quality Index' and '62,000 audience data points' with no source; neither the homepage nor the cureon page states a data provider, collection method or consent. The 29 September 2026 release as reproduced by TechIntelPro (PR Newswire mark) is a joint announcement ('Sovrn, a publisher technology and data company, and Multilocal ... today announced a partnership'). It quotes executives of both companies and states that Sovrn's commerce business 'adds consumer purchase and intent signals from activity across more than 50,000 retailers' for PMP deals on Sovrn's exchange. It gives no collection method, consent or legal basis. [129, 257] | supported with qualification | The external provider and source type are named in a joint release: anchor 3, on the same standard as Nexxen. | The external provider and source type are named in a joint release: anchor 3, on the same standard as Nexxen. | data/scores.json or data/matrix-extra.json: multilocal-curation/data_provenance |
| S-023 | {"v": 3, "grade": "D", "verification_status": "verified"} | cell verifier: qualified | Same structure as the IAS cell. Google's 4 November 2024 post names Multilocal among eight partners under the curation heading, which names no DSP; the 'Display & Video 360 and Yahoo' sentence sits under 'Programmatic buying streamlined for agencies' and describes the single agency deal ID negotiated with publishers. The Authorized Buyers page 'Curation for agencies (Beta)' links curated deal IDs to 'all DSPs integrated with agency deals' without naming any. Google's Ad Manager help page (©2026, Beta) lists 'Multilocal Media' as a curation partner. cureon page: 'Request a PMP and have it delivered directly to your DSP seat', with no DSP named; the homepage refers only to 'integrations across multiple SSPs, DSPs, and DMPs'. Neither the Sovrn release nor Index's 2024 release names a DSP. [196, 59, 256] | supported with qualification | As IAS Curation: partner documentation and an inference chain, Beta feature. | As IAS Curation: partner documentation and an inference chain, Beta feature. | data/scores.json or data/matrix-extra.json: multilocal-curation/activation_routes |
| S-024 | {"v": 3, "grade": "D", "verification_status": "verified"} | cell verifier: qualified | YMax page: 'Every URL is visible before you commit'; packages 'at the page level with full URL visibility' and makes them 'reviewable before you activate' (FAQ); 'Performance is tracked by individual page URL, by supply segment, and by creative variant'. No seller or publisher list, seller count or example deal; publishers appear only as 'premium publishers'. [11] | supported with qualification | Rationale corrected (anchor 4 needs seller and domain lists; disclosure mode public); score kept within the grade D cap. | Rationale corrected (anchor 4 needs seller and domain lists; disclosure mode public); score kept within the grade D cap. | data/scores.json or data/matrix-extra.json: ymax/composition_disclosure |
| S-025 | {"v": 3, "grade": "D", "verification_status": "verified"} | cell verifier: qualified | YMax page: 'connects page-level inventory intelligence with...' and 'page-level supply intelligence'. Companion Audiences FAQ: 'Start with any seed — your first-party data, YMax contextual visitors, site visitors, or converters'. Every impression is read for 'adaptive taxonomy signals', 'environment signals (daypart, geo, device, connection type)' and 'attention signals (swipes, tilts, touches, scrolls, engagement pattern)'. No third-party provider, collection method, intended use, permissions or legal basis. [11, 11] | supported with qualification | Rationale rebased on the cited page; score kept. | Rationale rebased on the cited page; score kept. | data/scores.json or data/matrix-extra.json: ymax/data_provenance |
| S-026 | {"v": 1, "grade": "D", "verification_status": "verified"} | cell verifier: contradicted | The 11 August 2022 launch post on the tag page says ID-free Custom AI 'uses AI to predict the likelihood of conversion based on privacy-safe behavioral signals like URL, geographic area and time of day'. It also contains 'ID-free is neither a new identifier nor is it contextual targeting'. So the post does name the source type of the model's inputs (bid-request signals: URL, geography, time of day) and the provider (Dstillery's own Custom AI). It does not say what data trains the model, and it gives no collection method, intended use, permissions or legal basis. The FAQ says 'the same audiences can be built as with standard ID-free via API', which ties the deal IDs to these ID-free audiences. [61, 70] | contradicted | The launch post names source type and provider; the scorer's 'no source type' was contradicted. | The launch post names source type and provider; the scorer's 'no source type' was contradicted. | data/scores.json or data/matrix-extra.json: dstillery-id-free-deal-ids/data_provenance |
| S-027 | {"v": 1, "grade": "D", "verification_status": "qualified"} | cell verifier: contradicted | In the 11 August 2022 launch post, Bryan Presti (Senior Ad Ops Specialist, KORTX, a Dstillery client) is quoted. His quote opens 'With Deal IDs, we can connect without an external API integration for programmatic ad buys'. The figures come in a separate sentence: 'Since we began using ID-free to help drive more purchases to Tropical Smoothie in March', results gave 'a $14.34 CPA for Display and $62.33 for Video', which he says rival or beat earlier ID-based campaigns. Those figures are for ID-free targeting from March 2022, before the deal IDs launched in August 2022. They are not a measurement of the deal product. No method, sample or period end is given. [129] | contradicted | Evidence is not inherited between products (cell rules); the figure predates the scored product. | Evidence is not inherited between products (cell rules); the figure predates the scored product. | data/scores.json or data/matrix-extra.json: dstillery-id-free-deal-ids/quality_evidence |
| S-028 | {"v": 2, "grade": "D", "verification_status": "qualified"} | cell verifier: qualified | The homepage was fetched on 3 November 2026: banner 'Semasio is now part of Samba', footer 'Copyright 2025'. The quotes are accurate, but they sit in other parts of the page. 'Target pages, not people' heads the Contextual Targeting product card. 'Replicates behavior stored on external ID sources' sits in a platform-level list ('Connect to all customers', next to 'Refines continuously based on Billions of page impression observations'). The HypertailPMP card says only 'Reach your ideal consumer in long-tail media spaces', 'Efficiency + Reach' and 'Engage your ideal target audience without waste', and names no data or source type. The HypertailPMP service page (/services/hypertail-pmp) returned HTTP 404 again on 3 November 2026. The register's 'PMPs built on Semasio contextual and audience data' is the provisional seed's definition, and no opened source supports it. [128] | supported with qualification | Access failure and inheritance from other products; n/e under the disclosure-mode rule. | Access failure and inheritance from other products; n/e under the disclosure-mode rule. | data/scores.json or data/matrix-extra.json: semasio-curated-deals/data_provenance |
| S-029 | {"v": 3, "grade": "D", "verification_status": "verified"} | cell verifier: verified | Release dated 15th September 2020. It contains 'we can curate supply direct from publishers', 'media consolidation into a central marketplace', 'eliminating non-essential pass throughs in the value chain', 'spearhead a new age of Supply Path Optimisation & Curated Marketplace offering in this region' and 'With added addressability through data partnerships'. It states no criterion, threshold or cadence. The quotes in the why are accurate. Whether this is 'logic named' (anchor 3) is disputed in anchor_issues. [6] | supported as stated | Sourcing language is not selection logic; same standard as WPP (1). | Sourcing language is not selection logic; same standard as WPP (1). | data/scores.json or data/matrix-extra.json: dentsu-curate/selection_transparency |
| S-030 | {"v": 1, "grade": "D", "verification_status": "verified"} | cell verifier: qualified | Sam Tan (Xandr) is quoted: 'increased control over who they buy from and the fees taken at each step in the supply chain' and 'drive increased transparency across media, technology and costs for buyers and sellers'. The release also says 'eliminating non-essential pass throughs in the value chain', and Naryani says 'curate the best quality inventory at the best price for our clients'. None of these gives a charge type, base, amount, payer, collector or period. The score of 1 stands, but the Xandr fee line is not the only pricing statement. [6] | supported with qualification | Rationale rebased on the release's wording; score kept. | Rationale rebased on the release's wording; score kept. | data/scores.json or data/matrix-extra.json: dentsu-curate/pricing_transparency |
| S-031 | {"v": 3, "grade": "D", "verification_status": "verified"} | cell verifier: qualified | S03 (the advertised scope) and S15 (the Agence 79 video campaign) returned HTTP 403. Adloox (opened, 17 Jul 2026): 'Select Desktop & Mobile Web' and 'Select Mobile, Desktop, Tablet devices (avoid CTV)'. Comscore release (opened, 10 Jun 2025): expands 'Comscore-Certified Deal IDs' in 'PubMatic's premium sell-side supply', 'including many of the top CTV publishers', with the PubMatic quote 'By enabling Comscore-Certified Deal IDs in our SSP'. It gives no launch date or live status and does not name Connect. [223, 12, 227, 119, 224, 228] | supported with qualification | The CTV evidence is weaker than first stated; the score is supported by web display and video. | The CTV evidence is weaker than first stated; the score is supported by web display and video. | data/scores.json or data/matrix-extra.json: pubmatic-connect/channel_execution |
| S-032 | {"v": 3, "grade": "C", "verification_status": "qualified"} | cell verifier: qualified | S07 (Last Updated October 26, 2026): 'You can target publishers or deals, not both'; 'You can add up to 10,000 domains and app bundles'; 'Lists cannot be used to exclude domains and app bundles' (predefined lists are for inclusion only, and exclusions are typed into the Exclude section, which 'takes priority'); 'up to four segment rule sets' to include and 'one segment rule set' to exclude; device types include three CTV types (Connected device, Connected TV, Set-top box); In-App and Web; Banner, Native, Video; countries, states and provinces, metro codes, 'Zip and postal codes (Beta)'; dayparting; 'minimum viewability probability'; 'All Tier 1 and Tier 2 categories are supported'; STV content genres; VCR (Beta), CTR (Beta), Custom Key Values (Beta), 'Targeting a deal in a Marketplace Package (Beta)', and also 'Networks and channels (Beta)' for STV. Blank publisher field: 'new publishers [14, 198, 109] | supported with qualification | Wording corrected; score kept. | Wording corrected; score kept. | data/scores.json or data/matrix-extra.json: index-marketplaces/selection_transparency |
| S-033 | {"v": 3, "grade": "C", "verification_status": "verified"} | cell verifier: qualified | S11 (Last Updated October 21, 2026): 49 vendor rows with markets (page counter reads '0 vendors'). S10 (May 04, 2026): 'LiveRamp Data Marketplace segments can only be used to target users in the United States'; Index 'reserves the right to update targeting for any deals' using them; no consent or legal-basis fields. S07: 'To target a Data Vendor's segment, the vendor must first grant you permission to use it.' However, S11's vendor descriptions do carry vendor-level collection and consent language for several vendors, e.g. AlikeAudience 'built exclusively from privacy-safe, consumer-consented data', Samba TV 'first-party data from millions of opted-in connected televisions', Nordic Data Resources 'Every segment is GDPR- and CCPA-compliant', Digiseg 'built on probabilistic household data', Lotame 'declared and demonstrated behavioral data from publisher partners'. [94, 240, 14] | supported with qualification | Wording corrected; score kept. | Wording corrected; score kept. | data/scores.json or data/matrix-extra.json: index-marketplaces/data_provenance |
| S-034 | {"v": 4, "grade": "C", "verification_status": "qualified"} | cell verifier: qualified | S07 documents the Web and In-App environments, Banner, Native and Video formats, three CTV device types and STV content genres as package settings. S04 (Jan 23, 2024): 'across various channels including streaming TV, display, video, and mobile app', 'currently in limited beta testing'. S02 (2026 product page): 'Streaming TV', 'Agentic Cultural Optimization for Premium CTV', 'Online Video' only as an entry in the site's Products list; no audio, DOOH or display. No page documents delivery per environment; the settings are configuration options. [14, 200, 187, 203, 318, 258] | supported with qualification | Same standard settled: documented product settings count as live availability under the grade C cap; the score of 4 stands and Microsoft's rationale is aligned. | Same standard settled: documented product settings count as live availability under the grade C cap; the score of 4 stands and Microsoft's rationale is aligned. | data/scores.json or data/matrix-extra.json: index-marketplaces/channel_execution |
| S-035 | {"v": 3, "grade": "D", "verification_status": "qualified"} | cell verifier: qualified | S16: logos with alt text 'TAG Brand Safety Certified', 'TAG Certified Agains Malware' (sic), 'TAG certified against fraud', 'Logo for iab gold standard certified', HUMAN, IAS, The Media Trust, IAB Tech Lab, under 'Industry certifications and collaboration'; no year, scope or product. S17: under 2019, 'Earned "TAG Platinum" status', 'we achieved Platinum status with the Trustworthy Accountability Group'. TAG registry (tagtoday.net/registry): opened, but its entries did not render to the fetch tool, so no Index Exchange entry could be read (not worked around). [19, 205, 319] | supported with qualification | A vendor's display of seals is a vendor assertion; without the certifier's record it is self-attestation (anchor 1). | A vendor's display of seals is a vendor assertion; without the certifier's record it is self-attestation (anchor 1). | data/scores.json or data/matrix-extra.json: index-marketplaces/assurance |
| S-036 | {"v": 3, "grade": "C", "verification_status": "qualified"} | cell verifier: qualified | S03 (ms.date 2026-02-18): universal categories 'defined by Microsoft Advertising. When Microsoft Advertising reviews inventory, we apply these categories based on the inventory's content'; seller categories; seller and exposed-publisher include/exclude; App & Web (default), App Only, Web Only; allowlists and blocklists, 'by default, the AppNexus Blocklist will always be applied'; segment and geography targeting; 'Thresholds (Viewability Threshold or Completion Rate Threshold) create a hard limit ... Any impressions with a predicted view or completion rate below the threshold won't be matched'. No category review cadence or deal update cadence. V-S04 (April 14, 2026): Max Reach lets curators 'expand their first-party segments to similar users' and adds 'contextual expansion, allowing curators to extend beyond addressable user IDs'; it 'leverages Microsoft first-party data and large langua [110, 111, 320, 110] | supported with qualification | Wording corrected; score kept. | Wording corrected; score kept. | data/scores.json or data/matrix-extra.json: microsoft-curate/selection_transparency |
| S-037 | {"v": 3, "grade": "C", "verification_status": "qualified"} | cell verifier: qualified | S03: 'Cross-device enhanced by Microsoft will be automatically enabled for your curated deal at no cost if you target a segment. If you prefer not to benefit from cross-device ... contact your account representative to opt out'; universal categories applied by Microsoft from inventory content; seller-defined categories. S01: curators use 'audience data' among their assets. V-S04: Max Reach 'leverages Microsoft first-party data and large language models (LLMs) to identify relevant URLs'. [56, 110, 320, 110] | supported with qualification | Wording corrected; score kept. | Wording corrected; score kept. | data/scores.json or data/matrix-extra.json: microsoft-curate/data_provenance |
| S-038 | {"v": 4, "grade": "C", "verification_status": "qualified"} | cell verifier: qualified | S03: 'Auction Type is always set to "Open Auction" for curators'; Dynamic CPM 'competes in either a first or second-price auction at the price bid by the buyer'; 'Use Market Price' falls back to open-auction pricing if seller YM floors are unavailable; 'You cannot change the selected buyer after the curated deal line item is created and saved.' S02 (2025-10-21): 'Curated transactions are not guaranteed deals and will not override any seller account configurations, including ad quality settings or floor price settings'; bids 'submitted to your final auction net of such fees or charges unless you agree otherwise with the applicable curator'; 'buyers can still bid on seller inventory via open exchange'. S04: Fixed / Standard / Market floor types; floors 'above, at, or even below a seller floor'. V-S01 (Jan 21, 2026): 'a single deal that is automatically eligible for all DV360 Partner IDs as [110, 111, 112, 113, 321, 110] | supported with qualification | Wording corrected; score kept within the grade C cap. | Wording corrected; score kept within the grade C cap. | data/scores.json or data/matrix-extra.json: microsoft-curate/auction_mechanics |
| S-039 | {"v": "ne", "grade": null, "verification_status": null} | cell verifier: consistency review | [56] | structural | Rationale aligned with the Index standard: n/e rests on the missing advertised scope. | Rationale aligned with the Index standard: n/e rests on the missing advertised scope. | data/scores.json or data/matrix-extra.json: microsoft-curate/channel_execution |
| S-040 | {"v": 3, "grade": "D", "verification_status": "qualified"} | cell verifier: qualified | SmartPMP page: 'SmartPMPs rely on 100% 1st party audience data, never 3rd party cookies.' Homepage and CognitivePMP page: 'Audigent's verified, opt-in data drives monetization for premium publisher and data partners including: Condé Nast, TransUnion, Warner Music Group, Penske, a360 Media, Fandom and many others.' DV360 post: 'Audiences are sourced from first-party publisher data', Experian 'deterministic purchase, intent, behavioral, and lifestyle signals' and The Weather Company. Platform Privacy Policy (last updated 2 September 2026; Predictive Pop, Inc. dba Audigent): cookies, beacons, tags, mobile SDKs; ID matching of 'publisher cookies, Audigent cookies, third party data provider cookies'; consent for segment building, legitimate interests otherwise; SmartPMP and curated deals not mentioned in the body. Source type and providers are named, so anchor 3 holds at grade D. [193, 116, 192, 190, 322] | supported with qualification | Wording corrected; score kept. | Wording corrected; score kept. | data/scores.json or data/matrix-extra.json: audigent/data_provenance |
| S-041 | {"v": 3, "grade": "D", "verification_status": "qualified"} | cell verifier: qualified | Release (14 Aug 2024): custom audiences from 'cookieless publisher data'; 'Lotame Analytics'. Capability page: 'AI-powered consumer insights' (the phrases 'cookieless publisher data' and 'market-level consumer data' are not on this page). PHD PDF: 'Lotame leveraged market-level consumer data and unique audience insights from its Analytics suite'; 'No modeling was done for the data used in the campaign'; 'Data is collected via loyalty / credit cards and online actions across publisher sites'. Services Privacy Notice ('Last modified on November 19, 2025'): pixels, tags, cookies, partner SDKs and APIs; device IDs, IP addresses, third-party IDs and emails come from business customers and data partners; EU legal basis 'consent'; TCF ID 95; 'curation' appears only in the site navigation, not the notice body. [13, 209, 125, 208] | supported with qualification | Wording corrected; score kept. | Wording corrected; score kept. | data/scores.json or data/matrix-extra.json: lotame/data_provenance |
| S-042 | {"v": 2, "grade": "D", "verification_status": "qualified"} | cell verifier: qualified | Google launch post (4 November 2024): 'Curation already supports Audigent, IAS, Liveramp, Lotame, Multilocal, Permutive, PrimeAudience and Scope3', packages 'currently available for Preferred Deals and Private Auctions via Display & Video 360 and Yahoo, with support for Programmatic Guaranteed and other DSPs coming in the months ahead'. The post does not call the feature Beta. Lotame capability page: 'is available to DSPs for scaled audience targeting via private marketplace (PMP)', no DSP named. Google Ad Manager help 'Configure curation settings (Beta)' (current, 2026): its curation-partner table lists Adelaide, Audigent/Experian (Audience), Audigent/Experian (Contextual), DoubleVerify, Integral Ad Science, LiveRamp, Multilocal Media, Permutive, PrimeAudience, Proximic by Comscore, SWYM.ai and System1; 'Lotame' does not appear anywhere on the page. The Authorized Buyers help page 'Cura [196] | supported with qualification | Basis of the availability label corrected; score kept. | Basis of the availability label corrected; score kept. | data/scores.json or data/matrix-extra.json: lotame/activation_routes |
| S-043 | {"v": 3, "grade": "D", "verification_status": "qualified"} | cell verifier: verified | Release: 'The marketplace provides an always-on, cross-channel pool of pre-vetted inventory across CTV, online video, display and audio' (present tense); opening paragraph 'higher-quality digital advertising inventory across streaming TV, video, display and audio'; 'Stagwell Curate is available immediately to clients across the Stagwell network'; partner sentence future tense: 'Partners including Freewheel and Magnite will provide access to deep, direct publisher relationships across streaming, video, web and audio'. Digiday narration: Stagwell 'is set to launch Stagwell Curate', no date. April 2026 release (ExchangeWire PressBox, 27 April 2026): CTV, premium video, streaming and linear TV only, all future tense ('will integrate', 'will power Stagwell Curate'), no display or audio, Magnite absent. FreeWheel Curation Hub page: CTV-centred ('premium CTV', video, a 'Programmatic Linear TV'  [5, 71, 72, 16, 5] | supported as stated | Same standard as TL Direct, recorded with the counterevidence; score kept at the grade D cap. | Same standard as TL Direct, recorded with the counterevidence; score kept at the grade D cap. | data/scores.json or data/matrix-extra.json: stagwell-curate/channel_execution |
| FINAL-01 | H4: holds for Microsoft Curate and Index Marketplaces, fails for Google packages built on Preferred Deals, unresolved for twelve | Outside AI review of the published edition: recode auction access by transaction type (Index exclusivity undocumented; Index fixed price has priority; Google Preferred Deals are a curator choice) | Index's 'Create a Marketplace Package' page (last updated 26 October 2026), read by a fetch agent on 4 November 2026: the creator selects First or Fixed with no default; the Auction type definition calls a fixed price auction 'an auction that has priority over all other auction bid types'; no exclusivity default. The review said 'glossary'; the definition is an on-page tooltip. Google's Preferred Deal is pre-auction by type but bundled at the curator's choice; the register's own taxonomy codes a curator's choice as configurable. [14, 24, 250] | supported with qualification | H4 recoded: holds for Microsoft Curate, unresolved for the other fourteen; Google and Index documented as configurable routes to priority. New claim CUR-FINAL-F01; CUR-E-F35 and CUR-INDEXMARKETPLACES-F05 reworded through the claims overlay. | review finding, checked at its primary source before any change | chapters 00, 04, 12, 13, 14; data/claims-extra.json; data/claims-overlay.json |
| FINAL-02 | Figure showed the first coding (62 unknown) while the text gave 61 after the verifier | Outside AI review of the published edition: reconcile the access figure's 62 unknown entries with the corrected 61 | The verifier's three Commerce Grid re-codings (CUR-E-F34) and the Index priority entry (CUR-FINAL-F01) are now applied before counting. [22, 23, 14] | supported as stated | scripts/derivations.mjs applies four corrections to the register before counting; the figure and text show 60 of 85 unknown (71%), priority documented for 4 rows. | review finding, checked at its primary source before any change | scripts/derivations.mjs; scripts/paper-figures.mjs; chapter 04 |
| FINAL-03 | H7: two deployments name a customer, a period, an executed decision and a transaction context | Outside AI review of the published edition: replace the exact count of two agent deployments; the published accounts leave the seller-side selection decision unclear; screen PubMatic's Havas-Telefónica campaign | PubMatic release of 9 June 2026 (read in the browser pane): agents 'autonomously selected CTV inventory' for Havas's Movistar campaign in Spain; no dates, no named agent. PubMatic's 5 January 2026 release says AgenticOS 'recommended campaign tactics, executed the buy' for Clubtails, with no inventory decision; its 27 April 2026 release claims 30 live campaigns and names Amnet/Interbev and Abovo Maxlead. The Magnite/LG accounts say the agents 'set up and activated' the plan. The agent stream had not screened PubMatic's 2026 releases. [313, 307, 312] | supported as stated | H7 stays supported with qualification, now resting on one platform-level vendor record (Havas/Movistar); the Geloso and LG records are coded as agent-led activation without a documented selection. New claims CUR-FINAL-F02 to F04; CUR-J-F21 reworded; agent tables, figure title and footer updated. | review finding, checked at its primary source before any change | chapters 00, 10, 12, 13, 14; plain.md; scripts/paper-figures.mjs; data/sources-extra.json |
| FINAL-04 | Review cited node 12976 for the June 2026 Havas campaign | Outside AI review of the published edition: citation for the Havas-Telefónica case (investors.pubmatic.com/node/12976/pdf) | Node 12976 is PubMatic's AgenticOS launch release of 5 January 2026 and does not mention Havas or Telefónica; the campaign is in PubMatic's release of 9 June 2026. [313, 312] | contradicted | The finding is applied with the correct source (CUR-FINAL-S01); the January release is registered separately (CUR-FINAL-S03) for the Geloso wording. | review finding, checked at its primary source before any change | data/sources-extra.json |
| FINAL-05 | Chapter 11 reported band sensitivity but not the evidence grades or a grade-C diagnostic | Outside AI review of the published edition: put the Deal Quotient evidence limits beside the bands (7 of 21 banded; 92 of 122 counted cells grade D; all seven bands move; pricing the sole Standing cell in four exchange rows; grade-C diagnostic leaves four admitted and no bands) | Recomputed from data/capability-matrix.json with build/quotient.mjs: banded 7; counted 122 (D 92, C 26, A/B 4); band-sensitive 7; pricing-only Standing 4 (Commerce Grid, Maestro, Microsoft Curate, TL Direct); C-or-better admits Maestro, Google, Index, Microsoft and bands no cohort. Every number in the review reproduced. [314] | supported as stated | New section 'What the bands rest on' in chapter 11, sentences in its standfirst and summary and in the abstract; derived checks dq_evidence_* added; claim CUR-FINAL-F05. The baseline method is unchanged. | review finding, checked at its primary source before any change | chapters 00, 11; scripts/derivations.mjs; data/claims-extra.json |
| FINAL-06 | Admission rule (live and Beta may be scored) not restated beside the scores; three admitted rows rest on unconfirmed availability | Outside AI review of the published edition: state whether unconfirmed products can enter the matrix | scoring-rules.json: live and beta products may be scored, announced and retired are listed. Google Ad Manager Curation is Beta, Stagwell Curate is live on its operator's claim, WPP Media Premium Marketplace has no page newer than 9 February 2023. Without the Google row the exchange field DQ falls from 46.7 to 45.4 and Microsoft Curate moves from Mid to Strong. [314] | supported as stated | Chapter 11 states the rule, names the three rows and gives the no-Beta sensitivity. | review finding, checked at its primary source before any change | chapter 11 |
| FINAL-07 | Hero: 'are now curated'; plain edition: 'Most of the time, that slot was put in a pack' | Outside AI review of the published edition: bring the headline and plain edition back to the measured units (Jounce's 72.7% is offered auctions, March 2025, global; 'now'; 'most of the time' a slot is packaged) | Jounce's figure is 72.7% of available RTB auctions in March 2025 on a global sample (CUR-B-F18). [310] | supported as stated | Hero now says 'By Jounce Media's global count for March 2025 ... were curated'; the plain edition drops the frequency claim and dates the count. | review finding, checked at its primary source before any change | paper.config.json; plain.md |
| FINAL-08 | Plain edition: 'One shows a list of sites for small packs' | Outside AI review of the published edition: Equativ documents publisher names for small deals, not a site list | Equativ's Create deals page shows a publisher-name table for deals of 50 publishers or fewer (CUR-NFA-S01). [3] | supported as stated | Plain edition: 'One names the publishers in small packs'. | review finding, checked at its primary source before any change | plain.md |
| FINAL-09 | 'No party without a stake has measured a named curated deal'; 'None publishes the amounts it charges' | Outside AI review of the published edition: use 'we found no independent matched evaluation of a named curated deal' and scope the fee claim to published rates for admitted products | CUR-B-F31 bounds the absence to the sources screened; CUR-H-F30 concerns published rate cards and amounts. [310] | supported as stated | Abstract, hero, chapter 7 and the plain edition use the bounded wording; fee sentences say charge rates or amounts of the admitted products. | review finding, checked at its primary source before any change | chapters 00, 06, 07; paper.config.json; plain.md |
| FINAL-10 | Chapter 6: 'No product discloses every charge with amount and base, so the supported rule is not met' | Outside AI review of the published edition: apply H2b's registered charge-type-and-basis test separately from Q3's amount-or-rate test | H2b supported rule: a majority disclose every applicable charge type with its basis; Q3 counts amount or rate and base. Six products score 3 (types and bases); the DSP take is outside every sell-side document (CUR-H-F31). | supported as stated | Chapter 6 and the appendix test H2b on types and bases only; the code stays unresolved on the registered rule. | review finding, checked at its primary source before any change | chapters 06, 14 |
| FINAL-11 | Interest mark shown only in the web matrix; no relationship flag | Outside AI review of the published edition: carry Lotame's interest flag into the score and matrix exports and add a relationship flag for Index Marketplaces and Magnite ClearLine Activate (Samba named as a data provider) | Index's List of Data Vendors (last updated 21 October 2026) lists Samba TV and Lotame; Magnite's blog of 9 September 2026 names Samba and Lotame among ClearLine Activate's direct third-party data partners (CUR-U-F11). OpenXSelect's partner logos could not be read as text on 4 November 2026, so OpenXSelect carries no flag. [94, 108] | supported as stated | Relationship field on Index Marketplaces and ClearLine; interest and relationship columns in capability-matrix-wide.csv and dq.csv; a second mark and a legend line in the matrix; sentence in chapter 11. | review finding, checked at its primary source before any change | build/build-matrix.mjs; build/instruments.mjs; build/make-csv.mjs; data/profile-overlay.json; chapter 11 |
| FINAL-12 | Chapter-end spill pages in chapters 4, 7, 10 and 12 and the downloads list | Outside AI review of the published edition: six sparse spill pages in the PDF | Measured with pypdf after the rebuild. | supported with qualification | The rewritten chapters 4 and 10 no longer spill; chapter 12 was tightened to remove its spill page. Spill pages remain at the end of a chapter 7 section, at the end of chapter 7 and at the end of the downloads list, plus two table continuations. | review finding, checked at its primary source before any change | chapter 12; dist PDF |
| FINAL-13 | Profile: 'a stated default of no floor or priority advantage'; SWOT: 'net-of-fee bidding with no stated priority' | Outside AI review of the published edition: Index profile and SWOT lines repeat the no-priority reading | Follows from FINAL-01 (CUR-FINAL-F01). [14] | supported as stated | Both lines corrected to say priority depends on the auction type the creator chooses. | review finding, checked at its primary source before any change | evidence/profiles/index-marketplaces.json; evidence/swot/index-marketplaces.json |
| FINAL-14 | Abstract opens 'The public record shows how much of the auction curation occupies, not what a buyer is buying.' | Outside AI review of the published edition: lead with 'A buyer can see that a deal is curated; the public record rarely shows what was selected, what each party charged, or whether the deal delivered' | Editorial suggestion; the abstract and chapter 12 already lead with this point. | structural | Not adopted; the existing lead sentence makes the same point. | editorial suggestion | none |



