No Fluff Advisory · Research · Verification Plain language

Research paper · November 2026

Verified Against What

ad verification sells buyers a verdict on whether an impression was viewable, valid, safe and suitable. This paper tests, product by product, what each verdict is accredited for, what independent evidence supports it, what third parties can see inside the large platforms, and whether a verdict changes what anyone pays.

Evgeny Popov · No Fluff Advisory

Research cutoff 11 November 2026 · Only information available by that date is used

Short on time? Read the plain-language version: the same findings in fewer, shorter words.

1458sources opened and registered
804claims in the research ledger
17vendors scored
20corrections logged

Abstract

Abstract

What does an ad-verification verdict establish, who can check it, and when does it decide money? This paper examines the US market through public product documents, standards, studies and contracts. Its evidence cutoff is 11 November 2026; status-sensitive facts are rechecked on 11 December 2026 and any change is published as a dated correction. It keeps measurement scope, detection evidence and contractual authority separate.

In short

The merged sold-unit grid contradicts H1a: 108 of 298 documented units hold MRC accreditation, or 36.2% . H1b is contradicted on the author's readings of Peer39's pre-bid wording and Protected by Mediaocean's CTV display claim, rather than on a regulator's finding . H2 is supported on the author's registered listed-terms reading, with determining effect in 2 of 16 readable entries and wider-contract sensitivities reported alongside . The independent sources screened supplied no qualifying sold-category error-rate test for a scored product, supporting H3 . TikTok's documented measurement-kit route contradicts H4's universal platform-supplied-data claim, at medium confidence . DoubleVerify's separation statement and ISBA's review pledge contradict H5 and H6 under the author's stated interpretations, with stricter readings disclosed . H7 remains unresolved because production claims and partial buyer reporting leave the verification-verdict-to-agent-decision link unestablished .

The question and the method

A verification verdict can classify traffic, measure an opportunity to see an ad or apply a content policy. The buyer also needs its observation route, measurable population, selected settings and permitted action. A vendor badge alone supplies none of those fields . The author's essays on verification's pricing power and advertising's missing adjudicator are rival framings tested against public terms, not corroborating evidence .

The hypotheses and scoring anchors were registered on 9 November 2026. AI agents produced 10 workstreams and 7 deep profiles under the author's direction. The source register holds 1458 records; the claim ledger holds 804 claims. Claude agents researched the workstreams and profiles and scored the calibration rows with separate refute-mode verifiers. An OpenAI model completed the later sweep checks, scored and self-checked the added rows, and fact-checked the chapters. The author checked the work and made the inclusion and verdict calls. Chapter 13 distinguishes those roles from a blind second score.

What the paper found

H1a: accredited coverage is below the rule

The common product-by-metric-family-by-environment grid records 80 accredited units out of 179 sold by suites (44.7%), 25 of 82 for IVT specialists (30.5%) and 3 of 37 for suitability specialists (8.1%). Pooled coverage is 108 of 298 (36.2%). H1a requires at least 75% pooled and no cohort below 50%; the merged result fails both conditions . illumaSentinel's sold-unit denominator remains unknown, rather than zero. The result describes the documented grid and MRC scope, not a spend-weighted share of impressions. Chapter 2.

H1b: scope wording requires interpretation

The author reads Peer39's reference to accredited measurement within a pre-bid signal as implying a pre-bid accreditation absent from the applicable Adloox record. A narrower reading could mean an accredited post-serve input feeding that signal. Protected by Mediaocean's CTV display-impression wording also exceeds scope on a literal reading. Those author interpretations make H1b contradicted; they establish no MRC compliance decision . Read the letter for the product, metric, environment, format, placement, language and standard. Chapter 2.

H2: some reports decide money

The listed-terms test supports H2 at 2/16, or 12.5%, under the author's ruling. The Trade Desk's supply-side clause gives accredited reports conclusive effect; Microsoft Invest had a partner-report refund route whose end Microsoft announced. These counterexamples defeat a universal claim that verification moves nothing . Wider public-contract populations and review-qualified or trigger-only clauses can change the share across the 25% threshold. No paid third-party-verdict settlement or matched price difference was found in the screened record; that bounded absence cannot establish what happens in private contracts . Chapter 4.

H3: public error-rate evidence remains missing

H3 is supported in the sources screened to 10 November 2026: no independent test reports a scored product's sold-category error rate with a labelled set and published denominator . Failure cases, measured prevalence and disagreement between vendors answer different questions. Smith and co-authors studied publisher-side labels with partly constructed overall ratings and small overlaps; their result measures agreement, not accuracy . The candidate IAS bot miss rates concern its unscored publisher tool and omit session counts . Vendor-protected comparisons use the protecting classifier and cannot reveal what it missed . Chapter 3.

H4: the route differs by platform

Meta, YouTube and Amazon document platform-supplied routes with different limits. TikTok's documented Open Measurement SDK integration lets a verifier's tag collect signals, contradicting H4 under its literal comparator . The route statements date from 2021–2022, TikTok controls the integration, and no 2026 source opened restates the full route. Snap and LinkedIn trackers leave the firing location unstated; a named partner on other platforms supplies no missing route. Chapter 5.

H5: a narrow published safeguard exists

DoubleVerify's Supply-side line and IAS's Publisher channel show seller revenue in FY2024, with IAS's line also including Publica's ad serving . The author accepts DoubleVerify's June 2024 structural-separation statement as the published safeguard that contradicts H5. It names no scope, responsible owner or enforcement for that separation and does not address the pending Nielsen-parent buyer. Ownership and seller revenue establish governance questions, not measurement bias . Chapter 7.

H6: a successor depends on the process test

ISBA's March 2025 guide publishes a floor and risk tiers. The author accepts its review pledge, named overseer and update trigger as a stated change process, making H6 contradicted. Requiring a published decision procedure instead leaves it unresolved. Separately, product references to GARM do not establish which dated definitions each implements . The public framework's existence therefore supplies no automatic evidence about a vendor's implemented classifier. Chapter 8.

H7: agent buying has an unresolved evidence link

The screened record contains schemas, connectors, tests and vendor production claims. Expanded reporting adds Butler/Till campaign context and quality-analysis work, but does not establish which actual verification verdict changed the buying agent's decision. No identified agreement giving that verdict financial force was found. The author retains unresolved under the stated interpretation of the rule's ambiguous named-buyer clause . Schema representation alone is already too narrow a description of the work. Chapter 10.

What a buyer can use

The Verdict Quotient scores 17 products on 12 dimensions within separate cohorts. Only suitability specialists meet the band-size gate. The field values are 60 for suites, 53 for IVT specialists and 39 for suitability specialists; cross-cohort comparison is invalid. The quotient measures documented capability, without estimating accuracy. Chapter 11 exposes the cells and sensitivity tests. Chapter 12 turns the evidence into requests for scoped letters, labelled tests, data routes, correction rights and the governing contract.

Interests

The author has worked at Samba TV, now branded Samba, since May 2025. Nielsen competes with it in TV measurement. DoubleVerify's agreed sale to Neptune BidCo US Inc., Nielsen's parent, remained pending at the 10 November 2026 check . He co-leads AgenticAdvertising.org's Signals & Measurement working group; No Fluff Advisory operates signals and sales agents in its registry . He previously worked at Verve Group, Hearts & Science and Lotame. No vendor funded, reviewed or saw the paper; every vendor was scored under the same published rules. Claude and OpenAI models performed the AI-assisted research and checks under his direction, and he checked the work. There was no human second scorer. The full interests and process statement appears in Chapter 13.

Chapter 1

What a verification verdict is

Ad verification sells a judgement about an ad impression against a stated condition, followed by a report or a control that acts on it. The conditions include viewability, invalid traffic, brand safety and suitability, while the delivery ranges from a list applied before bidding to measurement after an ad serves . This paper treats the condition, the data route and the power to act on the result as separate questions. Its comparisons follow the service a buyer purchases, with suites, invalid-traffic specialists and suitability specialists in separate cohorts.

In short

A viewable impression meets a pixel-and-time rule that establishes an opportunity to see an ad . Invalid-traffic filtration removes activity that falls within the Media Rating Council's general or sophisticated categories . Brand safety applies an exclusion floor, while suitability lets a brand choose its tolerance within content categories . A property classification and a classification of the particular image, video or audio next to an ad have different scopes . Tags, server records, software development kits and platform integrations expose different evidence to the company making the judgement . The buying-task register keeps those services apart from media sold with an embedded safety filter, publisher account protection and the platforms' own controls. A verifier's report, an exclusion, a block and a contractual right to money are separate uses, tested in Chapter 4.

The condition behind the label

A verdict needs a question narrow enough to answer. Was an impression in view for the required time? Did the measured traffic meet an invalid-traffic rule? Did the adjacent content fall into a category the advertiser excludes? The paper uses verdict for the resulting classification, measure or decision, together with the condition and scope that give it meaning. It records who measured the event, what the buyer selected and what action followed. A label without those fields leaves the purchase hard to assess.