## Appendix: verdicts, vendor universe and downloads


*The registered hypotheses and questions and how each was coded, the reasoned sample of companies and products screened, and the published data files behind the paper.*

### How each registered test was coded

Each hypothesis and question was registered on 2 November 2026 with its decision rules, before any research agent ran. The verdicts below apply those rules to the evidence read by the 3 November 2026 cutoff. Unresolved means the public sources screened do not settle a test; it does not mean the hypothesis is refuted.

| Test | Statement (short) | Verdict | Where |
|---|---|---|---|
| H1 | The US share of trading through multi-publisher curated deals rose from 2023 | Unresolved: no series in the sources screened identifies multi-publisher curated transactions on a constant, representative US definition; the ANA and Jounce series are reported as descriptive findings [310, 101, 311] | [Chapter 2](#share) |
| H2a | A curated deal's all-in cost exceeds the open auction's for comparable opportunities | Unresolved: no matched comparison on the same inventory, period and buyer is public [115, 216, 290, 64] | [Chapter 6](#fees) |
| H2b | Every charge in a scored product's deal path is disclosed with its basis | Unresolved on its registered rule: no buyer-facing fee screen was observed; on the type-and-basis test six products document the charges they control and none the full deal path, and 13 state no charge type [212, 231, 281, 115, 24] [116, 12, 209, 128] | [Chapter 6](#fees) |
| H3 | Curated deals deliver the quality they are sold on, measured independently | Unresolved: every quantified claim found comes from a vendor, its agency or a partner; no independent measurement tests a named deal's promise [119, 120, 121, 125] [310, 101, 311] | [Chapter 7](#quality) |
| H4 | Access register; the essay's no-default-advantage claim | Coded per product: holds for Microsoft Curate and is unresolved for the other fourteen exchange products; Google Ad Manager and Index document a priority the curator can choose, not a default [111, 250, 24, 266] [14] | [Chapter 4](#access) |
| H5 | Fewer independently controlled curation data suppliers sell in the US than in 2023 | Unresolved: documented independent suppliers fell from four of eleven to two of ten, but control is undocumented for six of the eleven 2023 suppliers [49, 41, 38, 36, 44, 34, 50] | [Chapter 5](#market) |
| H6 | A buyer can establish a named deal's sellers, selection rule and data provenance | Unresolved: the products split; none documents all three completely and one documents none [110, 248, 25, 11, 258] [6] | [Chapter 3](#contents) |
| H7 | Sell-side selection by a software agent reaches documented third-party production use | Supported with qualification: one record, PubMatic's June 2026 release on Havas's Movistar campaign, names a customer, a period bounded by its date, an executed selection and a live context; it is the vendor's own, platform-level and not independently measured [313] [93, 120, 172, 168, 171] | [Chapter 10](#agents) |
| Q1 | Who decides | Curators own the selection rules and exchanges run them; only five exchange platforms document a fee recipient and four a party able to change a live deal [115, 24, 25, 109] [115, 111] | [Chapter 1](#defining) |
| Q2 | The competition record | Liability found (17 April 2025), remedies decided (2 September 2026), a judgment proposed (2 October 2026) and nothing in force by 29 October 2026; none of the three court documents read mentions curated deals [80] [79, 156, 78] | [Chapter 9](#competition) |
| Q3 | The fee stack | No public source screened gives all four charges, with payer, collector, base and period, for one named curated deal; Adloox's $0.15 CPM is the one disclosed data-fee amount [115, 216, 290, 64] [12] | [Chapter 6](#fees) |
| Q4 | Consolidation since 2021 | Acquisitions, a take-private, a pending sale, a retirement and rebrands, each dated from its settling source; every data supplier whose code changed moved into an operating group or out of the market [49, 41, 38, 36, 44, 34, 50] | [Chapter 5](#market) |
| Q5 | Post-purchase control | No change-notification, audit, export or remedy term is documented in public for the products screened; the rest code access unknown, and contract terms are unknown for all [115, 24, 25, 109] | [Chapter 12](#next) |

### The vendor universe

The 40 companies and products in the vendor sample, a reasoned sample rather than a census, with owner, group and status as their own pages read on 3 November 2026. Products in the matrix are tagged. The product register gives the inclusion reason for every registered product, including the listed comparators, the late candidates (Onetag's curation platform and Permutive Curation) and the historical rows.

**Vendor universe**

| Company / product | Legal entity / owner | HQ | Scope group | Formats | Buying routes | Main geographies | Status at cutoff | Src |
|---|---|---|---|---|---|---|---|---|
| ClearLine (ClearLine Activate is an activation variant) | Magnite, Inc. (copyright; Nasdaq: MGNI per register/stream D) | not stated on page | exchange | not stated in body (page tags only: CTV, Live Streaming, Video on-demand, Audio, mobile, OTT) | not named (DSPs unnamed) | author is 'group SVP, US revenue'; no availability statement | live (2026-09-09; GA of the curation update not stated) | [108] |
| Connect (Connect for Publishers; Connect for Buyers in navigation) | PubMatic, Inc. (copyright; Nasdaq: PUBM per register/stream D) | not stated (footer: registered data broker in Texas) | exchange | CTV, video, mobile app, display (buyers page: inventory 'curated through our SSP and enriched by Connect's data signals') | PubMatic Activate or 'your DSP of choice' | no availability statement | live (pages undated, copyright 2026) | [223] |
| Index Marketplaces | Index Exchange (not named on page; private per register) | not stated | exchange | streaming TV / CTV; online video; 'open internet' | 'For DSPs' category only; none named | not stated | live (page undated, copyright 2026; latest item 2026-10-23) | [258] |
| OpenXSelect (with Curated Packages; OpenXBuild is a sibling suite) | OpenX (copyright; OpenX Technologies, Inc. per register) | not stated | exchange | video; display; CTV | 'demand-side platforms (DSPs)' unnamed | US statements: 'In the U.S.'; identity graph 'available in the U.S. only' | live (page undated, copyright 2026) | [17] |
| Microsoft Curate (login at curate.xandr.com) | Microsoft (Microsoft Advertising; Microsoft Corporation per register) | not stated | exchange | not stated | 'Microsoft Invest and all major DSPs' | not stated | live (documentation dated 2025-10-21) | [56] |
| Maestro by Equativ (page labelled Curation Platform) | Equativ (not named on page; Equativ SAS, Bridgepoint majority per register/stream D) | not stated | exchange | display; video; native; CTV; apps | none named | 'global omnichannel marketplace'; no US statement | live (page undated; items to 2026-10-28) | [15] |
| Nexxen Curated Marketplace | Nexxen International Ltd. (copyright; Nasdaq: NEXN per register) | not stated | exchange | 'across device types'; none named | 'DSP & SSP activation' navigation only | not stated | live (page undated, copyright 2026) | [254] |
| TL Direct | TripleLift (Vista Equity Partners per register/stream D) | 53 W. 23rd St., 12th Floor, New York (footer address) | exchange | web; mobile; CTV | none named; UI, API and MCP access | New York dateline; 'global scale' | live, generally available (2026-06-10) | [130] |
| FreeWheel Curation Hub (FreeWheel Advertiser Suite) | FreeWheel (copyright; Comcast per register/stream D) | not stated | exchange | CTV; (video via Buyer Cloud) | 'any demand-side platform (DSP) including Buyer Cloud' | not stated | live (page undated, copyright 2026) | [16] |
| Google Ad Manager Curation (Beta) ('Configure curation settings (Beta)') | Google (copyright; Alphabet Inc. per register) | not stated | exchange | not stated (signals: URL, App ID) | DSPs unnamed; buyer's agency seat; Deal ID shared by the curation partner | not stated | beta (page undated, ©2026) | [59] |
| Commerce Grid | Criteo ('built by Criteo'; Criteo S.A., Nasdaq: CRTO per register) | not stated | exchange | display; in-app; web; mobile; native; video; CTV | '60+ third party DSPs'; 'any DSP'; none named | '17,000+ global brands and agencies'; no US statement | live (page undated, copyright 2026) | [266] |
| Buyer Hub (BuyerHub Assistant; Buyer Hub Agent) | InMobi Advertising (entity not stated; InMobi Group per register) | not stated | platform | CTV (sample query only) | none named | US only in a sample query | live (agency access 'August 31st, 2026') | [174] |
| Sovrn Curated Deals (offer heading: PMP deals, attention deals, curated packages, Commerce Audiences) | Sovrn, Inc. (footer; register says Sovrn Holdings, Inc.) | not stated | exchange | CTV / streaming; web display and outstream video; mobile app | buyer logos as image filenames only (Amazon Ads, BidSwitch, Basis, Google DV360, TTD, Simplifi, Beeswax, RTB House); not treated as named routes | 'US / EU / CA / APAC' | live (page undated) | [277] |
| YMax (YMax.ai; YMax+ managed service) | Yieldmo (copyright; Yieldmo, Inc. per register) | not stated | exchange | video; CTV | 'your existing DSP as standard deal IDs'; none named | not stated | live (page undated, copyright 2026) | [11] |
| Brand+ Marketplace (formerly Smaato) | Verve Group, Inc. (copyright; Verve Group SE per register) [interest mark] | not stated | exchange | CTV; video streams; mobile apps; open web | none named | '150+ countries'; 'across NAM, APAC and EU' | live (page undated, copyright 2026) | [21] |
| SmartPMP (siblings ContextualPMP, CognitivePMP) | Audigent (Experian plc per register/stream D; not on page) | 200 Varick St. Suite 514, New York (office labelled Audigent NYC) | data | not stated | 'premium SSP platforms'; 'basic DSP integrations'; none named | NYC office; no availability statement | live (page undated) | [193] |
| Experian Curated Deals (with Deal Engine and Deal Marketplaces) | Experian Information Solutions, Inc. (copyright; Experian plc per register) | not stated | data | 'no matter the channel'; none named | 'demand-side platforms (DSPs) and supply-side platforms (SSPs)'; none named | not stated | live (page undated, copyright 2026) | [255] |
| Comscore-Certified Deal IDs (with AU) | Comscore, Inc. (NASDAQ: SCOR) | Reston, VA (dateline) | verification | CTV (top CTV publishers); others not stated | PubMatic SSP; DSPs unnamed | Reston, VA; 'global' | live (expanded 2025-06-10) | [227] |
| Adelaide high-AU PMPs | Adelaide Metrics Inc. (copyright) | not stated | verification | digital display; OLV; CTV | SSPs: PubMatic, Microsoft Advertising (Xandr), TripleLift, Magnite, OpenX; DSPs: Google DV360, The Trade Desk, Yahoo DSP | 'more than 20 of the world's largest advertising markets'; US not named | live (page undated, © 2019–2026) | [60] |
| IAS Curation (with Google Ad Manager) | Integral Ad Science, Inc. (copyright; Novacap private per register/stream D) | New York (dateline) | verification | not stated | Google Ad Manager | 'global advertisers' | live (launched 2024-11-04; listed as a Google curation partner 2026-11-03) | [58] |
| Lotame Curated Marketplaces (Curation) | Lotame Solutions, Inc. ('Epsilon Lotame'; Publicis Groupe) [interest mark] | not stated | data | connected TV marketplaces; PMPs for web and mobile | 'DSPs' via PMP; 'leading SSPs'; none named | not stated | live (page undated, copyright 2026) | [13] |
| Multilocal curation (cureon engine) | Multilocal Media (page; 'Ltd' not on page) | not stated | data | not stated | '21+ SSP curation marketplaces' (names only as image filenames); 'DSP seat' | site versions UK and US; 'Always-On Global Coverage' | live (page undated) | [256] |
| Dstillery ID-free Deal IDs | Dstillery (copyright; Dstillery, Inc. per register) | not stated | data | display; video; native | Microsoft Curate (build); 'any leading DSP' | not stated | unclear (post dated 2022-07-25, edited later) | [61] |
| NeuroX Curation | Seedtag (entity not stated; Seedtag Advertising S.L. per register) | not stated | data | web; CTV; OLV; display | 'push to your DSP'; 'standard Deal IDs into existing DSPs'; none named | not stated | live (page undated) | [128] |
| Semasio HypertailPMP (service line on homepage) | Semasio ('now part of Samba') [interest mark] | not stated | data | not stated | PubMatic (testimonial only) | Denmark references; no market statement | unclear (homepage copyright 2025; service page not retried) | [270] |
| Samba Audiences (no PMP product on page) | Samba (copyright; Samba TV, Inc. per register) [interest mark] | not stated | data | smart TVs; web; 'every screen' | none named | not stated | unclear (page 2026-11-02; segments only) | [268] |
| Adloox Attention Segments in PubMatic (via Qwarry) | Adloox ('now part of Scope3'; Apostra per register/stream D) | not stated | verification | desktop and mobile web; video; 'avoid CTV' | PubMatic Media Console (Create Auction Package Deal); DSP selected by the buyer | countries selectable; no market statement | live (page 2026-07-17) | [12] |
| Stagwell Curate | Stagwell (NASDAQ:STGW) | New York City (dateline) | agency | CTV / streaming TV; online video; display; audio | partners FreeWheel and Magnite; The Trade Desk's Sincera for models | NYC; '45+ countries' of specialists | live (available immediately, 2026-07-16) | [5] |
| Premium Marketplace (GroupM at launch) | WPP Media Limited (footer); WPP plc per register | not stated | agency | connected TV; digital video; display | Magnite; PubMatic | 'currently available in EMEA and North America' (2022) | unclear (article 2022-02-24; no later page) | [7] |
| Dentsu Curate (Supply Path Optimisation & Curated Marketplace offering) | Dentsu Aegis Network (2020; Dentsu Group Inc. per register) | not stated | agency | not stated | Xandr's technology platform | Hong Kong, South East Asia, Singapore; 'other APAC markets'; 30 markets | unclear (release 2020-09-15; no later page) | [6] |
| Roku Curate | Roku, Inc. (copyright; Nasdaq: ROKU per register) | not stated | platform | CTV (streaming) | Roku activates directly or sends to the advertiser's preferred DSP (unnamed) | US (site banner: 125M+ daily active users in the US) | live (page 2026-04-27) | [8] |
| Activate | PubMatic, Inc. | not stated | exchange | CTV; video | PubMatic's own buying solution | not stated | live (page undated, copyright 2026) | [223] |
| OpenXBuild | OpenX | not stated | exchange | not stated | not stated | not stated | live (named on the OpenXSelect page, copyright 2026) | [17] |
| Microsoft Invest | Microsoft | not stated | platform | not stated | named as a buying route for curated deals | not stated | unclear (named in documentation dated 2025-10-21; announced end 2026-02-28 not reflected) | [56] |
| Proximic by Comscore | Comscore, Inc. (NASDAQ: SCOR) | Reston, VA (dateline) | verification | not stated | Google Ad Manager curation partner list ('Proximic by Comscore'); PubMatic via Comscore-Certified Deal IDs | not stated | live (2025-06-10 release; Google partner list 2026-11-03) | [227] |
| LiveRamp, Inc. (Google curation partner listing) | LiveRamp, Inc. (Publicis Groupe since 2026-11-02 per register/stream D) | not stated | data | not stated | Google Ad Manager curation partner list | not stated | unclear (listed as a curation partner 2026-11-03; no LiveRamp deal product page opened) | [59] |
| DoubleVerify Inc. (Google curation partner listing) | DoubleVerify Holdings, Inc. (per register) | not stated | verification | not stated | Google Ad Manager curation partner list | not stated | unclear (listed as a curation partner 2026-11-03; DV Marketplace Suite page not opened in U) | [59] |
| Onetag Curation Platform (login link; no product page opened) | Onetag (entity not stated) | offices Pisa, Milan, London | exchange | open-web formats | '35+ global DSP partners'; none named | Italy and UK offices; no US statement | live (homepage undated) | [259] |
| Permutive Curation | Permutive (entity not stated) | not stated | data | functionality table: Web, iOS, Android, CTV, API Direct | Monetize; Index Exchange Marketplaces; PubMatic; Google Curated Deals; OpenX; 'other SSP curation seats' | not stated | live (documentation undated) | [267] |
| PrimeAudience; SWYM.ai; System1 (Google curation partner listings) | not established | not stated | data | not stated | Google Ad Manager curation partner list | not stated | unclear (listed 2026-11-03; no vendor page opened) | [59] |

### Downloads

- `sell-side-curation-2026.pdf` — the print edition: A4, with page numbers and chapter bookmarks
- `sell-side-curation-2026.md` — the reading edition in Markdown, with numbered references
- `data/source-register.csv` — every source with publication, event and access dates, type, geography, funding, access limits and reliability notes
- `data/claim-ledger.csv` — every material claim with sources, passage locator, scope, evidence class, review outcome, confidence and final wording
- `data/capability-matrix-wide.csv` — vendors × dimensions, scores only (0–5, n/e, n/a)
- `data/capability-matrix-long.csv` — one row per cell with availability, evidence grade, date, rationale and sources
- `data/capability-dimensions.csv` — the dimension definitions and anchors
- `data/dq.csv` — Deal Quotient per vendor: group means, coverage, band, the reweightings and every dimension score
- `data/dq-method.csv` — DQ groups, weights, shrinkage, field mean, band thresholds, and the reweighting, shrinkage and missing-evidence tests
- `data/dq-missing-evidence.csv` — every scored cell turned into n/e one at a time, with the DQ before and after
- `data/swot.csv` — every SWOT line for the profiled vendors with its basis (record, vendor documentation, vendor claim, inference) and sources
- `data/deep-profiles.csv` — the deep-profile summaries: what is sold, buyers, earnings, dependencies, fit, limits
- `data/deep-profile-records.csv` — every profiled capability, performance claim, financial figure and timeline event with sources
- `data/study-evidence.csv` — studies with design, sample, sponsor and limits
- `data/corrections-ledger.csv` — every change made after verification and challenge passes, with implementation and verification
- `data/vendor-universe.csv` — the reasoned vendor sample with scope group, owner, formats, routes and status
- `data/agentic-map.csv` — AI and agentic capabilities with status labels and observability
- `data/rules.csv` — rules and governance instruments by jurisdiction
- `data/estimates.csv` — market-size estimates with inclusions, exclusions and combinability
- `data/decision-tool.csv` — the decision tool rows
- `data/product-register.csv` — the product register: every scored or listed product with its purchased service, operating entity, owner, cohort, inclusion reason, aliases and US availability
- `data/hypotheses.csv` — the hypotheses registered before extraction, with the rules for supported, contradicted and unresolved
- `data/figure-data.json` — the canonical dataset behind every data figure
- `data/derived-checks.json` — every derived number with its formula and inputs
- `data/stats.json` — every count the prose and figures quote about the research itself, computed from the ledgers
- `data/scoring-rules.json` — dimension definitions, anchors, cell rules and the DQ definition, fixed before scoring



## References

The 331 sources cited in this paper, numbered in order of first citation. The full register of 389 sources opened during the research, with access dates, access limits and reliability notes, is in data/source-register.csv.

[1] (Author's work) The Great Repackaging (essay). Evgeny Popov, No Fluff Advisory, 2026-07-16. https://nofluffadvisory.com/writing/the-great-repackaging/ Accessed 2026-11-03.

[2] (Author's work) Proof and Story: The State of Contextual Advertising in 2026 (research paper). Evgeny Popov, No Fluff Advisory, 2026-08-30. https://nofluffadvisory.com/research/contextual-2026/ Accessed 2026-11-03.

[3] Create deals (Equativ Help Center, Maestro curation). Equativ, 2026-10-15. https://help.equativ.com/manage-deals/create-deals Accessed 2026-11-03.

[4] PubMatic Activate (product page). PubMatic, Inc.. https://pubmatic.com/activate/ Accessed 2026-11-03. Access: partial extract.

[5] Stagwell Launches Stagwell Curate. Stagwell Inc. (press release via ACCESS Newswire), 2026-07-16. https://www.stagwellglobal.com/stagwell-launches-stagwell-curate/ Accessed 2026-11-03.

[6] Dentsu Aegis Network launches new programmatic supply solution for brands. dentsu (Dentsu Aegis Network), Singapore newsroom, 2020-09-15. https://www.dentsu.com/sg/en/our-news/dentsu-aegis-network-launches-new-programmatic-supply-solution-for-brands Accessed 2026-11-03.

[7] Introducing Premium Marketplace. WPP Media (GroupM at the time), 2022-02-24. https://www.wppmedia.com/news/introducing-premium-marketplace Accessed 2026-11-03.

[8] What is Roku Curate?. Roku, Inc. (Roku Advertising), 2026-04-27. https://advertising.roku.com/learn/resources/what-is-roku-curate Accessed 2026-11-03.