The four metric families describe different properties of an impression. The Media Rating Council (MRC) publishes viewability and invalid-traffic requirements, while its content-level supplement addresses context, adjacency and brand safety . A campaign can therefore need more than one test. Read a vendor's claim of quality by asking which condition it covers, which impressions were measurable and what happened to those the service could not classify. The size of the reported percentage depends on that denominator.

For display ads, MRC's desktop guidelines version 2.0 of 18 August 2015 generally require at least 50% of the pixels in view for one continuous second, with a 30% pixel exception for large display ads of at least 242,500 pixels . Video generally requires at least 50% of the pixels in view for two continuous seconds while the video plays . The mobile guidelines of 28 June 2016 apply to mobile web and in-app measurement and retain format-specific pixel and duration rules . A buyer should record the format, environment and applicable rule before comparing rates.

MRC describes viewability as an opportunity to see, so the result alone establishes neither a person's attention nor a later response . The denominator also matters within viewability. The mobile guidelines require reporting of measurable and non-measurable impressions, and distinguish a measurable impression from one that meets the viewable threshold . A high rate among measurable impressions leaves a separate question about the share that could be measured. The attention paper owns attention scores and tests of attention, which this paper leaves outside its buying tasks.

Invalid traffic (IVT) is divided into general invalid traffic (GIVT) and sophisticated invalid traffic (SIVT) in MRC's June 2020 addendum . GIVT includes traffic identified through routine lists or standardised checks, while SIVT calls for more advanced analysis, corroboration or human intervention . The categories cover measurement exclusions as well as intentionally deceptive activity, and MRC requires services to disclose their decision rates and limitations . This paper calls an impression that survives the applicable filtration classified as valid. That wording keeps the reported classification attached to the process that produced it.

Safety and suitability require a content policy as well as a measurement method. ISBA's March 2025 Responsible Media Guide distinguishes a common safety floor from brand-specific suitability and publishes high, medium and low risk tiers . The floor identifies content to avoid, while the tiers express the advertiser's tolerance within categories . A buyer's selected tier belongs beside the result. Otherwise two reports can differ because the advertisers asked different questions, even before considering the classifiers' performance.

What was classified

The object classified can be a domain, an app, a page or the particular content adjacent to the ad. MRC's October 2025 representation policy distinguishes property-level ad verification based on text and contextual classification from enhanced content-level work on images, video and audio . Its 2018 supplement addresses discrete content measurement, metadata, adjacency, data freshness and human intervention . This paper records that object explicitly, because a conclusion about a property has a different evidential reach from a conclusion about a particular video.

The content-level supplement also allows a third-party service to use validated data from the platform, subject to requirements on the data and measurement . That makes the source of the information part of the result. Ask whether the verifier observed the content itself, received a content identifier, or analysed a platform-provided record. A platform integration can support a useful result, but its fields and exclusions still define what the verifier can establish. Chapter 5 records those routes by platform and metric.

The paper uses independent verification for measurement by a party other than the media seller or platform, and first-party validation for the seller's or platform's own measurement. This is a definition of the measurement relationship. It does not settle who funded a study, who pays the vendor, who owns it, or who can overturn its report. Those are separate fields. A buyer can use both types of measurement and still need to know which party controls the underlying observations.

When the verdict acts

A pre-bid service affects what the buyer considers before purchase. HUMAN's MediaGuard, for example, receives information from an OpenRTB bid request and returns an IVT prediction that the platform can use to drop the request . Peer39 sells categories applied through buying platforms, and Proximic describes pre-bid inventory filters for media buyers . The decision is made against information available at that point. A later report about the served impression answers a different question.

Post-bid measurement records an impression after the purchase or serve event. HUMAN's FraudSensor collects device, network and behavioural signals through detection tags and returns IVT classifications in reports . Pixalate documents tags and server-to-server event submission for its post-bid Analytics product, alongside separate pre-bid data feeds . Post-bid blocking is another action again. IAS describes pre-bid filtering and post-bid blocking for content outside a brand's settings, while Adloox describes page-context blocking and exclusion .

The timing should stay attached to the metric. A product can sell an IVT prediction before bidding, viewability measurement after delivery and suitability lists for a platform to execute. A buyer should ask which component produced each report and whether blocking was enabled for that placement. Merely finding the vendor's tag in an ad does not settle those settings. The useful record pairs the delivered metric with its collection route, stage and permitted action.

How the evidence reaches the verifier

A tag runs code or sends a request when the ad or page loads. HUMAN documents detection tags that gather device, network and behavioural signals, while Pixalate documents JavaScript tags, pixels, video verification scripts and server-submitted events . Each route has a distinct integration requirement. In the buying record, keep the tag's event separate from any model or classification subsequently applied to it. That makes it possible to ask which observations a missing or blocked tag would remove.

The Open Measurement Software Development Kit (OM SDK), maintained by IAB Tech Lab, provides measurement signals across supported apps and video environments . The Tech Lab's May 2024 CTV release describes signals such as device type and TV-off detection and names platform-specific integrations . Support for the SDK therefore needs an implementation record for the relevant player or device. Chapter 6 examines the CTV and in-app routes, including the distinction between an ad server's event and a device observation.

A server record can carry an impression identifier, time, app or device information supplied by another system. Pixalate explicitly supports server-to-server submission for platforms that cannot run a client-side tag . A crawler and classifier instead inspect content and assign categories. IAS describes analysis of page content and app metadata, supplemented by human review, and Channel Factory describes video classification used to build YouTube inclusion lists . Record the input and the output separately so that a source's description of a collection method does not become evidence of classification accuracy.

Platform feeds and application programming interfaces (APIs) form another route. Sightly says its YouTube suitability classifications use contextual signals from the YouTube Data API and that Google does not review those classifications . Its self-service list builder is consequently a different service from a platform's own safety control or a report measuring specific ad impressions . The table below preserves that distinction between selecting content and measuring delivery.

What the assurance covers

The paper reserves accreditation for MRC's scoped service record. MRC's representation policy distinguishes property-level from content-level accreditation and requires vendors to state the relevant capabilities and limits . TAG's Brand Safety Certified programme instead certifies a company's material operations in a stated geography, at entity level . The issuer, programme, entity and scope should therefore accompany the seal. Treat a vendor's own statement about compliance as self-attestation until the named body's record establishes what assurance it granted.

These assurance records answer a further question from the metric itself. A buyer still needs to know which product, setting and environment the purchase uses. A company-level seal cannot fill the product's missing method, and a method document cannot supply a test's error rate. The scoring rules give those attributes separate homes so that the same assurance does not earn credit throughout a row.

The service being bought

The register groups products by the output purchased. Cohort A contains suites sold to buyers for viewability, GIVT and SIVT, and at least one safety or suitability service across two or more environments. Cohort B contains impression-level IVT measurement or filtering for advertisers, agencies or trading platforms. Cohort C contains content safety or suitability classification sold for verification, exclusion or blocking. A product has one row for its buying task, even where it bundles several delivery components.

The test for a suitability product is whether the buyer can purchase the classification or control apart from managed media. Pixability's Self offer lets advertisers activate contextual and suitability data through their own Google Ads or Display & Video 360 account . Sightly's self-service offer includes YouTube inclusion and exclusion lists, and Channel Factory's August 2026 terms describe software accounts and paid plans alongside its services . These are recorded separately from the companies' managed campaigns and curated media offers.

HUMAN and Pixalate remain in the IVT cohort under this paper's primary-output rule, even though their menus also describe viewability or other quality signals . Peer39's suitability row and its Adloox verification suite remain separate purchased services, with Peer39's June 2026 acquisition announcement describing the addition of viewability and fraud detection . The method applies to the named service. Ownership alone supplies no missing capability, and one row's evidence cannot fill another row's cells.

Q2 records the products in the scoring population below. Product capabilities, delivery and environment descriptions are vendor documentation, unless the evidence column identifies a filing. The table describes the offer and leaves accreditation coverage to Chapter 2. Owner unknown means ultimate control was not established in the register, rather than an inference of independence. DSP means demand-side platform, the software through which a buyer bids. CTV means connected television. MFA means made-for-advertising. Environment labels in this table describe the vendor's offer, with each metric still requiring its own scope check.

ProductRecorded owner or operatorCohortMetrics soldMethod and stageBlocks or reports; vendor-described environmentsEvidence
DoubleVerify advertiser verificationDoubleVerify Holdings, public company at the baseline readAIVT, viewability, safety and suitabilityPre-bid segments and post-bid measurementAvoidance, blocking and reporting; web, apps, CTV and platform placements
IAS advertiser verificationNovacap affiliates, through Igloo Group ParentAIVT, viewability, safety and suitabilityTags, classification and platform integrations, pre-bid and post-bidFiltering, blocking and reporting; web, apps, CTV, audio and platforms
Adloox verificationPeer39AIVT, viewability, safety and suitabilityMeasurement, page analysis and DSP segments, pre-bid and post-bidExclusion, blocking and reporting; web, mobile and video/CTV offer
Protected by MediaoceanMediaoceanAIVT, viewability and suitabilityDetection and verification suite, post-bid with pre-bid controlsFiltration, controls and reporting; desktop, mobile web, apps and CTV, metric-specific
mFilterIt Ad Traffic ValidationmFilterIt; control unconfirmedAIVT, viewability, MFA, safety and suitabilityPre-bid avoidance and post-bid validationAvoidance and reporting; display, video/CTV and apps
HUMAN Ad Integrity SuiteHUMAN Security; control unconfirmedBGIVT and SIVT, with other quality signalsBid-request predictions and post-bid tagsPlatform-applied filtering and reporting; environment-specific implementations
Pixalate ad fraud protection and analyticsPixalate; control unconfirmedBIVT and display viewability, with quality analyticsPre-bid feeds and post-bid tags or server eventsBlocking feeds and reporting; websites, apps and CTV
Fraudlogix IVT detectionFraudlogix; control unconfirmedBGIVT and SIVT, with quality flagsPost-bid pixel analytics and pre-bid IP listsPlatform-applied blocking and reporting; high-volume real-time bidding traffic
Escalated IVT scoringEscalated Solutions; control unconfirmedBIVT and viewabilityPre-bid scoring and post-bid monitoringBlocking and reporting; display, video, apps and CTV
FouAnalyticsFounder-operated; control unconfirmedBIn-ad IVT analyticsJavaScript tag and post-bid classificationReporting; ads and websites where its tag executes, separate landing-page measurement for restricted platforms
Zefr brand safety and suitabilityZefr; control unconfirmedCContent safety and suitabilityClassification, pre-impression exclusions and post-campaign measurementExclusion and reporting; social-platform content
Peer39 brand safety and suitabilityP39 Tech; ownership self-description in registerCSafety and suitability categoriesPre-bid content classification and post-buy reportingBuyer-applied exclusions and reporting; web and CTV, by product scope
Proximic by ComscoreComscoreCSafety, suitability and misinformation filtersPre-bid contextual segmentsBuyer-applied exclusion; desktop, mobile, CTV and podcasts
illumaSentinelIlluma Technology; control unconfirmedCPage-level safety and suitabilityPre-bid classifier, activated in The Trade DeskAvoidance and reporting; open-web pages
ViewIQChannel Factory; Truelink investment, control unconfirmedCYouTube safety and suitabilityPre-bid classification and inclusion/exclusion listsBuyer-platform execution and reporting; YouTube
Pixability brand suitabilityPixability; control unconfirmedCYouTube suitabilityContextual segments and inclusion/exclusion lists before biddingBuyer-account activation and reporting; YouTube
Sightly Brand MentalitySightly; control unconfirmedCSuitability SmartListsYouTube API-based classification and list buildingInclusion/exclusion targeting; YouTube and programmatic offers, with separate CTV deals

Mantis Sentinel, a monitoring tag from Reach's Mantis, is listed rather than scored: its advertiser page describes the tag and its page-level output but gives neither separate purchase terms nor a live or beta status, which the scored-set rule requires.

A product comparison should hold the purchased task constant. Start with the output the buyer receives, then check the metric, environment, timing and action that make it useful. A publisher's account-protection tool and an advertiser's measurement service can both detect suspicious activity while answering different buying needs. The row boundary prevents that shared vocabulary from turning into a comparison the evidence cannot sustain. It also prevents a broad suite from appearing again as several specialist competitors.

For the same reason, the paper separates content inclusion from an observation of delivery. A list can express a brand's intended policy, while a report can describe what appeared during a campaign. To connect the two, ask for the settings applied, the period they covered and the relevant delivery record. That is a request for evidence about the named use. It should travel with the buyer's question before a coverage claim or summary score is used in procurement.

Who pays, and what the totals contain

The paying customer can be an advertiser, agency or trading platform, while publishers also buy measurement and quality tools. DoubleVerify's FY2025 filing separates advertiser Activation and Measurement from Supply-side revenue, and IAS's FY2024 filing separates Optimization, Measurement and Publisher revenue . Those labels describe different commercial perimeters. The buyer of a report and the party whose inventory it assesses can therefore sit in different contractual relationships with the same vendor.

DoubleVerify filed worldwide revenue of US$748.3 million for calendar 2025, including US$71.3 million from Supply-side customers . IAS filed worldwide revenue of US$530.1 million for calendar 2024, including US$76.5 million in its Publisher channel . The IAS line includes Publica's ad serving, while DoubleVerify's Supply-side line is a customer grouping . These company totals have stated years and scopes. They cannot establish the size of a common verification market or a market share without a comparable denominator.

Revenue also uses different units from product activity. DoubleVerify charges advertisers a measured transaction fee tied to the ads measured, while IAS describes recognising revenue from a fee per thousand impressions multiplied by measured volume . A buyer should keep the vendor's billing base beside its detection rate and its reported volume. Chapter 7 examines customer types, owners and safeguards, with the transaction record dated to the evidence read.

The register's boundaries

The product register holds 45 records, of which 19 are listed and 9 historical. A listed record documents a boundary decision, rather than a score of zero. Anura's Ad Tag Protect, for example, protects a publisher's own advertising account by withholding ads from flagged traffic, and its visitor-scoring offer addresses site visits . The scored IVT task instead requires a documented impression product serving a buyer or trading platform.

Contextual media and curated deals have their own papers in this series. GumGum's contextual advertising page and Seedtag's safety policy describe protections within their media offers . This paper requires a separately bought verification, exclusion or blocking service before treating that signal as a suitability row. Attention scores, content-rights protection and a platform's own advertising controls remain separate buying tasks. A broad use of the word safety cannot decide the product's cohort.

Historical records preserve earlier names without creating additional live competitors. Oracle's August 2024 end-of-life notice names Moat Analytics and Oracle Contextual Intelligence and ends all Oracle Advertising services on 30 September 2024 . DoubleVerify's filing records its purchases of Meetrics and OpenSlate . The register also keeps Method Media Intelligence as historical, with current commercial status unresolved after the domain served an unrelated IT-services page when screened . A dead or redirected page is an access and status observation. It cannot, alone, establish a corporate dissolution.

The questions a verdict has to survive

The author's essay Where the Surplus Settles argues that advertising verification has little pricing power when its verdict does not change a price . The Risk You Can Price identifies a missing adjudicator for disputed conversion counts, a narrower claim about cross-company outcome settlement . Here those are rival framings to test. They set questions about contractual authority, payment and redress, rather than conclusions about every verification product. Chapter 4 examines the actual terms that could support or defeat them.

H1a asks how much of the sold product-by-metric-by-environment grid holds MRC accreditation. H1b asks whether public accreditation claims stay inside the applicable record. Chapter 2 puts the service, metric, environment, format, placement, language and named standard beside each scope. That establishes what the assurance covers before using a company logo as shorthand.

H3 asks what independent tests establish about detection. Chapter 3 keeps a labelled error-rate test apart from a failure example, a prevalence study or agreement between classifiers. H4 asks how third parties obtain platform data and what they can do with it. Chapter 5 records the route for each partner, while Chapter 6 examines the device and server dependencies in CTV and apps.

H5 asks who pays the verifiers and what safeguards they publish, in Chapter 7. H6 asks whether a maintained common suitability framework exists after GARM, in Chapter 8. Chapter 9 keeps allegations, letters, procedural decisions and findings on the merits in their own records. They inform the questions a buyer asks and remain outside the Verdict Quotient.

H7 asks whether agent buying makes a verification result binding in an executed transaction, in Chapter 10. The Verdict Quotient compares documented capability within each buying task in Chapter 11, and Chapter 12 turns the evidence into requests for methods, scope, authority and correction rights. Chapter 13 records the evidence dates, verifier passes and unresolved items. Status is as read in November 2026; a recheck on 11 December 2026 is published as dated corrections.

Chapter 2

What MRC accreditation certifies

The Media Rating Council (MRC) accredits measurement processes against named standards and scopes. Its letters publish no product error rate . On the lists read on 10 November 2026, content-level brand safety and suitability accreditation is confined to YouTube . On this paper's reading, Peer 39's wording implies pre-bid accreditation beyond the listed scope, contradicting H1b . Every status is dated to its read and will be rechecked on 11 December 2026.

In short

MRC's letters certify conformance with named standards and publish no error rate . Its standards and guidance read on 10 November 2026 span February 2012 to July 2026 . TAG's four July 2026 certification programmes cover operating companies . Only Google's first-party controls and Zefr's third-party reporting hold content-level accreditation, both on YouTube with different placement and language limits . GumGum withdrew and its accreditation was revoked in September 2025 . IAS's property-level row has no current letter basis . H1b is contradicted on this paper's interpretation of Peer 39's wording and a literal reading of Protected by Mediaocean's display-impression claim on CTV . H1a is contradicted on the merged documented grid, with the denominator limits stated below .