[9] Takeaways from the April 2025 Jounce Report: Curation Fees. Playwire Technical Team (reviewed by Jayson Dubin), 2025-04-10. https://www.playwire.com/blog/april-2025-jounce-report Accessed 2026-11-03.

[10] Curated Packages and Advertising Deals. OpenX, 2026. https://www.openx.com/curated-packages/ Accessed 2026-11-03.

[11] YMax Platform (product page). Yieldmo, Inc., undated (copyright 2026). https://www.yieldmo.com/ymax-yieldmo/ Accessed 2026-11-03.

[12] How to activate Adloox Attention Segments in PubMatic (learning center). Adloox SAS (part of Scope3), 2026-07-17. https://www.adloox.com/learning-center/pubmatic Accessed 2026-11-03.

[13] Curation (capability page): Lotame's Curated Marketplaces. Lotame Solutions, Inc. (Epsilon Lotame, Publicis Groupe), 2026. https://www.lotame.com/capabilities/curation Accessed 2026-11-03.

[14] Create a Marketplace Package. Index Exchange (Knowledge Base), 2026-10-26. https://kb.indexexchange.com/marketplace_partners/deals/create_marketplace_package.htm Accessed 2026-11-03.

[15] Orchestrate Your Campaigns to Deliver Your Best Performance With Maestro by Equativ (curation page). Equativ SAS, undated (navigation items dated to 28 October 2026). https://equativ.com/advertisers/curation Accessed 2026-11-03.

[16] Curation Hub for Media Buyers. FreeWheel (Comcast Advertising), 2026. https://www.freewheel.com/advertisersuite/curationhub Accessed 2026-11-03.

[17] OpenXSelect | Curated Inventory and Supply-Side Targeting. OpenX, 2026. https://www.openx.com/openxselect/ Accessed 2026-11-03.

[18] Premium Marketplace: transparency (Perspective). WPP Media, Andrew Meaden, Global Head of Investment (GroupM at the time), 2023-02-09. https://www.wppmedia.com/news/premium-marketplace-transparency Accessed 2026-11-03.

[19] Exchange Quality. Index Exchange, 2026. https://www.indexexchange.com/platform/exchange-quality/ Accessed 2026-11-03.

[20] Connect for Publishers (PubMatic product page). PubMatic. https://pubmatic.com/products/connect-for-publishers/ Accessed 2026-11-03. Access: partial extract.

[21] Brand+ Marketplace (formerly Smaato) - Omni-channel Growth | Verve. Verve Group, Inc., 2026. https://verve.com/brand-marketplace-for-dsps/ Accessed 2026-11-03.

[22] Curation, deals and supply packages FAQs (Commerce Grid knowledge base). Criteo S.A. (docs.commercegrid.criteo.com). https://docs.commercegrid.criteo.com/kb/guide/en/curation-deals-supply-packages-faqs-0jNJ8abdmw/Steps/4064725 Accessed 2026-11-03.

[23] Commerce Grid Glossary. Criteo S.A. / The MediaGrid (publisher.docs.themediagrid.com), 2025-09-08. https://publisher.docs.themediagrid.com/ref/glossary.html Accessed 2026-11-03.

[24] Curation for data and curation partners (Beta). Google (Authorized Buyers Help), 2026. https://support.google.com/authorizedbuyers/answer/15501813 Accessed 2026-11-03.

[25] About Marketplace packages (Beta). Google (Google Ad Manager Help), 2026. https://support.google.com/admanager/answer/13355702?hl=en Accessed 2026-11-03. Access: partial extract.

[26] ANA Programmatic Media Supply Chain Transparency Study: Complete Report. Association of National Advertisers (ANA), with TAG TrustNet, PwC and Kroll, 2023-12-05. https://www.fiducia.eco/_files/ugd/de006c_22584ed38e8e4be5b74f81169043a451.pdf Accessed 2026-11-03.

[27] TripleLift acquired by Vista Equity Partners (company announcement). TripleLift, Inc., 2021-03-29. https://triplelift.com/press/triplelift-acquired-by-vista-equity-partners/ Accessed 2026-11-03.

[28] Form 8-K exhibit 99.1, Magnite, Inc.: Magnite closes SpotX acquisition. Magnite, Inc. (SEC EDGAR), 2021-04-30. https://www.sec.gov/Archives/edgar/data/1595974/000138713121005124/ex99-1.htm Accessed 2026-11-03.

[29] Equativ announces Bridgepoint as new majority shareholder (press release). Equativ, 2023-02-27. https://www.equativ.com/press/equativ-announces-bridgepoint-as-new-majority-shareholder-to-accelerate-scale-and-growth-of-the-independent-adtech-company Accessed 2026-11-03.

[30] Form 20-F, Nexxen International Ltd., fiscal year ended 31 December 2025. Nexxen International Ltd. (SEC EDGAR), 2026. https://www.sec.gov/Archives/edgar/data/0001849396/000117891326000712/zk2634437.htm Accessed 2026-11-03. Access: partial extract.

[31] Microsoft Annual Report 2022 (fiscal year ended 30 June 2022): Xandr acquisition. Microsoft Corporation, 2022-07. https://www.microsoft.com/investor/reports/ar22/index.html Accessed 2026-11-03. Access: partial extract.

[32] Equativ and Sharethrough merge (press release). Equativ, 2024-06-12. https://www.equativ.com/press/equativ-and-sharethrough-merge Accessed 2026-11-03.

[33] Equativ and Sharethrough will now operate under Equativ brand (press release). Equativ, 2025-06-09. https://www.equativ.com/press/equativ-and-sharethrough-will-now-operate-under-equativ-brand-solidifying-global-position-as-leading-end-to-end-media-platform Accessed 2026-11-03.

[34] Oracle Advertising EOL Frequently Asked Questions, Version 1 (copy hosted by ppc.land; oracle.com original returned 403). Oracle Corporation, 2024-06-17. https://ppc.land/content/files/2024/06/oracle-advertising-end-of-life-faqs.pdf Accessed 2026-11-03.

[35] Form 8-K, Outbrain Inc.: closing of the acquisition of Teads from Altice Teads. Outbrain Inc. (now Teads Holding Co.) (SEC EDGAR), 2025-02-03. https://www.sec.gov/Archives/edgar/data/1454938/000114036125002802/ny20037354x5_8k.htm Accessed 2026-11-03.

[36] Samba TV announces acquisition of Semasio, strengthening its position in contextual targeting (press release). Samba TV, Inc., 2024-10-31. https://samba.com/newsroom/samba-tv-announces-acquisition-of-semasio-strengthening-its-position-in-contextual-targeting Accessed 2026-11-03.

[37] Adloox is now part of Scope3 (company post). Adloox, 2024-11-05. https://adloox.com/blog/adloox-is-now-part-of-scope3 Accessed 2026-11-03.

[38] Experian accelerates identity and activation capabilities with acquisition of Audigent (press release). Experian plc, 2024-12-04. https://www.experianplc.com/newsroom/press-releases/2024/experian-accelerates-identity-and-activation-capabilities-with-a Accessed 2026-11-03.

[39] Full-year results FY25 (regulatory results announcement). Experian plc, 2025-05-14. https://www.experianplc.com/newsroom/press-releases/2025/full-year-results-fy25 Accessed 2026-11-03.

[40] Publicis to acquire Lotame, the world's leading independent end-to-end data solution (press release). Publicis Groupe S.A., 2025-03-06. https://www.publicisgroupe.com/en/news/press-releases/publicis-to-acquire-lotame-the-world-s-leading-independent-end-to-end-data-solution Accessed 2026-11-03.

[41] Half-year financial report 2025, note 3.1 (acquisitions): Lotame Solutions, Inc.. Publicis Groupe S.A., 2025-07. https://publicis-groupe.publispeak.com/half-year-financial-report-2025/article/25/ Accessed 2026-11-03.

[42] Form 8-K, Outbrain Inc.: corporate name change to Teads Holding Co. (Item 5.03) and ticker change (Item 8.01). Teads Holding Co. (SEC EDGAR), 2025-06-09. https://www.sec.gov/Archives/edgar/data/1454938/000095015725000478/form8-k.htm Accessed 2026-11-03.

[43] Form 8-K, Integral Ad Science Holding Corp. (Item 8.01): agreement and plan of merger with Igloo Group Parent, Inc.. Integral Ad Science Holding Corp. (SEC EDGAR), 2025-09-24. https://www.sec.gov/Archives/edgar/data/1842718/000119312525214050/d938971d8k.htm Accessed 2026-11-03.

[44] Form 8-K exhibit 99.1, Integral Ad Science Holding Corp.: IAS announces completion of acquisition by Novacap. Integral Ad Science Holding Corp. (SEC EDGAR), 2025-12-23. https://www.sec.gov/Archives/edgar/data/1842718/000114036125046475/ef20061707_ex99-1.htm Accessed 2026-11-03.

[45] Form 10-K, Comscore, Inc., fiscal year ended 31 December 2025. Comscore, Inc. (SEC EDGAR), 2026-03. https://www.sec.gov/Archives/edgar/data/1158172/000115817226000009/scor-20251231.htm Accessed 2026-11-03. Access: partial extract.

[46] Form 8-K, Comscore, Inc.: completion of the Recapitalization Transaction with Charter Communications Holding Company, LLC, Liberty Broadband Corporation and Pine Investor, LLC. Comscore, Inc. (SEC EDGAR), 2025-12-29. https://www.sec.gov/Archives/edgar/data/1158172/000115817225000104/scor-20251229.htm Accessed 2026-11-03.

[47] Form 8-K exhibit 99.1, Omnicom Group Inc.: Omnicom completes acquisition of Interpublic. Omnicom Group Inc. (SEC EDGAR), 2025-11-26. https://www.sec.gov/Archives/edgar/data/29989/000119312525300788/d24786dex991.htm Accessed 2026-11-03.

[48] Publicis to acquire LiveRamp to accelerate data co-creation for smarter agents (press release). Publicis Groupe S.A., 2026-05-17. https://www.publicisgroupe.com/en/news/press-releases/publicis-to-acquire-liveramp-to-accelerate-data-co-creation-for-smarter-agents Accessed 2026-11-03.

[49] Form 8-K, LiveRamp Holdings, Inc. (Items 2.01, 3.01, 3.03, 5.01, 5.02, 5.03, 9.01): completion of acquisition by MMS USA Holdings, Inc.. LiveRamp Holdings, Inc. (SEC EDGAR), 2026-11-02. https://www.sec.gov/Archives/edgar/data/733269/000110465926121353/tm2628592d27_8k.htm Accessed 2026-11-03.

[50] Peer39 acquires Adloox, extending brand safety, contextual intelligence and verification across DV360, Meta and Amazon DSP (company post). Peer39, Inc., 2026-06-03. https://www.peer39.com/blog/peer39-acquires-adloox-extending-brand-safety-contextual-intelligence-and-verification-across-dv360-meta-and-amazon-dsp Accessed 2026-11-03.

[51] Form 8-K, DoubleVerify Holdings, Inc. (Item 8.01): press release announcing the merger agreement with Neptune BidCo US Inc.. DoubleVerify Holdings, Inc. (SEC EDGAR), 2026-08-06. https://www.sec.gov/Archives/edgar/data/0001819928/000110465926092121/tm2621904d1_8k.htm Accessed 2026-11-03.

[52] Form 8-K exhibit 99.1, DoubleVerify Holdings, Inc.: third-quarter 2026 results and stockholder approval of the merger. DoubleVerify Holdings, Inc. (SEC EDGAR), 2026-10-30. https://www.sec.gov/Archives/edgar/data/0001819928/000110465926120670/dv-ex99d1.htm Accessed 2026-11-03.

[53] Form 8-K exhibit 99.1, DoubleVerify Holdings, Inc.: press release announcing the agreement to be acquired by Nielsen (Neptune BidCo US Inc.). DoubleVerify Holdings, Inc. (SEC EDGAR), 2026-08-06. https://www.sec.gov/Archives/edgar/data/1819928/000110465926092121/tm2621904d1_ex99-1.htm Accessed 2026-11-03.

[54] Introducing Apostra (company blog post). Apostra, Inc. (formerly Scope3), 2026-09-23. https://apostra.com/blog/introducing-apostra/ Accessed 2026-11-03.

[55] Multilocal secures backing of OAREX to accelerate global growth with new capital (company article). Multilocal Media Ltd, 2026-10-01. https://www.multilocal.media/articles/multilocal-secures-backing-of-oarex-to-accelerate-global-growth-with-new-capital Accessed 2026-11-03.

[56] About Microsoft Curate (Microsoft Learn). Microsoft Advertising (Microsoft Learn, xandr-docs), 2025-10-21. https://learn.microsoft.com/en-us/xandr/curate/about-curate Accessed 2026-11-03.

[57] Samba homepage. Samba TV, Inc. (Samba), 2026-11-02. https://www.samba.com/ Accessed 2026-11-03.

[58] IAS Launches Curation for Google Ad Manager (news page: 'ias-curation-google-ad-manager'). Integral Ad Science, Inc. (press release), 2024-11-04. https://integralads.com/news/ias-curation-google-ad-manager/ Accessed 2026-11-03.

[59] Configure curation settings (Beta). Google (Google Ad Manager Help), 2026. https://support.google.com/admanager/answer/15308224?hl=en Accessed 2026-11-03.

[60] Attention Metrics Made Easy: Programmatic Activation (resource page). Adelaide Metrics Inc.. https://adelaidemetrics.com/attention-resources/attention-metrics-made-easy-programmatic-activation Accessed 2026-11-03.

[61] FAQ: Deal IDs. Dstillery, Inc., 2022-07-25. https://dstillery.com/blog/faq-deal-ids/ Accessed 2026-11-03.

[62] Magnite Launches ClearLine, Giving Ad Agencies a New, Efficient Route to Premium Video Inventory (press release). Magnite, Inc., 2023-04-17. https://www.magnite.com/press/magnite-launches-clearline-giving-ad-agencies-a-new-efficient-route-to-premium-video-inventory-3/ Accessed 2026-11-03.

[63] Programmatic Transparency Benchmark: Q1 2025 Benchmark Findings. ANA, TAG TrustNet and Fiducia, 2025-05-21. https://s3.amazonaws.com/media.mediapost.com/uploads/ANAQ12025Programmatic.pdf Accessed 2026-11-03.

[64] Publishers: Don't Fear Curation Fees, They're Just The Same Old Ad Tech Tax. Anthony Vargas, AdExchanger, 2025-05-01. https://www.adexchanger.com/programmatic/publishers-dont-fear-curation-fees-theyre-just-the-same-old-ad-tech-tax/ Accessed 2026-11-03.

[65] The ANA Releases Q2 2025 Programmatic Transparency Benchmark Report, Reveals $26.8B in Wasted Programmatic Spend. Fiducia DLT Limited (benchmark technology partner), 2025-08-11. https://www.fiducia.eco/post/the-ana-releases-q2-2025-programmatic-transparency-benchmark-report-reveals-26-8b-in-wasted-program Accessed 2026-11-03.

[66] The ANA Q2 2025 Programmatic Transparency Benchmark Finds $26.8B in Wasted Programmatic Spend (press release). ANA, 2025-08-14. https://www.ana.net/content/show/id/pr-2025-08-programmatictrans Accessed 2026-11-03.

[67] Does Forbes operate 'made for advertising' pages and transact mis-declared ad inventory?. Adalytics Research, 2024 (page undated; status updates 20 March to 2 April 2024). https://adalytics.io/blog/ads-observed-on-www3-forbes-subdomain Accessed 2026-11-03.

[68] Are Amazon and other ad tech vendors serving ANA members' ads on 'Made for Arbitrage' websites in 2024?. Adalytics Research, 2024 (page undated; data collected January 2024; references a 7 March 2024 ANA event). https://adalytics.io/blog/ads-observed-on-made-for-advertising-sites-in-january-2024 Accessed 2026-11-03.

[69] On pre-bid bot detection and filtration: Are ad tech vendors serving US Government and Fortune 500 brands' digital ads to bots?. Adalytics Research, undated page (latest observations November 2024; DSP analysis labelled October 2024). https://adalytics.io/blog/prebid-bot-filtration Accessed 2026-11-03.

[70] Blog tag: Deal IDs (two posts; 'Dstillery's privacy-safe audience targeting solution can now activate Deal IDs', 11 August 2022). Dstillery, Inc., 2022-08-11. https://dstillery.com/blog/tag/deal-ids/ Accessed 2026-11-03.

[71] Stagwell is building its own AI media curation marketplace. Digiday, Sam Bradley, 2026-07-16. https://digiday.com/media-buying/stagwell-is-building-its-own-ai-media-curation-marketplace/ Accessed 2026-11-03.

[72] Stagwell (STGW) Builds AI-Powered, Outcome-First TV Advertising Platform with FreeWheel. ExchangeWire PressBox (reproduced Stagwell/FreeWheel release), 2026-04-27. https://www.exchangewire.com/blog/2026/04/27/stagwell-stgw-builds-ai-powered-outcome-first-tv-advertising-platform-with-freewheel/ Accessed 2026-11-03.