What an accreditation certifies

MRC accredits one service's measurement process at a time. It is an industry-funded body whose members include advertisers, agencies, media sellers and measurement firms, and its audits are performed by CPA firms that MRC engages and the audited service pays for ; EY appears on the copy lists of its letters to HUMAN (formerly White Ops), Pixalate and Zefr . Each letter certifies that the service's reported statistics comply with the standards it names and with the procedures the service described. The audit tests conformance and reports no detection or error rate . Zefr's release of 9 March 2026 glosses its content-level accreditation as confirming "validity, reliability, and governance"; MRC's letter uses neither of the first two words .

MRC's response to Adalytics on 11 April 2025 requires impression-level GIVT filtration after serving, while pre-bid blocking is optional . Pre-bid audits test production and accuracy of supplied files and databases, while the platform decides whether to bid . HUMAN's accreditation covers pre-bid detection rather than prevention of serving, and MRC says it has no billing insight .

MRC's February 2024 policy makes the letter the authoritative scope reference . The digital list read on 10 November 2026 held 41 services without an update stamp . Superseded letters may disappear, and quarterly reports can omit continuations . Withdrawal records an exit from the audit process without establishing an accuracy failure .

Accreditation sits beside two neighbours that Chapter 1 defines: TAG certification, which covers an operating company rather than a product, and vendor self-attestation, such as a "Brand Safety" label from a vendor MRC does not audit .

The standards and their versions

The table lists issuer documents read on 9 and 10 November 2026 and the letters or seals that cite them . Guidance dates, accreditation dates and software releases describe different events . The July 2026 AI guidance was only scanned in the earlier pass, so this chapter draws no compliance conclusion from it .

IssuerDocumentVersion and dateWhat it coversNamed in the letters or seals ofEvidence
MRCViewable Ad Impression Measurement Guidelines (desktop)Version 2.0, 18 August 2015Viewable-impression thresholdsEvery viewability letter
MRCMobile Viewable Ad Impression Measurement GuidelinesFinal, 28 June 2016Viewability on mobile web and in-appEvery viewability letter
MRCInvalid Traffic Detection and Filtration Standards Addendum (IVT 2.0)Issued October 2015; updated June 2020Detection and filtration of GIVT and of sophisticated invalid traffic (SIVT)DoubleVerify, IAS, HUMAN, Pixalate, Protected by Mediaocean and Adloox; Zefr's YouTube viewability letter for the GIVT provisions only
MRC2024 IVT Interim UpdatesFinal, 24 April 2024Privacy; spoofed CTV domains and bundle IDs; property-level considerations for SIVT vendorsCited as interim guidance in most 2025 and 2026 letters
MRCStatement on IVT requirements and processes for pre-bid11 April 2025GIVT filtered after serving; pre-bid filtration not required; what pre-bid audits coverA statement, not a standard
IAB and MRCGuidelines for the Conduct of Ad Verification14 February 2012Property-level ad verificationDoubleVerify; Protected by Mediaocean's property-level letter; Zefr's and Google's content-level letters
MRC, with the American Association of Advertising Agencies (4A's), the Association of National Advertisers (ANA) and IABSupplement: Enhanced Content Level Context and Brand SafetyFinal, September 2018Content-level brand safety; reporting granular enough for buyers' own suitability settings; platform data accessProtected by Mediaocean's property-level letter; Zefr's and Google's content-level letters
MRCServer-Side Ad Insertion and OTT GuidanceFinal, August 2021Measurement of server-inserted ads and of video delivered to TVsProtected by Mediaocean's property-level letter
MRCPolicy for Accreditation Representations by MRC-Accredited Services15 February 2024How accredited services may describe their accreditationA policy, not a standard
MRCPolicy for Property-Level Ad Verification Representations18 October 2025, with a six-month grace period"Brand Safety" labels and disclosures for property-level servicesA policy, not a standard
MRCGuidance on Existing Standards for AI Use in Media Measurement8 July 2026Maps AI-related requirements in seven MRC documents; new AI standards in phases to early 2027Guidance
IAB Tech LabOpen Measurement SDK, JavaScript clients1.6.12, 2 November 2026Measurement interface for in-app, web video and CTV; device attestation since 1.6.0 of 29 October 2025DoubleVerify's letter (in-app viewability including OM SDK traffic); required by TAG's Certified Against Fraud 11.0 for verifiers under its device-attestation guidance
IAB Tech Labads.cert 2.0Page last updated 14 February 2024Authenticated Connections published; Authenticated Delivery and Authenticated Devices marked futureNone found
TAGBrand Safety Certified Guidelines5.0, July 2026Brand-safety certification of the operating companySeals held by DoubleVerify, IAS, Adloox, Channel Factory and illuma
TAGCertified Against Fraud Guidelines11.0, July 2026Anti-fraud certification of the operating companySeals held by DoubleVerify, IAS, HUMAN, Pixalate, Protected by Mediaocean and Adloox
TAGCertified Against Malvertising Guidelines7.0, July 2026, renamed from Certified Against MalwareAnti-malvertising certification of the operating companyNo seal found among the companies searched
TAGCertified for Transparency GuidelinesVersion 2, July 2026Transparency certification of the operating companySeals held by DoubleVerify, IAS and Protected by Mediaocean

MRC's standards page read on 10 November 2026 names no separate suitability guideline . The 2018 supplement requires discrete content reporting for buyers' risk settings, points to the 4 A's framework, and requires direct measurement or validated platform data for third-party content-level accreditation . It also requires disclosures of expected errors, sampling uncertainty and technical limits, while making false-positive likelihood optional . These disclosures do not supply a public product error rate . Chapter 5 covers data routes and Chapter 8 the frameworks.

Two MRC policies govern what accredited services may say. The February 2024 policy asks them to state what is and is not accredited, bars claims for anything not accredited, including services under review, and restricts "first" and "only" claims . The policy of 18 October 2025, addressed to the services MRC audits, bars accredited property-level services from positioning themselves or their metrics as "Brand Safety", or as content-level, without content-level capability in the environment, after a six-month grace period that by the paper's computation ran to about 18 April 2026; suitability language stays allowed with disclosures . It requires property-level vendors to disclose that images, video and audio are excluded, among other limits, and says MRC will not consider new property-level submissions without commitments to content-level capability . It does not bind vendors MRC does not audit, and MRC does not say whether it reaches an audited vendor's unaudited products, such as the "Brand Safety" products Zefr sells off YouTube, so that question is unresolved .

In the MRC records screened to 10 November 2026, no decision applying the October 2025 representation policy was found . DoubleVerify's letter of 13 April 2026 still uses Brand Safety metric names, while its one-sheet uses Brand Risk Floor Avoidance . Protected Media's pre-policy letter uses Brand Safety names with a property-level caveat . These discrepancies support no compliance finding .

The Tech Lab's ads.cert 2.0, a protocol for authenticating programmatic counterparties, has a reference repository whose last commit is dated 8 March 2023, and the pages read publish no adoption count for it or for OM SDK .

TAG certifies companies

TAG reissued all four of its certification guidelines in July 2026: Brand Safety Certified 5.0, whose change log runs back to version 1.0 of September 2020, when the Digital Trading Standards Group of the UK's former Joint Industry Committee for Web Standards (JICWEBS) merged into TAG's programme; Certified Against Fraud 11.0, which makes intermediaries use PaymentID and drops a billing-notification requirement for server-side ad insertion (SSAI) on CTV; Certified Against Malvertising 7.0, renamed from Certified Against Malware; and Certified for Transparency version 2 . TAG certifies the operating entity, not a product, and its brand-safety guidelines require a content verification service to disclose its capabilities and to review complaints of wrongful classification objectively .

TAG's registry read on 10 November 2026 lists DoubleVerify and IAS with global fraud and brand-safety seals confirmed through independent validation and TAG-validated transparency seals for North America and Europe . Both appear on TAG's Platinum page . Zefr and Escalated were enrolled without seals, while searches for Peer 39, Comscore and Fraudlogix returned no entry under the names tried . TAG seals do not enter H1a's MRC numerator.

MRC's decisions, 2023 to 2026

Between January 2023 and September 2026, MRC's quarterly reports record grants that widened viewability, invalid-traffic and platform reporting, and revocations of verification services or platforms' brand-safety controls in March 2023, November 2023, June 2024 and September 2025, each after a withdrawal from the process or a product's deprecation . The table lists the grants, revocations, denials and removals, and the continuations on which current scopes rest.

DecisionServiceActionScopeEvidence
January 2023GumGum (Verity API)Grant, continue, denyContent-level classification and URL reporting granted on CTV (English); desktop and mobile web continued; in-app denied
March 2023Facebook Instant Articles (Meta, first-party)RevokeContent-level brand-safety controls on Instant Articles, a deprecated product; in-stream video remained
May 2023IASGrantCTV video tracked ads and rendered and viewable impressions, certified traffic only
September 2023AdlooxGrantIn-app display and video viewability, exposure metrics and decision rate
November 2023Method MIRevokeRendered display ad impressions, after a hiatus, on withdrawal from the audit
March 2024DoubleVerifyGrant, continue, denyGranted: Serbo-Croatian (domain), Korean and Hindi (CTV), CTV video viewability including SIVT, pre-bid decisioning; denied: pre-bid Authentic Attention
March 2024IAS (YouTube through Google Ads Data Hub)GrantThird-party processing of YouTube video impressions and viewability on desktop, mobile web and in-app
April 2024IASGrantSIVT filtration and reporting on CTV
June 2024Oracle Moat Analytics, Moat's YouTube and Meta reporting, Oracle Contextual IntelligenceRevokeOn withdrawal from the process and planned deprecation
Letter of 9 October 2024IASContinue, in no quarterly reportProperty-level ad verification for display and video, English only, on desktop and mobile web, with the platform's other metrics
April 2025PixalateContinueAd counts, impressions and SSAI metrics with SIVT on desktop, mobile web, in-app and CTV; display viewability on the first three
May 2025GumGumContinueContent-level brand safety for text, image and video, English, on desktop, mobile web and CTV, open web
June 2025Meta (first-party)Grant, continueFacebook and Instagram Feed controls and reporting granted (English, Spanish); Facebook in-stream continued (37 languages); misinformation excluded
September 2025Protected MediaGrantProperty-level ad verification on desktop and mobile web, English, Spanish and Portuguese
September 2025GumGumRevokeContent-level brand safety and URL reporting on the open web, on withdrawal from the process
September 2025Meta (first-party)RevokeFeed and in-stream content-level controls and reporting, on withdrawal from the process
November 2025IAS reporting for Amazon's demand-side platform (DSP), server to serverGrantImpressions, viewability and tracked ads with SIVT on desktop, mobile web and in-app; Fire TV and Twitch excluded
November 2025Zefr (YouTube through Google Ads Data Hub)GrantVideo impressions and viewability on desktop, mobile web and in-app; skippable, non-skippable and bumper ads
December 2025TikTokRemoved from in-processLack of intent to engage in a third-party audit
January 2026IAS Attention Measurement; IAS In-Game MeasurementRemoved from in-processLack of intent to engage in a third-party audit
January 2026Zefr (YouTube content-level)GrantEnglish in-stream video and music video on desktop, mobile web, in-app and CTV, with exclusions
April 2026DoubleVerifyGrant, continueTikTok in-app video viewability with GIVT and SIVT; French on CTV and five domain-level languages; attention reporting (the attention paper's subject)
April 2026HUMANGrant, continueViewability on desktop, mobile web and in-app granted; pre-bid and post-bid GIVT and SIVT on four environments continued
April 2026Google (YouTube, first-party)Grant, continueShorts granted; in-stream continued
June 2026IASGrant, continuePre-bid viewability and IVT-avoidance segments on desktop, mobile web and in-app granted; platform continued "excluding Ad verification"
June 2026Peer39 Adloox (formerly Scope3 France Adloox)ContinueDisplay and video tracked ads, impressions and viewability on desktop, mobile web and in-app; SIVT on desktop and mobile web
September 2026Protected MediaContinueSame scope as April 2026

Two gaps shape the record. IAS's continuation of 9 October 2024, the last letter found granting its property-level ad verification, appears in no quarterly report and is read from an Internet Archive copy, because the live file returns HTTP 404 . Pixalate's scope still rests on its letter of 8 April 2025: none of the five reports from July 2025 to September 2026 records a Pixalate decision, and a missing entry is not evidence that no continuation occurred .

Accreditation by product

The scored products hold most of their MRC accreditation in viewability and invalid traffic, as the figure shows. Nine of the scored products in the figure have no MRC entry at all: mFilterIt among the suites; Fraudlogix and FouAnalytics among the invalid-traffic specialists; and Peer39's own product, Proximic by Comscore, illumaSentinel, ViewIQ, Pixability and Sightly's Brand Mentality among the suitability specialists . Escalated has nothing accredited, but MRC lists its audit as in process as of June 2026 .

MRC STATUS BY SCORED PRODUCT AND METRIC FAMILY, 10 NOVEMBER 2026DisplayviewabilityVideoviewabilityGIVTSIVTPre-bid IVTProperty-levelverificationContent-levelsafetyDoubleVerifyD MW App+1 platf.all 4+3 platf.all 4all 4+1 platf.all 4all 4IASD MW App+2 platf.all 4+3 platf.all 4all 4+1 platf.D MW Appno letterin auditAdloox verificationD MW AppD MW AppD MW AppD MWProtectedD MW AppD MW Appall 4all 4D MWmFilterItHUMAND MW AppD MW Appall 4all 4all 4PixalateD MW Appin auditall 4all 4in auditFraudlogixEscalatedin auditin auditin auditFouAnalyticsZefrYTYTYTPeer39ProximicillumaSentinelViewIQPixabilitySightlyAccredited (environments shown)In MRC audit, not accreditedListed without a current letterRevokedNo MRC entryNot sold in this familyD desktop, MW mobile web, App in-app, CTV connected TV; all 4 = all four; +n platf. = the number of large platformswhere a platform-specific service is also accredited (YouTube via Ads Data Hub, Meta, TikTok, Amazon server toserver). IAS’s property-level row is listed on MRC’s page without a current letter (the last letter granting it isdated 9 October 2024). Accreditation certifies a measurement process for the listed metric, environment and scope;its audits can include accuracy testing, but the letters publish no product error-rate estimate. Read on MRC’saccredited and under-review lists and letters, 10 November 2026; status-sensitive and rechecked at the freeze.
Nine of seventeen scored products have no MRC entry, and only Zefr holds content-level accreditation. MRC status for each scored product, by metric family and environment, from MRC’s accredited and under-review digital lists and the letters linked from them, read on 10 November 2026. A product with no MRC entry may still sell the family; the matrix scores accreditation against what each product sells (chapter 11). Content-level accreditation outside YouTube is held by no scored product.Units: status category · Scope: the 17 scored products of the product register, MRC digital accreditation, US · Sources Status-sensitive: MRC’s page carries no update date. "No MRC entry" is bounded to MRC’s digital accredited and under-review lists screened that day.Data: data/figure-data.json#accreditation

Content-level brand safety and suitability accreditation is held by two services, both for YouTube, with different placements . Google's is first-party validation of its own controls, not third-party verification: in-stream video and, since an April 2026 decision, Shorts, on desktop, mobile web and in-app according to the list . Zefr's is third-party labelling and reporting of English-language in-stream video and music video on desktop, mobile web, in-app and CTV, with Shorts excluded . On the list read on 10 November 2026, no open-web service holds content-level accreditation, and Zefr is the only scored product that does; both content-level letters cite the 2018 supplement . Zefr also sells content-level classification on Meta, TikTok, Snap and Reddit, none of it accredited . IAS's YouTube brand-safety audit is listed as in process as of April 2023, Meta's pipeline that supplies brand-suitability data to third parties is under review, and MRC's under-review list names no DoubleVerify service .

In May 2025 MRC continued GumGum's English-language content-level accreditation for open-web text, image and video on desktop, mobile web and CTV . GumGum's content-level accreditation was revoked in September 2025 after the contextual paper recorded it; this paper corrects that record rather than repeating it. The revocation followed GumGum's withdrawal from the accreditation process, and Meta's first-party content-level accreditation was revoked the same month on the same ground . A withdrawal is not a demonstrated accuracy failure .

DoubleVerify's and Protected by Mediaocean's accreditations in this area are property-level ad verification, which MRC's 2025 policy treats as classification of properties from text or keywords, with images, video and audio excluded . DoubleVerify holds it on desktop, mobile web, in-app and CTV under its letter of 13 April 2026, in about 176 languages for binary classification, 56 at domain level, 41 at page level and ten on CTV, and its own one-sheet states that this is not content-level brand safety . Protected by Mediaocean holds it on desktop and mobile web only, from an initial grant of 25 September 2025 covering display reporting and blocking and video reporting in English, Spanish and Portuguese .

IAS's property-level row has no letter on MRC's site. The list shows IAS with property-level ad verification on desktop and mobile web, but the last MRC letter whose body grants it is the continuation of 9 October 2024 (display and video, English only), which the list linked in every capture read from 28 April 2025 to 16 June 2026 . The letter of 9 July 2026 that the list now links omits ad verification, MRC's report for April to June 2026 says that continuation excluded it, and MRC's report for July to September 2026 records no IAS action . Whether the accreditation lapsed, continues on an unreported cycle or is shown in error is unresolved, and for H1a the row has no current letter basis .