[73] Stagwell Inc. (NASDAQ: STGW) Reports Results for the Three and Nine Months Ended September 30, 2026 (Form 8-K, Exhibit 99.1). Stagwell Inc. (SEC EDGAR), 2026-10-29. https://www.sec.gov/Archives/edgar/data/0000876883/000087688326000030/stgw2026930pr.htm Accessed 2026-11-03.

[74] Stagwell Inc. Form 10-Q for the quarter ended September 30, 2026. Stagwell Inc. (SEC EDGAR), 2026-10. https://www.sec.gov/Archives/edgar/data/0000876883/000087688326000033/stgw-20260930.htm Accessed 2026-11-03. Access: partial extract.

[75] Kroger launches private marketplace to connect brands with consumers. Supermarket News, Jennifer Strailey, 2021-10-20. https://www.supermarketnews.com/foodservice-retail/kroger-launches-private-marketplace-to-connect-brands-with-consumers Accessed 2026-11-03.

[76] We're here for off-platform (Instacart Ads page). Instacart (Maplebear Inc.). https://company.instacart.com/ads/were-here-for-off-platform Accessed 2026-11-03. Access: partial extract.

[77] PubMatic and Butler/Till Launch AdCP-Enabled Agentic AI Campaign Across CTV (syndicated copy of the 15 December 2025 release). PubMatic, Inc. and Butler/Till, via MarTech Series, 2025-12-16. https://martechseries.com/tv-advertising/pubmatic-and-butler-till-launch-adcp-enabled-agentic-ai-campaign-across-ctv/ Accessed 2026-11-03.

[78] United States v. Google LLC, No. 1:23-cv-00108 (E.D. Va.), Memorandum Opinion (liability), Dkt 1410. US District Court for the Eastern District of Virginia (Brinkema, J.), published by the US Department of Justice, 2025-04-17. https://www.justice.gov/atr/media/1412561/dl?inline Accessed 2026-11-03.

[79] [Proposed] Final Judgment, Dkt 1861-1 (attachment to the parties' Joint Notice of Proposed Final Judgment). United States, seventeen plaintiff states and Google LLC (joint filing, E.D. Va.), 2026-10-02. https://storage.courtlistener.com/recap/gov.uscourts.vaed.533508/gov.uscourts.vaed.533508.1861.1.pdf Accessed 2026-11-03.

[80] CourtListener docket, United States v. Google LLC, 1:23-cv-00108 (E.D. Va.), entries 1855-1861. Free Law Project (CourtListener RECAP archive), 2026-10-29. https://www.courtlistener.com/docket/66753787/united-states-v-google-llc/?filed_after=2026-08-15&order_by=desc Accessed 2026-11-03.

[81] RECAP search results, S.D.N.Y., Kargo / Sovrn / TRUSTX / Teads / Index Exchange (docket cards, November 2025 to October 2026). Free Law Project (CourtListener RECAP archive), 2026-11-03. https://www.courtlistener.com/?type=r&q=Kargo%20OR%20Sovrn%20OR%20TRUSTX%20OR%20Teads%20OR%20%22Index%20Exchange%22&court=nysd&filed_after=11%2F01%2F2025&order_by=dateFiled%20desc Accessed 2026-11-03. Access: partial extract.

[82] RECAP search results, E.D. Va., OpenX / PubMatic / Magnite / Index Exchange v. Google LLC (five docket cards). Free Law Project (CourtListener RECAP archive), 2026-11-03. https://www.courtlistener.com/?type=r&q=OpenX%20OR%20PubMatic%20OR%20Magnite%20OR%20%22Index%20Exchange%22&court=vaed&filed_after=08%2F01%2F2025&order_by=dateFiled%20desc Accessed 2026-11-03. Access: partial extract.

[83] Case T-794/25: Action brought on 20 November 2025 – Google and Alphabet v Commission (OJ C/2026/185). Court of Justice of the European Union, via EUR-Lex (Publications Office), 2026-01-12. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:62025TN0794 Accessed 2026-11-03.

[84] Vermont becomes 23rd state to enact consumer privacy law. IAPP (International Association of Privacy Professionals), 2026-06-17. https://iapp.org/news/a/vermont-becomes-23rd-state-to-enact-consumer-privacy-law Accessed 2026-11-03.

[85] US State Privacy Legislation Tracker 2025: Comprehensive Consumer Privacy Bills (chart). IAPP (International Association of Privacy Professionals), 2025-11-24. https://prod.iapp.org/media/pdf/resource_center/State_Comp_Privacy_Law_Chart.pdf Accessed 2026-11-03.

[86] IAB TCF: Belgian Market Court upholds €250,000 fine against IAB for GDPR violation. Lewis Silkin LLP, 2025-05-27. https://www.lewissilkin.com/en/insights/2025/05/27/iab-tcf-belgian-market-court-upholds-250-000-fine-against-iab-for-gdpr-violatio-102kyon Accessed 2026-11-03.

[87] LG and Magnite launch agentic home screen ads with a winning NFL quarterback (Alyssa Boyle). AdExchanger, 2026-10-14. https://adexchanger.com/tv/lg-and-magnite-launch-agentic-home-screen-ads-with-a-winning-nfl-quarterback Accessed 2026-11-03.

[88] AdCP documentation: Signals protocol walkthrough (3.2.8 path). AgenticAdvertising.org, 2026. https://docs.adcontextprotocol.org/dist/docs/3.2.8/signals/overview Accessed 2026-11-03.

[89] The Rebuild: How Agentic Advertising Will Redefine the Future of Media (Scope3 news post, now an Apostra placeholder). Scope3 / Apostra. https://apostra.com/news/the-rebuild-how-agentic-advertising-will-redefine-the-future-of-media Accessed 2026-11-03. Access: access error.

[90] Why Magnite Built a Seller Agent — and What It Signals for AdCP (company blog). Magnite, Inc. (Magnite Team), 2026-01-06. https://www.magnite.com/blog/why-magnite-built-a-seller-agent-and-what-it-signals-for-adcp/ Accessed 2026-11-03.

[91] IABTechLab/buyer-agent and IABTechLab/seller-agent release and tag records (GitHub API). IAB Tech Lab, 2026-08-18. https://api.github.com/repos/IABTechLab/seller-agent/releases?per_page=30 Accessed 2026-11-03.

[92] AAMP (Agentic Advertising Management Protocols) standard page. IAB Tech Lab, 2026-10-27. https://iabtechlab.com/standards/aamp-agentic-advertising-management-protocols/ Accessed 2026-11-03.

[93] Beverage marketer sees cost savings with agentic media buying test (Sara Karlovitch). Marketing Dive (Industry Dive), 2026-03-17. https://www.marketingdive.com/news/beverage-marketer-sees-cost-savings-with-agentic-media-buying-test/814905/ Accessed 2026-11-03.

[94] List of Data Vendors. Index Exchange (Knowledge Base), 2026-10-21. https://kb.indexexchange.com/integrations/list_of_data_vendors.htm Accessed 2026-11-03.

[95] OpenRTB 2.6 Implementation Notes (implementation.md), section 7.3 PMP & Direct Deals. IAB Tech Lab (OpenRTB Working Group), 2026. https://github.com/InteractiveAdvertisingBureau/openrtb2.x/blob/main/implementation.md Accessed 2026-11-03.

[96] Deal Sync API Specification (deal1.0.md; the Tech Lab page calls it the Deals API, version 1.0). IAB Tech Lab (repository IABTechLab/deal-api), 2026-02-06. https://github.com/IABTechLab/deal-api/blob/main/deal1.0.md Accessed 2026-11-03.

[97] AdCP documentation: introduction (agent roles, GA statement, Prebid Sales Agent status). AgenticAdvertising.org, 2026. https://docs.adcontextprotocol.org/docs/intro Accessed 2026-11-03.

[98] Programmatic Transparency Benchmark: Q3 2025 Benchmark Findings. ANA, TAG TrustNet and Fiducia, 2025-11-05. https://s3.amazonaws.com/media.mediapost.com/uploads/Q32025ProgrammaticTransparencyBenchmark.pdf Accessed 2026-11-03.

[99] Programmatic Transparency Benchmark: Q1 2026 Benchmark Findings. ANA, TAG TrustNet and Fiducia, 2026-06-01. https://www.fiducia.eco/_files/ugd/de006c_21911c212fab454bbd16b20517f333ae.pdf Accessed 2026-11-03.

[100] Programmatic Transparency Benchmark: Q4 2025 Benchmark Findings. ANA, TAG TrustNet and Fiducia, 2026-02-23. https://www.fiducia.eco/_files/ugd/de006c_730e6f3f60554d2983e0b75321eac55d.pdf Accessed 2026-11-03.

[101] The State Of The Open Internet 2025. Jounce Media, 2025 (undated PDF; charts carry March 2025 data). https://www.jouncemedia.com/build/resources/2025_Jounce_Media_State_Of_The_Open_Internet_2afe364829.pdf Accessed 2026-11-03.

[102] Programmatic Transparency Benchmark: 2024 Benchmark Findings Update (December 2024). ANA and TAG TrustNet, with Fiducia, 2024-12. https://fiducia.eco/_files/ugd/de006c_0dc02b31ad01468baa172ff0b0e82197.pdf Accessed 2026-11-03.

[103] ANA Programmatic Transparency Benchmark Reveals Widening Gap Between Top and Bottom Advertisers (press release). ANA, 2026-05-27. https://www.ana.net/content/show/id/pr-2026-05-programmatic Accessed 2026-11-03.

[104] The State Of The Open Internet (2023 edition). Jounce Media, 2023. https://www.jouncemedia.com/resources/Jounce+Media+2023+State+Of+The+Open+Internet.pdf Accessed 2026-11-03.

[105] Time for change and transparency in programmatic advertising (ISBA Programmatic Supply Chain Transparency Study, 2020). ISBA, in association with the AOP; work carried out by PwC, 2020-05-06. https://isba.org.uk/article/time-change-and-transparency-programmatic-advertising Accessed 2026-11-03. Access: partial extract.

[106] ISBA Programmatic supply chain transparency study II summary: Test of the Taskforce Financial Audit Toolkit. PwC for ISBA, 2023-01. https://www.isba.org.uk/system/files/media/documents/2023-01/ISBA%20%20PwC%20programmatic%20supply%20chain%20study%20II%20%28summary%29-%2018%20January%202023.pdf Accessed 2026-11-03. Access: partial extract.

[107] Magnite To Unify Curation and Activation Within ClearLine. Magnite, Inc. (press release), 2025-10-01. https://www.magnite.com/press/clearline-curation/ Accessed 2026-11-03.

[108] Bridging the Gap Between Data and Supply: Inside ClearLine's Self-Serve Audience Capabilities. Magnite (blog), 2026-09-09. https://www.magnite.com/blog/bridging-the-gap-between-data-and-supply-inside-clearlines-self-serve-audience-capabilities/ Accessed 2026-11-03.

[109] Understanding Index Marketplaces transactions for Marketplace Partners. Index Exchange (Knowledge Base), 2026-09-09. https://kb.indexexchange.com/marketplace_partners/marketplace_fees_partners.htm Accessed 2026-11-03.

[110] Microsoft Curate - Create a Curated Deal Line Item. Microsoft Advertising (Microsoft Learn), 2026-02-18. https://learn.microsoft.com/en-us/xandr/curate/create-a-curated-deal-line-item Accessed 2026-11-03.

[111] Microsoft Curate - Curated Deals (client overview and common questions). Microsoft Advertising (Microsoft Learn), 2025-10-21. https://learn.microsoft.com/en-us/xandr/curate/curated-deals Accessed 2026-11-03.

[112] Curated Deal Floors. Microsoft Advertising (Microsoft Learn), 2025-10-21. https://learn.microsoft.com/en-us/xandr/curate/curated-deal-floors Accessed 2026-11-03.

[113] Curation Optimization. Microsoft Advertising (Microsoft Learn, Microsoft Curate documentation), 2025-10-21. https://learn.microsoft.com/en-us/xandr/curate/curation-optimization Accessed 2026-11-03.

[114] Get started with deal management (Maestro help center). Equativ (help.equativ.com), 2026-04-23. https://help.equativ.com/manage-deals/get-started-with-deal-management Accessed 2026-11-03.

[115] Curator Margins. Microsoft Advertising (Microsoft Learn), 2025-10-21. https://learn.microsoft.com/en-us/xandr/curate/curator-margins Accessed 2026-11-03.

[116] Audigent | The Leading Data Activation, Curation and Identity Platform (homepage). Audigent (a part of Experian), undated. https://audigent.com/ Accessed 2026-11-03.

[117] TL Direct | Curation on Your Terms. TripleLift, 2026. https://triplelift.com/tl-direct/ Accessed 2026-11-03.

[118] Asymmetric information in programmatic advertising: Three studies on adverse selection, mechanism choices, and fee structures (doctoral thesis record). Francesco Balocco, Erasmus University Rotterdam (supervisors Ting Li, Eric van Heck), 2023-09-08. https://pure.eur.nl/en/publications/asymmetric-information-in-programmatic-advertising-three-studies-/ Accessed 2026-11-03. Access: abstract only.

[119] PubMatic and Agence 79 Achieve 5x Higher CTR vs Open Market with Programmatic Video Curation in France (press release, Paris). PubMatic, Inc. (press release), 2026-06-11. https://pubmatic.com/news/pubmatic-and-agence-79-achieve-higher-ctr/ Accessed 2026-11-03.

[120] Chalice AI and OpenX Bring AI-Powered Curation to High-Quality Video. OpenX Technologies, Inc. and Chalice AI (press release), 2026-06-16. https://www.openx.com/press-releases/chalice-ai-and-openx-bring-ai-powered-curation-to-high-quality-video/ Accessed 2026-11-03.

[121] Microsoft Curate | Microsoft Advertising (product page: 'Curate by the numbers'). Microsoft Advertising, undated. https://about.ads.microsoft.com/en/solutions/technology/microsoft-curate Accessed 2026-11-03.

[122] SmartPMPs Outperform DMP Data & Open Exchange Inventory (case study page). Audigent (a part of Experian), undated. https://audigent.com/case-studies/smartpmp-outperforms-open-exchange Accessed 2026-11-03.

[123] Audigent Improves Alcon Sign-Ups Using Curated PMPs (case study page). Audigent (a part of Experian), undated. https://audigent.com/case-studies/audigent-improves-alcon-sign-ups-using-curated-pmps Accessed 2026-11-03.

[124] Lotame Launches Curation to Help Brands Find Their Audience (press release, GlobeNewswire, syndicated copy on AAP). Lotame Solutions, Inc. (press release), 2024-08-14. https://aapnews.aap.com.au/news/globenewswire9204553 Accessed 2026-11-03.

[125] Think Global, Act Local: How Lotame and PHD EMEA Partnered to Execute a Curated Marketplace for Global CPG Drinks Brand (case study PDF). Lotame Solutions, Inc., undated (hosted under '2025_Case_Studies'). https://resources.lotame.com/hubfs/2025_Case_Studies/CPG-Alchohol_PHD_Curation_Lotame-Case-Study.pdf Accessed 2026-11-03.

[126] Adelaide Partners with Magnite, PubMatic, TRUSTX, Sovrn, Xandr, Google Display & Video 360, and Other Supply-Side Partners to Launch the Industry's First High-Attention Marketplaces Worldwide. Adelaide Metrics Inc. (press release), 2023-03-07. https://www.adelaidemetrics.com/news/adelaide-partners-with-magnite-pubmatic-trustx-sovrn-xandr-google-display-amp-video-360-and-other-supply-side-partners-to-launch-the-industrys-first-high-attention-marketplaces-worldwide Accessed 2026-11-03.

[127] Teads and Adelaide Partner To Launch High-Attention Omnichannel Private Marketplaces. Adelaide Metrics Inc. (release hosted on Adelaide's blog; Teads announced), 2023-08-17. https://www.adelaidemetrics.com/blog/teads-and-adelaide-partner-to-launch-high-attention-omnichannel-private-marketplaces Accessed 2026-11-03.

[128] Seedtag NeuroX Curation: The Self-Serve Interface to the Neuro-Contextual Exchange (product page). Seedtag Advertising S.L., undated. https://www.seedtag.com/curation Accessed 2026-11-03.

[129] Multilocal (homepage). Multilocal Media Ltd, undated. https://www.multilocal.media/ Accessed 2026-11-03.

[130] TripleLift Launches TL Direct, Its Self-Serve Platform (general availability). TripleLift, Inc. (press release), 2026-06-10. https://triplelift.com/press/triplelift-launches-tl-direct-its-self-serve-platform/ Accessed 2026-11-03.

[131] Opinion and Order on issue preclusion, In re Google Digital Advertising Antitrust Litigation, 21-md-3010 (PKC), Dkt 146 in 1:23-cv-05177. United States District Court for the Southern District of New York (Senior Judge P. Kevin Castel), 2025-10-27. https://www.sdnyblog.com/files/2025/10/23-cv-05177-In-Re-Google-Digital-Advertising-Antitrust-Litigation.pdf Accessed 2026-11-03.

[132] CourtListener docket, Equativ SAS v. Google LLC, 1:26-cv-00140 (S.D.N.Y.), pages 1 and 2. Free Law Project (CourtListener RECAP archive), 2026-10-24. https://www.courtlistener.com/docket/72117542/equativ-sas-v-google-llc/ Accessed 2026-11-03.

[133] AT.40670 - Google - Adtech and Data-related practices (competition case register page). European Commission (competition case search), 2026-01-15. https://competition-cases.ec.europa.eu/cases/AT.40670 Accessed 2026-11-03.

[134] Commission fines Google EUR 2.95 billion over abusive practices in online advertising technology (press release IP/25/1992). European Commission (Directorate-General for Competition), 2025-09-05. https://ec.europa.eu/commission/presscorner/api/files/document/print/en/ip_25_1992/IP_25_1992_EN.pdf Accessed 2026-11-03.

[135] Our proposal in response to the European Commission's decision about our ad tech business fully addresses concerns and minimizes disruption for businesses. Google (The Keyword blog), 2025-11-14. https://blog.google/company-news/inside-google/around-the-globe/google-europe/european-commission-ad-tech-response/ Accessed 2026-11-03.

[136] InfoCuria case information: T-794/25 Google and Alphabet v Commission. Court of Justice of the European Union (InfoCuria), 2026-01-12. https://infocuria.curia.europa.eu/tabs/affair?lang=en&sort=AFF_NUM-DESC&searchTerm=%2522T%252D794%252F25%2522&publishedId=T-794%2F25 Accessed 2026-11-03.

[137] Answer given by Executive Vice-President Ribera on behalf of the European Commission to written question E-000132/2026. European Commission, via the European Parliament's parliamentary questions register, 2026-03-12. https://www.europarl.europa.eu/doceo/document/E-10-2026-000132-ASW_EN.html Accessed 2026-11-03.

[138] Investigation into suspected anti-competitive conduct by Google in ad tech (CMA case page). Competition and Markets Authority (UK), 2026-05-19. https://www.gov.uk/cma-cases/investigation-into-suspected-anti-competitive-conduct-by-google-in-ad-tech Accessed 2026-11-03.

[139] OpenRTB 2.6 specification (2.6.md on main; release 2.6-202606). IAB Tech Lab (OpenRTB Working Group), 2026-06-11. https://github.com/InteractiveAdvertisingBureau/openrtb2.x/blob/main/2.6.md Accessed 2026-11-03.

[140] OpenRTB Specification v3.0 (March 2022), Object: Deal and Object: Item. IAB Tech Lab, 2022-03. https://github.com/InteractiveAdvertisingBureau/openrtb/blob/master/OpenRTB%20v3.0%20FINAL.md Accessed 2026-11-03.

[141] Ads.txt Version 1.1 specification. IAB Tech Lab (Programmatic Supply Chain working group), 2022-07. https://github.com/InteractiveAdvertisingBureau/Supply-Chain-Validation/blob/main/ads.txt%20v1.1.md Accessed 2026-11-03.

[142] Sellers.json specification 1.0. IAB Tech Lab (OpenRTB Working Group), 2019-07. https://github.com/InteractiveAdvertisingBureau/Supply-Chain-Validation/blob/main/sellers-json.md Accessed 2026-11-03.

[143] OpenRTB SupplyChain object (supplychainobject.md). IAB Tech Lab (OpenRTB Working Group), 2019-07-31. https://github.com/InteractiveAdvertisingBureau/openrtb/blob/main/supplychainobject.md Accessed 2026-11-03.

[144] Curated Audiences and launching the Curation Framework Working Group (blog post). Hillary Slattery, IAB Tech Lab, 2024-12-16. https://iabtechlab.com/curated-audiences-and-launching-the-curation-framework-working-group/ Accessed 2026-11-03.

[145] Deals in Prebid (docs.prebid.org/adops/deals.html). Prebid.org. https://docs.prebid.org/adops/deals.html Accessed 2026-11-03.

[146] MRC Viewable Ad Impression Measurement Guidelines (Desktop), Version 2.0 (Final with 2015 additions). Media Rating Council (with the IAB Emerging Innovations Task Force), 2015-08-18. https://mediaratingcouncil.org/sites/default/files/Standards/081815%20Viewable%20Ad%20Impression%20Guideline_v2.0_Final.pdf Accessed 2026-11-03.

[147] MRC Mobile Viewable Ad Impression Measurement Guidelines, Final Version. Media Rating Council, 2016-06-28. https://mediaratingcouncil.org/sites/default/files/Standards/062816%20Mobile%20Viewable%20Guidelines%20Final.pdf Accessed 2026-11-03.

[148] Invalid Traffic Detection and Filtration Standards Addendum, June 2020 Update (Final) (IVT 2.0). Media Rating Council, 2020-06. https://mediaratingcouncil.org/sites/default/files/Standards/IVT%20Addendum%20Update%20062520.pdf Accessed 2026-11-03.

[149] 2024 IVT Interim Updates (memo to digital measurement vendors subject to MRC audit). George Ivie, Laris Oliveri and Ron Pinelli, Media Rating Council, 2024-04-24. https://mediaratingcouncil.org/sites/default/files/Standards/2024_IVT_Interim_Updates_FINAL.pdf Accessed 2026-11-03.

[150] MRC Cross-Media Audience Measurement Standards (Phase I Video), Final. Media Rating Council (sponsoring associations: 4A's, ANA, IAB, VAB), 2019-09. https://www.mediaratingcouncil.org/sites/default/files/Standards/MRC%20Cross-Media%20Audience%20Measurement%20Standards%20%28Phase%20I%20Video%29%20Final.pdf Accessed 2026-11-03.

[151] MRC Digital Advertising Auction Transparency Standards. Media Rating Council (steer team: MRC, Omnicom as lead, ANA, WFA, 4As, IAB Tech Lab), 2026-01. https://mediaratingcouncil.org/sites/default/files/Standards/MRC_Digital_Advertising_Auction_Transparency_Standards_FINAL_0.pdf Accessed 2026-11-03.

[152] Certified Against Fraud (CAF) Program page. Trustworthy Accountability Group (TAG). https://www.tagtoday.net/fraud Accessed 2026-11-03.

[153] Certified Against Malvertising (CAM) Program page (threat-sharing). Trustworthy Accountability Group (TAG). https://www.tagtoday.net/threat-sharing Accessed 2026-11-03.

[154] Brand Safety Certified Program page. Trustworthy Accountability Group (TAG). https://www.tagtoday.net/brand-safety Accessed 2026-11-03.

[155] Certified for Transparency program page. Trustworthy Accountability Group (TAG). https://www.tagtoday.net/transparency Accessed 2026-11-03.

[156] Memorandum Opinion, United States et al. v. Google LLC, No. 1:23-cv-108 (LMB/JFA), E.D. Va., Dkt 1858 (remedies). United States District Court for the Eastern District of Virginia (Judge Leonie M. Brinkema), 2026-09-02. https://storage.courtlistener.com/recap/gov.uscourts.vaed.533508/gov.uscourts.vaed.533508.1858.0_1.pdf Accessed 2026-11-03.

[157] Google's general search and search advertising services (CMA strategic market status case page). Competition and Markets Authority (UK), 2026-09-23. https://www.gov.uk/cma-cases/googles-general-search-and-search-advertising-services Accessed 2026-11-03.

[158] Google's mobile platform (CMA strategic market status case page). Competition and Markets Authority (UK), 2026-08-14. https://www.gov.uk/cma-cases/googles-mobile-platform Accessed 2026-11-03.

[159] Regulations (California Privacy Protection Agency rulemaking page). California Privacy Protection Agency, 2026. https://cppa.ca.gov/regulations/ Accessed 2026-11-03.

[160] FTC Finalizes Order Banning Mobilewalla from Selling Sensitive Location Data (press release). Federal Trade Commission, 2025-01-14. https://www.ftc.gov/news-events/news/press-releases/2025/01/ftc-finalizes-order-banning-mobilewalla-selling-sensitive-location-data Accessed 2026-11-03.

[161] Judgment of the Court (Fourth Chamber) of 7 March 2024, IAB Europe v Gegevensbeschermingsautoriteit, Case C-604/22, ECLI:EU:C:2024:214. Court of Justice of the European Union (via EUR-Lex), 2024-03-07. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:62022CJ0604 Accessed 2026-11-03.

[162] Belgian Market Court Confirms Limited Role of IAB Europe In The TCF (statement). IAB Europe, 2025-05-14. https://iabeurope.eu/belgian-market-court-confirms-limited-role-of-iab-europe-in-tcf/ Accessed 2026-11-03.

[163] The Transparency & Consent Framework (TCF) v2.3 (framework page). IAB Europe, 2025-04. https://iabeurope.eu/transparency-consent-framework/ Accessed 2026-11-03.

[164] AdCP README on main (protocol surfaces table). AgenticAdvertising.org / adcontextprotocol GitHub organisation, 2026-10-28. https://github.com/adcontextprotocol/adcp/blob/main/README.md Accessed 2026-11-03.

[165] AdCP agent registry API (agents list). AgenticAdvertising.org, 2026-11-03. https://agenticadvertising.org/api/registry/agents?limit=200 Accessed 2026-11-03.

[166] The Ad Context Protocol aims to make sense of agentic ad demand (James Hercher). AdExchanger, 2025-10-15. https://www.adexchanger.com/publishers/the-ad-context-protocol-aims-to-make-sense-of-agentic-ad-demand/ Accessed 2026-11-03.

[167] adcontextprotocol.org homepage ('AdCP 3.2 is now generally available'; 'It's already happening'). AgenticAdvertising.org, 2026. https://adcontextprotocol.org/ Accessed 2026-11-03.

[168] IABTechLab/OpenProposal repository (README, release tag v3.0-draft-1). IAB Tech Lab, 2026-10-09. https://github.com/IABTechLab/OpenProposal Accessed 2026-11-03.

[169] IABTechLab/seller-agent repository (Seller Agent SDK reference implementation; README, releases, commits). IAB Tech Lab, 2026-10-28. https://github.com/IABTechLab/seller-agent Accessed 2026-11-03.

[170] IAB Tech Lab Introduces AAMP 3.0 with OpenProposal (press release). IAB Tech Lab, 2026-09-22. https://iabtechlab.com/press-releases/iab-tech-lab-introduces-aamp-3-0-with-openproposal/ Accessed 2026-11-03.

[171] prebid/salesagent repository (README, pyproject.toml, GitHub API repo and commits). Prebid.org, 2026-10-28. https://github.com/prebid/salesagent Accessed 2026-11-03.

[172] LG Ad Solutions and Magnite Expand Partnership to Advance Agentic Buying on the Home Screen (GlobeNewswire release, Finviz copy). Magnite, Inc. and LG Ad Solutions, 2026-10-14. https://finviz.com/news/402336/lg-ad-solutions-and-magnite-expand-partnership-to-advance-agentic-buying-on-the-home-screen Accessed 2026-11-03.

[173] Kargo announces closed beta of Project KERA (ExchangeWire PressBox, reproduced release). Kargo Global, Inc. via ExchangeWire PressBox, 2026-04-07. https://www.exchangewire.com/blog/2026/04/07/kargo-announces-closed-beta-of-project-kera/ Accessed 2026-11-03.

[174] Buyer Hub - An AI Agent for Media Buyers. InMobi Advertising, 2026. https://go.inmobi.com/buyer-hub-2026/ Accessed 2026-11-03. Access: partial extract.

[175] AdCP repository releases (GitHub releases API, 15 most recent entries, and releases/latest). AgenticAdvertising.org / adcontextprotocol GitHub organisation, 2026-10-31. https://api.github.com/repos/adcontextprotocol/adcp/releases?per_page=15 Accessed 2026-11-03.

[176] AdCP documentation: Media Buy walkthrough (seller-planned campaign from brief to proposal acceptance). AgenticAdvertising.org, 2026. https://docs.adcontextprotocol.org/docs/media-buy/ Accessed 2026-11-03.

[177] AgenticAdvertising.org homepage (members, directory, production claim). AgenticAdvertising.org, 2026. https://agenticadvertising.org/ Accessed 2026-11-03.

[178] Prebid Sales Agent documentation page. Prebid.org, 2026. https://docs.prebid.org/agents/salesagent.html Accessed 2026-11-03.

[179] IAB Tech Lab press-release index (page 1). IAB Tech Lab, 2026-11-03. https://iabtechlab.com/press-releases/ Accessed 2026-11-03.

[180] IABTechLab/buyer-agent repository (Buyer Agent SDK reference implementation; GitHub API). IAB Tech Lab, 2026-10-28. https://github.com/IABTechLab/buyer-agent Accessed 2026-11-03.

[181] IABTechLab/agentic-direct repository (Core Orchestration; README and GitHub API). IAB Tech Lab, 2026-01-28. https://github.com/IABTechLab/agentic-direct Accessed 2026-11-03.

[182] IAB Tech Lab Tools Portal: Agent Registry. IAB Tech Lab, 2026. https://tools.iabtechlab.com/agent-registry Accessed 2026-11-03. Access: access error.

[183] Stagwell (STGW) Launches The Media Machine, Full Lifecycle Agentic Media Operating System (press release, ACCESS Newswire). Stagwell Inc., 2026-06-19. https://www.stagwellglobal.com/stagwell-stgw-launches-the-media-machine-full-lifecycle-agentic-media-operating-system/ Accessed 2026-11-03.

[184] Magnite Advances Agentic Vision as Major Media Owners, Buyers and Technology Players Leverage Orchestration Layer (GlobeNewswire release, Finviz copy). Magnite, Inc., 2026-09-16. https://finviz.com/news/392579/magnite-advances-agentic-vision-as-major-media-owners-buyers-and-technology-players-leverage-orchestration-layer Accessed 2026-11-03.

[185] Empowering businesses for a future that is conversational, personal, and agentic (Microsoft Advertising blog, Kya Sainsbury-Carter). Microsoft Advertising, 2025-05-14. https://about.ads.microsoft.com/en/blog/post/may-2025/empowering-businesses-for-a-future-that-is-conversational-personal-and-agentic Accessed 2026-11-03.

[186] Microsoft Advertising blog author page, Nick Bolt (listing of 'From Proprietary Assets to Profitable Media Business: The Power of Curation', 10 September 2026). Microsoft Advertising, 2026-10-21. https://about.ads.microsoft.com/en/blog/author/nick-bolt Accessed 2026-11-03. Access: partial extract.

[187] Index Exchange Announces Marketplaces (launch release). Index Exchange Inc., 2024-01-23. https://www.indexexchange.com/press/press-releases/announces-marketplaces Accessed 2026-11-03.

[188] Index Marketplaces Accelerates; Index Exchange Adds Data Vendor Ecosystem to Power Sell-Side Audience Activation. Index Exchange (press release), 2025-10-07. https://www.indexexchange.com/press/press-releases/index-exchange-data-vendor-ecosystem Accessed 2026-11-03.

[189] Nexxen's Curated Marketplace Enables Partners to Unlock New Value from Data for Smarter Curation (general availability). Nexxen International Ltd. (press release), 2025-10-01. https://nexxen.com/nexxens-curated-marketplace-enables-partners-to-unlock-new-value-from-data-for-smarter-curation/ Accessed 2026-11-03.

[190] Audigent curated inventory packages now available in DV360, delivered through Index Marketplaces (blog post). Experian Marketing Services, 2025-08-20. https://www.experian.com/blogs/marketing-forward/audigent-curated-inventory-packages-arrive-in-dv360-delivered-through-index-marketplaces/ Accessed 2026-11-03.

[191] Audigent now offers turnkey open internet deals through Amazon DSP (blog post). Experian Marketing Services, 2025-08-25. https://www.experian.com/blogs/marketing-forward/audigent-now-offers-turnkey-open-internet-deals-through-amazon-dsp/ Accessed 2026-11-03.

[192] Comscore CognitivePMP Marketplace. Audigent (a part of Experian). https://audigent.com/comscore-cognitivepmp-marketplace Accessed 2026-11-03.

[193] SmartPMP (product page). Audigent (a part of Experian). https://audigent.com/smart-pmps Accessed 2026-11-03.

[194] ContextualPMP (product page). Audigent (a part of Experian). https://audigent.com/contextual-pmps Accessed 2026-11-03.

[195] Financial Times: TAG recognizes leading companies across ad industry worldwide for achieving 2026 certifications (news page). TAG (Trustworthy Accountability Group), 2026-03-06. https://www.tagtoday.net/news/financial-times-tag-recognizes-leading-companies-across-ad-industry-worldwide-for-achieving-2026-certifications Accessed 2026-11-03. Access: partial extract.

[196] New ways to enhance agency efficiency with Google Ad Manager (curation launch post). Google (Peentoo Patel, Director, Product Management, Google Ad Manager), 2024-11-04. https://blog.google/products/admanager/new-ways-to-enhance-agency-efficiency/ Accessed 2026-11-03.

[197] Understanding fees for using segments. Index Exchange (Knowledge Base), 2026-10-26. https://kb.indexexchange.com/marketplace_partners/deals/understanding_segment_fees.htm Accessed 2026-11-03.