Pre-bid detection of invalid traffic is accredited for three services: DoubleVerify and HUMAN on desktop, mobile web, in-app and CTV, and IAS on desktop, mobile web and in-app from an initial grant in June 2026, although MRC had said on 11 April 2025 that IAS was then neither audited nor accredited for it . Adloox, owned by Peer39 since June 2026, holds SIVT filtration on desktop and mobile web only . Peer39's only accredited service is that Adloox service, scored as its own row because evidence is not inherited between rows, and Comscore's own page claims MRC accreditation only for Comscore TV . HUMAN's June 2026 suite sells brand safety and suitability with no MRC accreditation or listed audit for it , and Pixalate's brand-safety ratings of apps and sites are property-level and neither accredited nor under review .

On CTV, SIVT filtration is accredited for eight services on MRC's list, five of them scored products, and video viewability for two, DoubleVerify and IAS, both limited to certified apps; Chapter 6 reads that record . On the nine large platforms, third-party measurement of a platform's own placements is accredited on four and third-party content-level classification on one, YouTube, through Zefr; Chapter 5 reads what those accreditations rest on .

Every accreditation above is narrowed by its letter. The table gives the placement, language, data and environment limits that H1a's units inherit.

ProductService and letterWhat the letter narrowsEvidence
DoubleVerifyQuality Analytics Platform, 13 April 2026On CTV, only certified apps count toward video viewable impressions and Fully On-Screen metrics, while other CTV viewability metrics may include tested, non-certified apps; no display viewability on CTV
DoubleVerifySame letterProperty-level languages: about 176 binary, 56 at domain level, 41 at page level, 10 on CTV
DoubleVerifySame letterPre-bid brand safety and suitability segments on desktop, mobile web and in-app only; CTV pre-bid limited to fraud and IVT, Fully On-Screen, app-store categories and site and app lists
DoubleVerifyMeta, YouTube and TikTok reportingMeta: Facebook and Instagram feed placements, net of GIVT; YouTube: in-stream and bumper through Ads Data Hub, net of GIVT, with Shorts, in-feed and masthead out of scope; TikTok: in-app video with GIVT and SIVT filtration
IASCampaign Monitor, Firewall and Viewability, 9 July 2026CTV video metrics and SIVT for certified applications only; metrics under the mobile rich-media ad interface (MRAID) out of scope
IASSame letterPre-bid viewability and IVT-avoidance segments (IP and user-agent segments, Fraud IP List) on desktop, mobile web and in-app; no CTV
IASProperty-level row on the listNo current letter; the last grant, of 9 October 2024, covered display and video, English only, on desktop and mobile web
IASAmazon DSP server-to-server reporting, 7 November 2025Amazon Retail, IMDb, Amazon Publisher Services and third-party exchange supply; Fire TV and Twitch excluded
IASMeta reporting, 9 July 2026Placements audited for Meta's own measurement, many excluded; net of GIVT; IAS cannot identify IVT inside Meta independently
AdlooxPeer39 Adloox, 29 June 2026SIVT filtration on desktop and mobile web only; in-app metrics net of GIVT; no pre-bid metric
Protected by MediaoceanProtected Media, 22 September 2026On CTV, video impressions only, by non-SDK measurement; no CTV viewability
Protected by MediaoceanProperty Level Ad Verification, 25 September 2025Desktop and mobile web; display reporting and blocking, video reporting only; English, Spanish and Portuguese; gambling and misinformation categories, IVT ad verification, custom bundle lists and rules, sentiment and non-matching language excluded
HUMANAd Measurement Platform, 14 April 2026Viewability on desktop, mobile web and in-app only; the list's SIVT row applies to backend detection and IVT predictions; pre-bid accreditation covers detection, not prevention
PixalateAnalytics platform, 8 April 2025Display viewability on desktop, mobile web and in-app only; video viewability and pre-bid SIVT in audit as of September 2025; no CTV viewability accredited or under review
ZefrYouTube content-level, 20 January 2026English-language in-stream video and music video; Shorts, YouTube TV, Google TV, Masthead, Google Video Partner videos and volumetric data excluded
ZefrYouTube through Ads Data Hub, 6 November 2025In-stream and bumper video on desktop, mobile web and in-app; GIVT filtration only; in-feed, masthead, Shorts and non-YouTube Demand Gen inventory excluded
Google (first-party, not scored)YouTube Brand Safety, 28 May 2026Google Ads and Display & Video 360 only; in-stream and Shorts; video discovery, masthead, YouTube Kids, non-YouTube inventory and live streams excluded; no CTV on the list

Whether vendors' claims stay within scope

H1b is contradicted on this paper's reading of what the pages imply. Its rule requires at least one current claim that states or implies an unlisted metric, environment, format or placement. The Peer 39 and Protected readings below meet that test only under the interpretations stated beside them .

Peer 39 places accredited viewability measurement inside its description of the pre-bid signal . The paper reads this as implying pre-bid accreditation, which MRC's separate post-serve Adloox letter does not grant . An alternative reading is that the signal incorporates accredited post-serve data. The author retains the broader-scope interpretation for H1b; it is an author judgement, not an MRC finding . Peer 39's acquisition post lists the accredited metrics more narrowly .

Protected by Mediaocean's home page says it is MRC accredited for rendered display and video ad impressions on desktop, mobile web, in-app and CTV. MRC's letter of 22 September 2026 accredits display impressions in the first three environments and, on CTV, video impressions only, by non-SDK measurement; read literally, the claim extends display impressions to CTV . The page's SIVT sentence, which names all platforms including CTV, is wider on CTV by the same reading, while the company page's narrower claim of viewability and SIVT detection stays within scope .

The Zefr profile found a third claim of the same kind. Zefr's company page lists, as a 2025 milestone, MRC accreditation for viewability and IVT, while MRC accredits Zefr's YouTube video impressions and viewability with GIVT filtration only and lists Zefr under no invalid-traffic metric; the profile codes the claim broader by implication .

The table separates ambiguous claims, unreadable scope and priority or uniqueness claims outside H1b's metric-scope test . The verifier pass moved IAS's Quality Impression and HUMAN's MRC brief to ambiguous . Whether a dated release still online counts as current remains an author decision.

ProductPage, and dateWhat it saysMRC's listed scopeCoding
Peer39 brand safety and suitabilityVerification page, undated, read 10 November 2026Viewability measurement described as MRC-accredited and built into the pre-bid signalNo entry for Peer39's product; the Peer39 Adloox service is accredited post-serve only, on desktop, mobile web and in-app (29 June 2026)Broader on this paper’s reading of the pre-bid wording, with a post-serve-input alternative
Protected by MediaoceanHome page, 2026 footer, read 10 November 2026Accredited for rendered display and video impressions on desktop, mobile web, in-app and CTV, and for SIVT on all platforms including CTVDisplay impressions on desktop, mobile web and in-app; on CTV, video impressions only, by non-SDK measurement (22 September 2026)Broader on a literal reading
Zefr brand safety and suitabilityCompany page, read 10 November 2026A 2025 milestone of MRC accreditation for viewability and IVT; a 2019 milestone of industry accreditations for brand safety and suitability that names no bodyYouTube video impressions and viewability with GIVT filtration only; no IVT metric listed; first Zefr decision found November 2025Broader by implication in the Zefr profile's coding; the 2019 milestone not checkable
Pixalate ad fraud protection and analyticsBlog post of 27 January 2022, live on 10 November 2026Pre-bid blocking described as MRC-accreditedNo accredited Pixalate pre-bid service; pre-bid SIVT in audit as of September 2025Broader if a live dated post counts as current (the author's call)
DoubleVerify advertiser verificationMRC one-sheet, PDF created 8 April 2026, read 10 November 2026Pre-bid property-level brand suitability avoidance and Authentic Brand Suitability segments listed against desktop, mobile web, in-app and CTVPre-bid brand safety and suitability segments on desktop, mobile web and in-app; on CTV, fraud and IVT, Fully On-Screen, app-store categories and site and app lists (13 April 2026)Ambiguous: within scope if the one-sheet's footnote 2, which lists those CTV components, governs the rows
DoubleVerify advertiser verificationFraud, brand-suitability and viewability pages, undated, read 10 November 2026DV Authentic Ad called an MRC-accredited metric, defined to include the intended geographyAuthentic Ads and Authentic Rate listed without a geography note; DoubleVerify's one-sheet says they are not accredited when campaign settings include geographic targetingAmbiguous: within MRC's listed scope, broader than DoubleVerify's own disclosed basis
DoubleVerify advertiser verificationRelease of 24 April 2024, live on 10 November 2026Accreditation extended to CTV pre-bid property-level brand suitability, contextual and Fully On-screen segmentsBrand safety, suitability and contextual pre-bid segments on desktop, mobile web and in-app; app-store categories and site and app lists accredited on CTV (13 April 2026)Ambiguous: a dated release, which may describe the app-level CTV segments
IAS advertiser verificationQuality Impressions and ad-fraud pages, read 10 November 2026Quality Impression called MRC-accredited, defined to need a suitable environment and the correct geographyQuality Ads on desktop, mobile web and in-app; the July 2026 letter excludes ad verification and lists no brand-safety or geography metric, which only the November 2025 Amazon DSP letter listsAmbiguous
HUMAN Ad Integrity SuiteMRC solution brief, PDF created 16 August 2023, still served on 10 November 2026Calls its pre-bid accreditation one for "mitigation" of SIVT and offers end-to-end SIVT protectionPre-bid IVT detection; MRC says detection, not prevention of ad serving (11 April 2025)Ambiguous
Pixalate ad fraud protection and analyticsMedia Ratings Terminal FAQ, undated, read 10 November 2026App and domain IVT rates from a global data pool described as MRC-accredited IVT measurementThe Analytics platform's reported statistics (8 April 2025)Ambiguous: whether ratings derived from accredited detection are the accredited service is MRC's judgment
Zefr brand safety and suitabilityBrand-safety solution page, read 10 November 2026MRC logo beside content-level brand safety and suitability and viewability, on a page presenting YouTube, Meta, TikTok, Snap and RedditYouTube onlyAmbiguous: unqualified rather than expressly broader
Adloox verificationReleases of 25 January 2023, 23 September 2024 (dateline June 2024) and 24 October 2024The company described as MRC-accredited in announcements of Equativ pre-bid measurement, Display & Video 360 segments and Meta brand safety and suitability measurement, with no disclaimerNone of those services is accreditedAmbiguous: the descriptor attaches to the company, and the releases are dated
Adloox verificationCompany page, read 10 November 2026Accredited since 2014 for comprehensive ad measurement and traffic validation, with no footnote or link to specificsImpressions and viewability on desktop, mobile web and in-app; SIVT on desktop and mobile webNot checkable: no metric or environment stated
IAS advertiser verificationAd-fraud page, read 10 November 2026IAS the first verification provider accredited for SIVT on CTVMRC's list captures of 24 January 2023 and 23 February 2024 show CTV SIVT for DoubleVerify, HUMAN, Oracle Moat, Pixalate and Protected Media before IAS's April 2024 grantPriority claim, outside the scope test; contradicted by MRC's record
Pixalate ad fraud protection and analyticsAnalytics page, read 10 November 2026Its 12 SSAI metrics called unique to PixalateIAS is also accredited for SSAI video metrics; MRC does not generally approve "only" claimsUniqueness claim, outside the scope test