[198] How Index Marketplaces work. Index Exchange (Knowledge Base), 2026-05-04. https://kb.indexexchange.com/marketplace_partners/what_are_marketplaces.htm Accessed 2026-11-03.

[199] How floors work in Index Marketplaces. Index Exchange (Knowledge Base), 2026-05-04. https://kb.indexexchange.com/marketplace_partners/auction_mechanics/how_floors_work_marketplaces.htm Accessed 2026-11-03.

[200] Index Marketplaces (product page). Index Exchange, 2026. https://www.indexexchange.com/platform/index-marketplaces/ Accessed 2026-11-03.

[201] How SWYM.ai Leveraged Index Marketplaces To Deliver 56% Boost in Client KPIs for Butler/Till. Index Exchange (editorial blog, 'Index Editor'), 2025-04-09. https://www.indexexchange.com/2025/04/09/swym-ai-leveraged-index-marketplaces/ Accessed 2026-11-03.

[202] How a Global Hotel Brand Cut Cost per Booking by 41% With dentsu, Chalice AI, and Index Cloud. Index Exchange (editorial blog, 'Index Editor'), 2026-10-06. https://www.indexexchange.com/2026/10/06/hotel-brand-dentsu-chalice-ai-index-cloud/ Accessed 2026-11-03.

[203] Context at Streaming TV Scale: PMG, IRIS.TV, and Index Exchange Drive Brand Lift. Index Exchange (editorial blog, 'Index Editor'), 2025-11-12. https://www.indexexchange.com/2025/11/12/context-streaming-tv-scale-pmg-iris-tv/ Accessed 2026-11-03.

[204] Adelaide Joins Index Marketplaces, Bringing Attention-Based Media Quality to Sell-Side Curation. Index Exchange Inc. (press release), 2026-01-07. https://www.indexexchange.com/press/press-releases/adelaide-joins-index-marketplaces/ Accessed 2026-11-03.

[205] About Index Exchange. Index Exchange, 2026. https://www.indexexchange.com/about/ Accessed 2026-11-03.

[206] Accessing demand through Index Marketplaces (publishers). Index Exchange (Knowledge Base), 2026-05-04. https://kb.indexexchange.com/publishers/marketplaces/access_marketplace_demand.htm Accessed 2026-11-03.

[207] Onetag partners with Lotame (release). Onetag, 2025-10-02. https://www.onetag.com/?p=6129 Accessed 2026-11-03.

[208] Services Privacy Notice. Lotame, a division of Epsilon Data Management, LLC, 2025-11-19. https://www.lotame.com/privacy/services-privacy-notice/ Accessed 2026-11-03.

[209] Lotame Unveils Curation to Enhance Global Advertising Performance and Solve Addressability Gaps. Lotame Solutions, Inc., 2024-08-14. https://www.lotame.com/resources/lotame-unveils-curation-to-solve-addressability-gaps/ Accessed 2026-11-03.

[210] Global CPG Brand Soaks Up Deep Consumer Insights (case study page: PHD and Lotame Curated Marketplaces). Lotame, 2025-02-25. https://www.lotame.com/resources/case-study-phd-lotame-curated-marketplaces/ Accessed 2026-11-03.

[211] Lotame Security Summary (Public V2.1.1). Lotame Solutions, Inc.. https://resources.lotame.com/hubfs/~Lotame.com%20Content/Memberships,%20Affiliations%20and%20Oversight/Lotame_Security_ISO27001_Summary.pdf Accessed 2026-11-03.

[212] Form 10-K, Magnite, Inc., fiscal year ended 31 December 2025. Magnite, Inc. (SEC EDGAR), 2026-02. https://www.sec.gov/Archives/edgar/data/1595974/000159597426000007/mgni-20251231.htm Accessed 2026-11-03. Access: partial extract.

[213] Culture Hive gains second SSP in 30 days as Magnite opens ClearLine to CRS (Luís Rijo). PPC Land, 2026-07-23. https://ppc.land/culture-hive-gains-second-ssp-in-30-days-as-magnite-opens-clearline-to-crs/ Accessed 2026-11-03.

[214] Form 10-Q, Magnite, Inc., quarterly period ended 30 June 2026. Magnite, Inc. (SEC EDGAR, accession 0001628280-26-053343), 2026. https://www.sec.gov/Archives/edgar/data/0001595974/000162828026053343/mgni-20260630.htm Accessed 2026-11-03. Access: partial extract.

[215] Magnite and Cognitiv Announce Deep Learning Integration for Real-Time Curation (GlobeNewswire release, Stock Titan copy). Magnite, Inc. and Cognitiv, 2026-01-06. https://www.stocktitan.net/news/MGNI/magnite-and-cognitiv-announce-deep-learning-integration-for-real-v8hg2q5tip4l.html Accessed 2026-11-03.

[216] Microsoft Monetize - Charges for Sellers. Microsoft Advertising (Microsoft Learn, xandr-docs), 2025-10-21. https://learn.microsoft.com/en-us/xandr/monetize/charges-for-sellers Accessed 2026-11-03.

[217] Microsoft Corporation Form 10-K for the fiscal year ended June 30, 2022. Microsoft Corporation (SEC EDGAR), 2022-07-28. https://www.sec.gov/Archives/edgar/data/789019/000156459022026876/msft-10k_20220630.htm Accessed 2026-11-03. Access: partial extract.

[218] Microsoft Corporation Form 10-K for the fiscal year ended June 30, 2026. Microsoft Corporation (SEC EDGAR), 2026-07-29. https://www.sec.gov/Archives/edgar/data/789019/000119312526323660/msft-20260630.htm Accessed 2026-11-03. Access: partial extract.

[219] MRC continuing accreditation letter to Microsoft Corporation (Microsoft Advertising), with TAG Certified Against Fraud addendum. Media Rating Council, Inc. (George W. Ivie), 2025-01-31. https://mediaratingcouncil.org/sites/default/files/Accreditation%20Letters/Microsoft/GI013125_Microsoft%20Advertising%20Accreditation%20Letter.pdf Accessed 2026-11-03.

[220] From Proprietary Assets to Profitable Media Business: The Power of Curation. Microsoft Advertising (blog; Nick Bolt, Senior Product Marketer), 2026-09-10. https://about.ads.microsoft.com/en/blog/post/september-2026/from-proprietary-assets-to-profitable-media-business-the-power-of-curation Accessed 2026-11-03.

[221] Putting our assets to work for publishers and advertisers. Microsoft Advertising (blog; Jennifer Creegan, General Manager, Global Marketing & Operations), 2023-06-18. https://about.ads.microsoft.com/en/blog/post/june-2023/putting-our-assets-to-work-for-publishers-and-advertisers Accessed 2026-11-03.

[222] Media Buyer Console: Create, Manage and Target Deals All-in-one-place. PubMatic, Inc., undated (Copyright © 2026). https://pubmatic.com/products/media-buyer-console/ Accessed 2026-11-03.

[223] PubMatic for Buyers: Gen AI-Powered Media Buying Platform for Advertisers & Agencies (solutions page). PubMatic, Inc., 2026. https://pubmatic.com/solutions/buyers/ Accessed 2026-11-03.

[224] Auction Packages Storefront. PubMatic, Inc., undated (Copyright © 2026). https://pubmatic.com/auction-packages/ Accessed 2026-11-03.

[225] The Power Of Optimization In Curation. PubMatic, Inc. (blog; Mike Chowla, VP, Product Management), 2025-03-19. https://pubmatic.com/blog/the-power-of-optimization-in-curation/ Accessed 2026-11-03.

[226] PubMatic Weighs In On The Debate Around Curation Fees. AdExchanger (Anthony Vargas), interviewing Rajeev Goel, co-founder and CEO of PubMatic, 2025-03-07. https://www.adexchanger.com/platforms/pubmatic-weighs-in-on-the-debate-around-curation-fees/ Accessed 2026-11-03.

[227] Comscore and Adelaide Expand Access to High-Quality, Attention-Optimized Curated Deals across PubMatic's platform. Comscore, Inc. (press release), 2025-06-10. https://www.comscore.com/Insights/Press-Releases/2025/6/Comscore-and-Adelaide-Expand-Access Accessed 2026-11-03.

[228] Video & VCR Refresh Auction Package (United States region page). PubMatic, Inc., undated. https://pubmatic.com/auction-packages/us/video-vcr-refresh-us/ Accessed 2026-11-03. Access: partial extract.

[229] CTV Auction Package (United States region page). PubMatic, Inc.. https://pubmatic.com/auction-packages/us/ctv-us/ Accessed 2026-11-03. Access: partial extract.

[230] PubMatic’s Connect Platform for Curation Delivers for Publishers: Drives 25%+ Higher eCPMs, Incremental Revenue and Diverse Demand. PubMatic, Inc. (press release), 2025-03-20. https://pubmatic.com/news/pubmatics-connect-platform-for-curation-delivers-for-publishers-drives-25-higher-ecpms-incremental-revenue-and-diverse-demand/ Accessed 2026-11-03.

[231] Form 10-K, PubMatic, Inc., fiscal year ended 31 December 2025. PubMatic, Inc. (SEC EDGAR), 2026-02. https://www.sec.gov/Archives/edgar/data/1422930/000142293026000010/pubm-20251231.htm Accessed 2026-11-03. Access: partial extract.

[232] Form 10-Q, PubMatic, Inc., quarter ended 30 June 2026. PubMatic, Inc. (SEC EDGAR, accession 0001422930-26-000030), 2026-08-06. https://www.sec.gov/Archives/edgar/data/1422930/000142293026000030/pubm-20260630.htm Accessed 2026-11-03.

[233] Our Deal Management Agent: Bringing Scaled AI-Adoption To PMPs, PGs, And Auction Packages. PubMatic, Inc. (blog; Yigit Gursoy, Senior Product Manager), 2026-04-28. https://pubmatic.com/blog/our-deal-management-agent-bringing-scaled-ai-adoption-to-pmps-pgs-and-auction-packages/ Accessed 2026-11-03.

[234] Stagwell Inc. Earnings Presentation, Second Quarter 2026 (Form 8-K, Exhibit 99.2). Stagwell Inc. (SEC EDGAR), 2026-07-30. https://www.sec.gov/Archives/edgar/data/876883/000087688326000023/a2q26earningspresentatio.htm Accessed 2026-11-03.

[235] EDGAR full-text search: "Stagwell Curate" (10-Q, 10-K, 8-K) and "Curate" in Stagwell filings, 1 January to 3 November 2026. U.S. Securities and Exchange Commission (EDGAR full-text search). https://efts.sec.gov/LATEST/search-index?q="Curate"&ciks=0000876883&dateRange=custom&startdt=2026-01-01&enddt=2026-11-03 Accessed 2026-11-03.

[236] Stagwell Inc. Annual Report on Form 10-K for the fiscal year ended December 31, 2025. Stagwell Inc. (SEC EDGAR), 2026-03-13. https://www.sec.gov/Archives/edgar/data/876883/000087688326000010/stgw-20251231.htm Accessed 2026-11-03. Access: partial extract.

[237] The Trade Desk Announces OpenSincera. The Trade Desk, Inc. (press room), 2025-05-13. https://www.thetradedesk.com/press-room/the-trade-desk-announces-opensincera Accessed 2026-11-03.

[238] Stagwell Inc. Earnings Presentation, Third Quarter 2026 (Form 8-K, Exhibit 99.2). Stagwell Inc. (SEC EDGAR), 2026-10-29. https://www.sec.gov/Archives/edgar/data/876883/000087688326000030/a3q26earningspresentatio.htm Accessed 2026-11-03.

[239] Audigent, a part of Experian, brings premium first-party audiences to Nexxen (blog post). Experian Marketing Services, 2025-08-18. https://www.experian.com/blogs/marketing-forward/audigent-a-part-of-experian-brings-its-premium-first-party-audiences-to-nexxen Accessed 2026-11-03.

[240] Data usage conditions. Index Exchange (Knowledge Base), 2026-05-04. https://kb.indexexchange.com/marketplace_vendors/data_partner_vendors/data_usage_conditions.htm Accessed 2026-11-03.

[241] Announcing ISO 27001 certification (Lotame resource URL; redirects to Schellman certificate directory entry 1571242-15). Lotame; Schellman (certificate directory). https://www.lotame.com/resources/announcing-iso-27001-certification Accessed 2026-11-03. Access: partial extract.

[242] TAG Recognizes Leading Companies Across Ad Industry Worldwide for Achieving 2026 Certifications (press release). Trustworthy Accountability Group (TAG), 2026-03-05. https://www.tagtoday.net/pressreleases/raising-the-bar-tag-applauds-2026-certified-leaders-elevating-industry-standards-for-fraud-brand-safety-malvertising-and-transparency Accessed 2026-11-03.

[243] Responsible Advertising & Data Governance. Magnite, Inc., undated (Copyright 2026). https://magnite.com/responsible-advertising-and-data-governance Accessed 2026-11-03.

[244] Comscore Launches Certified Deal IDs in Magnite's SSP, Bringing Trusted Content Rankings to Programmatic Buying. Comscore, Inc., 2025-05-05. https://www.comscore.com/Insights/Press-Releases/2025/5/Comscore-Launches-Certified-Deal-IDs-in-Magnites-SSP Accessed 2026-11-03.

[245] Xandr Brings Xandr Curate to Market Opening New Revenue Streams for Asset Owners (republished release). MarTech Series (author line 'Business Wire'; release text by Xandr), 2021-04-27. https://martechseries.com/sales-marketing/programmatic-buying/xandr-brings-xandr-curate-market-opening-new-revenue-streams-asset-owners/ Accessed 2026-11-03.

[246] Connect for Buyers: Improved ROI, Cookies Optional. PubMatic, Inc., undated (Copyright © 2026 PubMatic, Inc.). https://pubmatic.com/products/connect-for-buyers/ Accessed 2026-11-03.

[247] EDGAR submissions record, CIK 0000876883 (Stagwell Inc). U.S. Securities and Exchange Commission. https://data.sec.gov/submissions/CIK0000876883.json Accessed 2026-11-03.

[248] Curation for agencies (Beta). Google (Authorized Buyers Help), 2026. https://support.google.com/authorizedbuyers/answer/15282785 Accessed 2026-11-03.

[249] Deals API - IAB Tech Lab (standards page). IAB Tech Lab, 2026-02-06. https://iabtechlab.com/standards/dealsapi/ Accessed 2026-11-03.

[250] Transaction types. Google (Google Ad Manager Help), 2026. https://support.google.com/admanager/answer/2805834?hl=en Accessed 2026-11-03.

[251] ads.txt - Authorized Digital Sellers (standards page). IAB Tech Lab, 2026-09-14. https://iabtechlab.com/standards/ads-txt/ Accessed 2026-11-03.

[252] iabtechlab.com site search results for 'curation'. IAB Tech Lab (site search), 2026-11-03. https://iabtechlab.com/?s=curation Accessed 2026-11-03.

[253] The Evolution of Curation: From Static Deals to Real-Time Precision (member perspective). Keith Arrowsmith, Onetag (hosted by IAB Tech Lab, tagged member-perspective), 2025-09-23. https://iabtechlab.com/the-evolution-of-curation-from-static-deals-to-real-time-precision/ Accessed 2026-11-03.

[254] Nexxen Curated Marketplace (product page). Nexxen International Ltd., undated (copyright 2026). https://nexxen.com/nexxen-curated-marketplace/ Accessed 2026-11-03.

[255] Curation (Consumer View): Curated Deals, Deal Engine, Deal Marketplaces (product page). Experian Marketing Services (Experian Information Solutions, Inc.), undated (copyright 2026; hero image named 'CV-Curation-Hero-2026'). https://www.experian.com/marketing/consumer-view/curation Accessed 2026-11-03.

[256] cureon: the curation engine (product page). Multilocal Media Ltd. https://www.multilocal.media/cureon Accessed 2026-11-03.

[257] Sovrn partners with Multilocal to expand curated deals with commerce signals (PR Newswire release as reproduced). Sovrn Holdings, Inc. and Multilocal Media Ltd, via TechIntelPro, 2026-09-29. https://techintelpro.com/news/ai/enterprise-ai/sovrn-partners-with-multilocal-to-expand-curated-deals-with-commerce-signals Accessed 2026-11-03.

[258] Index Marketplaces - Smarter Supply. Better Outcomes.. Index Exchange, 2026. https://www.indexexchange.com/marketplaces/ Accessed 2026-11-03.

[259] Onetag homepage. Onetag. https://www.onetag.com/ Accessed 2026-11-03.

[260] FreeWheel Advertiser Suite (FreeWheel Curation Hub section). FreeWheel Media, Inc. (Comcast Advertising), undated (copyright 2026). https://www.freewheel.com/advertisersuite Accessed 2026-11-03.

[261] IAB Tech Lab Releases Deals API to Standardize Programmatic Deal Sync (press release). IAB Tech Lab, 2025-12-04. https://iabtechlab.com/press-releases/iab-tech-lab-releases-deals-api-to-standardize-programmatic-deal-sync/ Accessed 2026-11-03.