Other screened claims omit limits without demonstrably extending scope, including certified-app restrictions and Zefr's English-only in-stream restriction . Channel Factory, illuma, Fraudlogix and Escalated made no MRC claim in that screen, nor did HUMAN's June 2026 suite release .

The earlier screen could read at least one page or release for each product it covered . It did not establish the accreditation-claim position of mFilterIt, FouAnalytics, Pixability and Sightly. Unscreened pages cannot be coded unreadable, so the register-wide unresolved gate has not been evaluated completely. Retaining H1b's contradicted verdict requires accepting the stated scope interpretations and a counterexample-first reading of the rule.

How H1a counts units

H1a predicted that at least three quarters of the units the scored products sell hold current MRC accreditation. A unit is a product, a metric family and an environment. The six families are display viewability, video viewability, GIVT, SIVT, property-level ad verification, and content-level brand safety or suitability; the environments are desktop web, mobile web, in-app, CTV, audio and digital out-of-home, plus each large platform on which a metric is sold. Attention metrics belong to the attention paper and are not units here. What a product sells comes from its own documentation, recorded in the product register; what is accredited comes from MRC's accredited list, under-review list and letters at the freeze, read for product, metric, environment, format, placement, language and standard. A letter's placement and language exclusions narrow what an accredited unit covers: Zefr's English-only in-stream grant does not reach the Shorts, Google Video Partner inventory or other languages Zefr sells on YouTube .

Under-review units count as unaccredited, and IAS's property-level row has no current letter basis . H1a reports pooled, cohort, environment and content-level shares with both counts. Its registered support rule requires at least 75% overall and no cohort below 50%. It is contradicted below either threshold and unresolved if sold units cannot be established for more than a quarter of scored products.

The grid matters as much as the list. In a preliminary count during the research, whether Zefr's YouTube viewability reporting counted as one unit or three, with or without GIVT units, moved the suitability cohort's share across the 50% floor . One grid has to apply to every product, and the units come from the scored register.

The merged documented grid records 108 accredited units out of 298 sold (36.2%): cohort A 80/179 (44.7%), cohort B 25/82 (30.5%) and cohort C 3/37 (8.1%). Content-level brand safety or suitability accounts for 1 accredited unit out of 61 sold units in that family; 1 further content-level unit is under review. Across all families, 12 units are under review (1 in A, 11 in B and none in C), counted as unaccredited. H1a is contradicted: the pooled share is below the 75% rule and every cohort is below its 50% floor. These are documented product×metric-family×environment units, not shares of impressions or spend. illumaSentinel has an unknown sold-unit denominator and remains outside this documented tally; unknown is not zero, and those omissions do not meet the registered more-than-a-quarter unresolved gate. Status rests on the November records, pending the 11 December 2026 freeze recheck . One counting ruling matters here: MRC's accredited list still shows IAS's property-level verification on desktop and mobile web, while the July 2026 letter linked from it omits them. The grid follows the list, as it does for every product, and counts both units as accredited; excluding them gives 106 of 298 (35.6%). Neither reading changes IAS's accreditation score or the H1a verdict .

MRC-ACCREDITED SHARE OF THE UNITS EACH COHORT SELLS0%10%20%30%40%A · Suites44.7% 80 of 179 units, 1 in auditB · IVT specialists30.5% 25 of 82 units, 11 in auditC · Suitability specialists8.1% 3 of 37 unitsAll scored products36.2% 108 of 298 units, 12 in auditA unit is a product x metric family x environment that the product documents as sold, narrowed by the letters’ placement and languageexclusions. Units in audit count as not accredited.
Under half of what the scored products sell is MRC-accredited. Accredited units divided by sold units, for each cohort and pooled, from the cell-verified scoring of each product (hypothesis H1a). A sold unit is counted only where the product’s own documents or a verified record show it is sold.Units: percent of sold units · Scope: the scored products at 10 November 2026 · Sources The denominator depends on what each vendor documents as sold; an undocumented product line is not counted.Data: data/figure-data.json#h1a

Chapter 3

What the independent record shows

In the sources screened to 10 November 2026, no independent test reports an error rate for a scored product on a labelled set with a published denominator, supporting H3 . The record measures prevalence, vendor disagreement and failure cases . Adalytics' candidate miss rates concern IAS's unscored publisher tool and give no session counts . Neither those cases nor vendors' replies supply the test H3 requires .

In short

The ANA's 2023 study measured IVT at 0.78% of impressions and MFA at 21%, using vendors' and a data supplier's classifications . DoubleVerify and IAS announced their MFA products after that study's window . Adalytics' ten reports naming a verifier are observational and publish no product error-rate denominator . Smith et al. found vendor disagreement on news articles, with labels partly constructed by the authors and no ground truth . The candidate bot miss rates concern IAS Publisher Optimization, which this paper does not score, and give no session counts . H3 is supported on this dated, bounded screen . Vendors' protected-versus-unprotected comparisons use their own classifiers and publish no arm sizes . Industry ad-fraud totals are sponsored projections or models with incompatible definitions .

What the ANA's 2023 study measured

The ANA's study measured what the participants' vendors and one data supplier found in 21 advertisers' buys; it tested no product. The Programmatic Media Supply Chain Transparency Study analysed log-level data from 21 ANA members, of 67 that wanted to take part, covering $123 million of open-web programmatic spend and 35.5 billion impressions from September 2022 to January 2023 . Three demand-side platforms (DSPs), six supply-side platforms (SSPs) and three verification companies, DoubleVerify, IAS and Oracle Moat, supplied data and had no role in the analysis (p. 17); the panel selected itself . The Complete Report (December 2023) is cited by printed page from a public copy hosted by Adslot . The logs were reconciled by TAG TrustNet, a joint initiative of Fiducia and TAG, the Trustworthy Accountability Group, which certifies verification vendors .

The participating vendors' classifications were matched to DSP logs . IVT was 0.78% of impressions and 0.69% of spend, median measurability was 86%, and 85% of measurable impressions were viewable . These are measured prevalence, and the ANA's attribution of low IVT to protection is its interpretation .

Domains matched to DeepSee.io's 4,500-site MFA list took 21% of impressions and 15% of spend in the subset analysed . The ANA reported good viewability and low IVT and described the sites as usually brand safe . Yet brand safety was outside the study, and the Q3 2025 benchmark cites inconsistent vendor methods as the reason .

The participating DoubleVerify and IAS services announced MFA products after the study window, in September and October 2023 respectively . DoubleVerify's earlier on-request handling remains undated, and no participant contract establishing it in force was found in the screened sources . Adloox, which supplied no study data, did advertise an MFA-labelled pre-bid segment by December 2022, but equivalence to the September 2023 joint definition is undocumented . The ANA advised testing the later offerings against what they promised .