[262] Integral Ad Science homepage (Solutions navigation). Integral Ad Science, Inc., 2026. https://integralads.com/ Accessed 2026-11-03.

[263] sellers.json Supply Chain Transparency (standards page). IAB Tech Lab, 2026-10-07. https://iabtechlab.com/sellers-json/ Accessed 2026-11-03.

[264] openrtb2.x releases. IAB Tech Lab (GitHub repository InteractiveAdvertisingBureau/openrtb2.x), 2026-06-11. https://github.com/InteractiveAdvertisingBureau/openrtb2.x/releases Accessed 2026-11-03.

[265] docs.prebid.org sitemap.xml. Prebid.org. https://docs.prebid.org/sitemap.xml Accessed 2026-11-03.

[266] Commerce Grid - The world's first commerce SSP, built by Criteo. Criteo S.A., 2026. https://www.criteo.com/platform/commerce-grid/ Accessed 2026-11-03.

[267] Curation (Permutive product documentation). Permutive. https://docs.permutive.com/products/curation Accessed 2026-11-03.

[268] Samba Audiences (product page). Samba TV, Inc. (Samba), 2026-11-02. https://www.samba.com/audiences Accessed 2026-11-03.

[269] HypertailPMP (service page). Semasio GmbH (part of Samba). https://www.semasio.com/services/hypertail-pmp Accessed 2026-11-03. Access: partial extract.

[270] Semasio homepage. Semasio GmbH (part of Samba), 2025. https://www.semasio.com/ Accessed 2026-11-03.

[271] Introducing new pathways to programmatic CTV at scale. Walmart Connect, Ryan Mayward, GM & SVP, 2026-04-27. https://www.walmartconnect.com/resources/articles/2026/new-pathways-to-programmatic-CTV-at-scale Accessed 2026-11-03.

[272] Magnite partners with Walmart Connect to power Connect Select marketplace. Magnite, Inc. (Magnite Team blog), 2026-04-28. https://www.magnite.com/blog/magnite-partners-with-walmart-connect-to-power-connect-select-marketplace/ Accessed 2026-11-03.

[273] Marketplace overview - Display & Video 360 Help. Google LLC (Display & Video 360 Help). https://support.google.com/displayvideo/answer/6224754?hl=en Accessed 2026-11-03.

[274] Yahoo Advertising to launch Yahoo Backstage, enabling a direct path to premium publisher inventory. Yahoo Inc. (press), 2023-06-19. https://www.yahooinc.com/press/yahoo-advertising-to-launch-yahoo-backstage-enabling-a-direct-path-to-premium-publisher-inventory Accessed 2026-11-03.

[275] Yahoo Backstage (Yahoo DSP product page). Yahoo Inc.. https://www.yahooinc.com/yahoo-dsp/backstage Accessed 2026-11-03.

[276] Direct Access (Viant supply path optimization program). Viant Technology Inc.. https://www.viantinc.com/solutions/publisher/direct-access/ Accessed 2026-11-03.

[277] Advertising Demand - Sovrn, Inc.. Sovrn Holdings, Inc.. https://www.sovrn.com/advertising-demand/ Accessed 2026-11-03.

[278] Programmatic Transparency Benchmark: Q2 2026 Benchmark Findings (report page). ANA, 2026-08-26. https://www.ana.net/miccontent/show/id/rr-2026-08-programmatic-transparency-benchmark-q2 Accessed 2026-11-03. Access: partial extract.

[279] ANA Programmatic Transparency Benchmark (package page listing all editions). ANA, 2026. https://www.ana.net/micpackage/show/id/97114 Accessed 2026-11-03.

[280] The State Of The Open Internet (resource page). Jounce Media, undated. https://www.jouncemedia.com/resources/market-outlook Accessed 2026-11-03.

[281] Jounce Media home page (research, data, client portal). Jounce Media, undated. https://www.jouncemedia.com/ Accessed 2026-11-03.

[282] Form 10-K, Criteo S.A., fiscal year ended 31 December 2025. Criteo S.A. (SEC EDGAR), 2026-02. https://www.sec.gov/Archives/edgar/data/1576427/000157642726000014/crto-20251231.htm Accessed 2026-11-03. Access: partial extract.

[283] Second Programmatic Supply Chain Transparency Study (knowledge page). ISBA and PwC, 2023-01-18. https://www.isba.org.uk/knowledge/second-programmatic-supply-chain-transparency-study Accessed 2026-11-03. Access: partial extract.

[284] Form 10-K, The Trade Desk, Inc., fiscal year ended 31 December 2025. The Trade Desk, Inc. (SEC EDGAR), 2026-02. https://www.sec.gov/Archives/edgar/data/1671933/000167193326000014/ttd-20251231.htm Accessed 2026-11-03. Access: partial extract.

[285] Form 10-K, Viant Technology Inc., fiscal year ended 31 December 2025. Viant Technology Inc. (SEC EDGAR), 2026-03. https://www.sec.gov/Archives/edgar/data/1828791/000182879126000019/dsp-20251231.htm Accessed 2026-11-03. Access: partial extract.

[286] ANA Releases 2024 Programmatic Benchmark Study (press release). ANA, 2024-12-12. https://ana.net/content/show/id/pr-2024-12-programmatic Accessed 2026-11-03.

[287] ANA Benchmark Reference Guide and Methodology (TrueKPI Framework). Fiducia DLT Limited (for the ANA benchmark), undated. https://www.fiducia.eco/methodology Accessed 2026-11-03.

[288] Online platforms and digital advertising market study, final report, Appendix R: Fees in the adtech stack. UK Competition and Markets Authority (CMA), 2020-07-01. https://assets.publishing.service.gov.uk/media/5efb22ebe90e075c4e144c59/Appendix_R_-_fees_in_the_adtech_stack.pdf Accessed 2026-11-03.

[289] Curation - friend or foe? (blog post). The Ozone Project, 2025-07-10. https://ozoneproject.com/blog/curation-friend-or-foe/ Accessed 2026-11-03.

[290] Configure curation settings (Beta) - Google Ad Manager Help. Google. https://support.google.com/admanager/answer/15308224 Accessed 2026-11-03.

[291] TripleLift Finds 99% of Buyers Consider Deal Curation Important to Media Strategy, Yet Only 21% Rate Their Execution as Very Effective (press release). TripleLift, 2026-06-17. https://triplelift.com/press/triplelift-finds-99-of-buyers-consider-deal-curation-important-to-media-strategy/ Accessed 2026-11-03.

[292] PubMatic and Agence 79 Achieve 5X Higher CTR vs Open Market with Programmatic Video Curation in France (customer story: curation). PubMatic, Inc. (case study page), undated (copyright 2026; the companion release is dated 11 June 2026). https://pubmatic.com/case-studies/pubmatic-and-agence-79-drive-higher-ctr Accessed 2026-11-03.

[293] Multilocal case study: media owner (Maglr publication). Multilocal Media Ltd, unknown. https://multilocal.maglr.com/case-study-media-owner/media-owner Accessed 2026-11-03. Access: partial extract.

[294] KINESSO, Matterkind and Reprise come together to form new unified entity. KINESSO (IPG Mediabrands), Japan site, 2023-09-21. https://jp.kinesso.com/en/insights/kinesso-matterkind-and-reprise-come-together-to-form-new-unified-entity/ Accessed 2026-11-03.

[295] Offsite Display and Collective TV (offering page). Albertsons Media Collective. https://www.albertsonsmediacollective.com/offerings/offsite-display-and-collective-tv/ Accessed 2026-11-03.

[296] How Albertsons Media Collective Is Redefining Advertiser Value: Matching CPG Investment with Enterprise Media. Albertsons Media Collective (blog), 2025. https://www.albertsonsmediacollective.com/news/blog/blog-details/2025/How-Albertsons-Media-Collective-Is-Redefining-Advertiser-Value-Matching-CPG-Investment-with-Enterprise-Media/default.aspx Accessed 2026-11-03.

[297] Clean pipes, clear value: why OpenPath is a win for advertisers. The Trade Desk, Inc., 2026-02-24. https://www.thetradedesk.com/resources/clean-pipes-clear-value-why-openpath-is-a-win-for-advertisers Accessed 2026-11-03.

[298] Inventory tier brand suitability control (What's new). Amazon Ads, 2025-11-14. https://advertising.amazon.com/resources/whats-new/inventory-tier-brand-suitability-control Accessed 2026-11-03.

[299] Supply quality (Amazon Ads library news article). Amazon Ads. https://advertising.amazon.com/library/news/supply-quality Accessed 2026-11-03.

[300] OpenXBuild (product page). OpenX Technologies, Inc.. https://www.openx.com/openxbuild/ Accessed 2026-11-03.

[301] Complaint, OpenX Technologies, Inc. and OpenX Ltd. v. Google LLC, E.D. Va. No. 1:25-cv-1282, Dkt 1. OpenX Technologies, Inc. and OpenX Ltd. (plaintiffs); copy hosted by PPC Land, 2025-08-04. https://ppc.land/content/files/2025/08/OpenX-Technologies--Inc.--and-OpenX-Ltd.-v.-Google-LLC.pdf Accessed 2026-11-03.

[302] PubMatic Files Lawsuit Against Google to Restore Fair Competition and Innovation in Digital Advertising (press release, Exhibit 99.1 to Form 8-K). PubMatic, Inc., 2025-09-08. https://www.sec.gov/Archives/edgar/data/1422930/000142293025000044/ex991-pressrelease.htm Accessed 2026-11-03.

[303] Magnite Files Lawsuit Against Google (press release, Exhibit 99.1 to Form 8-K). Magnite, Inc., 2025-09-16. https://www.sec.gov/Archives/edgar/data/1595974/000159597425000035/a991pressreleaseofmagnitei.htm Accessed 2026-11-03.

[304] CourtListener RECAP search result: Index Exchange Inc. v. Google LLC, 1:25-cv-10477 (S.D.N.Y.). Free Law Project (CourtListener), 2025-12-17. https://www.courtlistener.com/?type=r&q=&docket_number=1%3A25-cv-10477&court=nysd Accessed 2026-11-03. Access: partial extract.

[305] Teads Files Lawsuit Against Google Seeking Financial Damages Following Federal Court Antitrust Ruling (press release, Exhibit 99.1 to Form 8-K). Teads Holding Co., 2026-08-03. https://www.sec.gov/Archives/edgar/data/0001454938/000145493826000048/exh991pr.htm Accessed 2026-11-03.

[306] IABTechLab/AAMP hub repository README (release and versioning model). IAB Tech Lab, 2026-09-22. https://github.com/IABTechLab/AAMP Accessed 2026-11-03.

[307] PubMatic’s AgenticOS Accelerates Globally as Agentic Campaigns Unlock Efficiency and Performance (press release). PubMatic, 2026-04-27. https://investors.pubmatic.com/news-releases/news-release-details/pubmatics-agenticos-accelerates-globally-agentic-campaigns Accessed 2026-11-04.

[308] Certifications (TAG programme overview). Trustworthy Accountability Group (TAG). https://www.tagtoday.net/certifications Accessed 2026-11-03.

[309] ANA Q4 2025 Programmatic Transparency Benchmark Shows Media Quality Is Now Driving Measurable Performance Outcomes (press release). ANA, 2026-02-25. https://www.ana.net/content/show/id/pr-2026-02-programatic Accessed 2026-11-03.

[310] ANA Industry Benchmark: Glossary of Key Terms. Fiducia DLT Limited (for the ANA benchmark), undated. https://www.fiducia.eco/glossary Accessed 2026-11-03.

[311] Adalytics blog index (pages 1 and 2). Adalytics Research, undated. https://adalytics.io/blog Accessed 2026-11-03.

[312] PubMatic Launches AgenticOS, the Operating System for Agent-to-Agent Advertising (press release). PubMatic, 2026-01-05. https://investors.pubmatic.com/news-releases/news-release-details/pubmatic-launches-agenticos-operating-system-agent-agent Accessed 2026-11-04.

[313] PubMatic and Havas Launch the First Agentic CTV Campaign in Spain for Telefónica, Achieving 18% Lower CPM Vs Target (press release). PubMatic, 2026-06-09. https://investors.pubmatic.com/news-releases/news-release-details/pubmatic-and-havas-launch-first-agentic-ctv-campaign-spain Accessed 2026-11-04.

[314] (Author's work) Who Decides on the Sell Side: published capability matrix and Deal Quotient code (data/capability-matrix.json, build/quotient.mjs). Evgeny Popov, No Fluff Advisory, 2026-11-03. https://nofluffadvisory.com/research/sell-side-curation-2026/data/capability-matrix.json Accessed 2026-11-04.

[315] Experian: Full-year financial report (RNS, Investegate). Experian plc (via Investegate), 2025-05-14. https://www.investegate.co.uk/announcement/rns/experian--expn/full-year-financial-report/8877075 Accessed 2026-11-03. Access: partial extract.

[316] Experian: Full-year financial report (RNS, London South East). Experian plc (via lse.co.uk), 2025-05-14. https://www.lse.co.uk/rns/EXPN/full-year-financial-report-cj8g1sp1zh5crsc.html Accessed 2026-11-03. Access: partial extract.

[317] Create meta-deals (Equativ Help Center, Maestro curation). Equativ, 2026-08-05. https://help.equativ.com/manage-deals/create-meta-deals Accessed 2026-11-03.

[318] A First-to-Market Win: dentsu X and Dos Equis Tap Gracenote To Reach College Football Viewers. Index Exchange (editorial blog, 'Index Editor'), 2026-04-27. https://www.indexexchange.com/2026/04/27/market-win-dentsu-x-dos-equis-tap-gracenote/ Accessed 2026-11-03.

[319] TAG Registry. TAG (Trustworthy Accountability Group). https://tagtoday.net/registry Accessed 2026-11-03. Access: partial extract.

[320] Publisher release roundup: Q1 2026 edition (exact heading not captured; title from the URL slug). Microsoft Advertising (blog; Nick Bolt), 2026-04-14. https://about.ads.microsoft.com/en/blog/post/april-2026/publisher-release-roundup-q1-2026-edition Accessed 2026-11-03.

[321] Microsoft Curate (January 21, 2026) DV360 Central Partner ID Support for Multi-Buyer Deals. Microsoft Advertising (Microsoft Learn, xandr-docs), 2026-01-21. https://learn.microsoft.com/en-us/xandr/curate/release-notes-2026-01-21 Accessed 2026-11-03.

[322] Platform Privacy Policy. Predictive Pop, Inc., doing business as Audigent, 2026-09-02. https://audigent.com/platform-privacy-policy Accessed 2026-11-03.

[323] Deal Sync (Commerce Grid knowledge base). Criteo S.A. (docs.commercegrid.criteo.com). https://docs.commercegrid.criteo.com/kb/guide/en/deal-sync-20YlGkM5L3/Steps/2888995 Accessed 2026-11-03.

[324] Advertisers put SSPs and curators under the microscope in sell-side push for ad tech fee transparency (OpenX news page reproducing a Digiday item). Digiday, reproduced on openx.com, 2025-03-12. https://www.openx.com/news/advertisers-put-ssps-and-curators-under-the-microscope-in-sell-side-push-for-ad-tech-fee-transparency/ Accessed 2026-11-03. Access: partial extract.

[325] Terms and Conditions (website). Index Exchange, 2026. https://www.indexexchange.com/terms Accessed 2026-11-03.

[326] Publicis to acquire Lotame the world's leading independent end-to-end data solution (Epsilon pressroom copy). Epsilon (Publicis Groupe), 2025-03-06. https://www.epsilon.com/us/about-us/pressroom/lotame-acquisition Accessed 2026-11-03.

[327] California Data Broker Registry, registration data (registry.csv): Epsilon Data Management, LLC (Broker ID 1026). California Privacy Protection Agency (registry); entry submitted by Epsilon Data Management, LLC, 2026-10-30. https://cppa.ca.gov/data_broker_registry/registry.csv Accessed 2026-11-03. Access: partial extract.

[328] Xandr Brings Xandr Curate to Market Opening New Revenue Streams for Asset Owners (Business Wire original). Xandr (Business Wire), 2021-04-26 (from the URL; not read). https://www.businesswire.com/news/home/20210426005974/en/Xandr-Brings-Xandr-Curate-to-Market-Opening-New-Revenue-Streams-for-Asset-Owners Accessed 2026-11-03. Access: partial extract.

[329] EDGAR company filings list, Microsoft Corp (CIK 0000789019), Form 10-K. U.S. Securities and Exchange Commission (EDGAR). https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0000789019&type=10-K&dateb=&owner=include&count=10 Accessed 2026-11-03.

[330] EDGAR company filings list, PubMatic, Inc., Form 10-Q. U.S. Securities and Exchange Commission (EDGAR), live record. https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=1422930&type=10-Q&dateb=&owner=include&count=10 Accessed 2026-11-03.

[331] Stagwell Inc. Definitive Proxy Statement (Schedule 14A) for the 2026 Annual Meeting. Stagwell Inc. (SEC EDGAR), 2026-04-28. https://www.sec.gov/Archives/edgar/data/876883/000110465926050360/tm261497-1_def14a.htm Accessed 2026-11-03. Access: partial extract.