The benchmark that followed

The ANA's quarterly Programmatic Transparency Benchmark repeats the design at a larger scale and keeps its limit: it reports prevalence as its suppliers measure it and tests none of them . IVT, measurability and viewability come from log data matched between DSPs and the advertisers' own verification providers, MFA data from DeepSee.io and, from the third quarter of 2025, brand-risk data from Mobian, a contextual vendor; the panel changes every quarter .

ReleasePeriodMarketers (participating / active)ImpressionsSpendIVTMFAEvidence
2023 study (Complete Report)Sep 2022 to Jan 20232135.5 billion$123 million0.78% of impressions, 0.69% of spend21% of impressions, 15% of spend (subset)
2024 Benchmark, Findings UpdateApr to Sep 202437 / 2138.5 billion$235 millionnot in the passages readmedian 1.1% (10% in the 2023 study)
Q3 2025Jul to Sep 202539 / 2117.5 billion$142 million0.4% of spend (waterfall)0.5% of spend (waterfall; 0.4% in the summary)
Q4 2025Oct to Dec 202554 / 3516.7 billion$142 million0.4% of spend0.4% of spend
Q1 2026Jan to Mar 202686 / 6620.9 billion$160 million0.2% of spend1.1% of spend
Q2 2026 (release read in Fiducia's repost; the report is behind a form)Apr to Jun 2026not readnot readnot readspend lost "held at 0.1%"not read

The benchmark's changing panel and supplier classifications cannot separate cleaner supply from improved detection . It warns that waterfall values and medians differ, and the Q2 2026 release's description of 0.1% IVT as unchanged conflicts with Q1's printed 0.2% . Mobian's separate claim that legacy systems misclassified roughly a third of professional news pages gives no labels or denominator . A July 2026 ANA, TAG and Fiducia analysis found AI-generated pages passing the screened controls but named no verifier, leaving H3 unchanged .

Adalytics' reports, case by case

Adalytics' reports document failure modes: cases, without denominators, of ads carrying a verifier's code that ran where buyers would not want them or reached non-human traffic. From June 2023 to March 2025 it published ten reports that name a verification vendor, all observational, and the three that bear most on DoubleVerify, IAS and HUMAN make no claim about magnitude or relative abundance . Adalytics sells analytics to advertisers and declares that no publisher or ad-tech vendor funds its research .

Report (subject)Dated, and on what basisVerifiers named, and their roleMethod and samplePeriod observedEstimandEvidence
Google TrueView and Google Video PartnersPublic by 27 June 2023 (Google's reply that day; first archived 28 June)IAS and DoubleVerify, said to rely on YouTube's data feeds; MRCCase report; no sample statedNot statedFailure-mode cases
MRC's answers on TrueViewJuly 2023 (sitemap 14 July; first archived 23 July)MRC; Oracle Moat, DoubleVerify, IASRelays MRC's statement that no third party was audited or accredited for Google brand safety2023Record of accreditation scope
YouTube advertisers and children's data17 August 2023 (sitemap and first archive)IAS and DoubleVerify in YouTube's account of beta tests, IAS also as a recipient of one landing page's data; Channel Factory once, as a vendor Apple appeared to useCase report; non-exhaustive list of brandsJune and July 2023Cases, not about verifiers' classifications
Search ads on objectionable pagesNovember 2023 (first archived 28 November; sitemap 6 December)HUMAN and DoubleVerify as advertisers whose search ads appeared on such pagesCase report2023Cases, not about verifiers
MFA sites in January 2024March 2024 (sitemap 11 March; first archived 18 March)IAS, DoubleVerify, Oracle Moat, Pixalate, HUMAN, Fou Analytics: tags in ads on MFA sites22 sites judged MFA; panel, crawlers and archives; brands counted, no impression denominatorJanuary 2024Failure-mode cases
Forbes www3 subdomainUndated; public by 4 April 2024 (first archived that day); Forbes made contact on 20 March, the page-URL mis-declaration was no longer observable from 22 March and the subdomain was offline from 2 AprilIAS, DoubleVerify, Oracle Moat, HUMAN tags in ads; their publisher code absentOne publisher; mis-declared bid requests to four SSPsTo 22 March 2024One failure-mode case
User IDs in ad auctions10 May 2024 (sitemap and first archive)HUMAN and Oracle Moat as an SSP's declared partners; MRCOne case2024Case, not about verifiers' classifications
"AI" brand safety technology7 August 2024 (first archive; same-day replies)IAS publisher key-values; DoubleVerify and IAS tags; MRCConvenience sampleExplicit-page examples undated; dated examples April 2023 to July 2024Failure-mode cases, both directions
Image host imgbb.com and ibb.co7 February 2025 (first archived that day); the senators' letters cite "Jan. 2025", its data cutoffDoubleVerify and IAS tags on explicit adult pages; MRCThousands of URLScan.io recordings and its own sessions; advertisers' domain-level account relayed2017 to January 2025Failure-mode cases and a second-hand domain-level account
Pre-bid bot filtration28 March 2025 (sitemap; dated replies)DoubleVerify, IAS, HUMAN, MRCThree crawl datasets: over a petabyte, over two million websites, over seven years; no counts published2017 to March 2025Failure-mode cases; one candidate rate (IAS flag)

The explicit-pages report of 7 August 2024 began when a multinational brand, told that its campaigns were "100% brand safe", asked for a review; the phrase is this report's . It showed IAS's publisher-side endpoint returning "veryLow" or "low" on all seven brand-safety keys, as IAS's own guide defines them, for explicit pages, mostly Fandom wikis, and ads carrying DoubleVerify or IAS code there; the same endpoint labelled mainstream news "medium" or "high" . The sample was one of convenience, and the report disclaims any claim about a classifier's sensitivity, specificity or accuracy . Its key-values come from a publisher product, not the IAS advertiser verification this paper scores .

The image-host report of 7 February 2025 works at the level of a domain. It reviewed thousands of URLScan.io recordings of imgbb.com and ibb.co from 2017 to 2024 and relayed several advertisers' statements that their vendors had marked 100% of measured impressions there safe or suitable, a second-hand, domain-level account rather than a measured rate . The ads it showed, some with DoubleVerify or IAS tags, sat beside adult explicit pages; its one archived instance involving putative child sexual abuse material (CSAM) was reported to the authorities and not shown, and advertisers could not count ads beside CSAM for lack of page-level reporting . A domain that hosts illegal images is not the same as an ad beside one.

The pre-bid bot report of 28 March 2025 shows ads served to bots in three crawl datasets, URLScan.io, HTTP Archive and one it does not name . Its statement that advertisers were billed rests on non-public log files and invoices from its work for Fortune 500 advertisers, and it disclaims any finding of fault, magnitude or prevalence . Whether URLScan's crawler declares itself is disputed: DoubleVerify and MRC say its user agent does not, while HTTP Archive is on the Spiders and Bots list of the Interactive Advertising Bureau (IAB) . Senator Warner's letters, dated 24 March 2025 and released on 28 March, asked the Federal Trade Commission (FTC) and the Justice Department to investigate DoubleVerify, IAS and HUMAN, and footnote 6 of the FTC letter describes the failures as general invalid traffic (GIVT), not advertising fraud under industry definitions; the letters record questions, not findings . The report's HUMAN cases concern platforms that, MRC says, make their own bid decisions on HUMAN's signals, and no HUMAN response was found by 10 November 2026 . DoubleVerify sued Adalytics over this report on 12 May 2025, and the pleading-stage ruling of 27 April 2026 took DoubleVerify's allegations as true, made no finding on the report's accuracy and left the case to continue (Chapter 9) .

One figure is framed as a detection error rate for a named product. Treating its crawls as "100% confirmed bot traffic", the report says the "fr" flag of IAS's Publisher Optimization marked the declared HTTP Archive bot valid 16% of the time in 2022 to 2024 crawls (17% in its summary) and URLScan.io's bot 77% of the time in 2019 to 2024 crawls of over 1,200 websites, without publishing the number of sessions behind either share . IAS's own implementation guide of 11 July 2024 defines the flag as "Whether the ad represents invalid traffic" . MRC said on 11 April 2025 that IAS's publisher tools were neither audited nor accredited, and its list read on 10 November 2026 still shows none, though it now lists IAS's pre-bid IVT detection for desktop, mobile web and in-app, accredited from June 2026 .

What the vendors and MRC answered

Every answer corrects something, and none publishes the labelled data that would settle the case it answers. H3 judges a response on its merits where it corrects scope, data, decoding, staleness, the denominator or the billing stage, without requiring a new test of equal size .

DateRespondentReport answeredWhat it correctsData publishedEvidence
27 June 2023, updated 13 JulyGoogleGoogle Video PartnersMethod; cites DoubleVerify's and IAS's 93% viewability on Google Video Partners as contradictionNo counts
12 July 2023IASGoogle Video PartnersData: 3.4% of its YouTube and partner impressions, 92.74% viewable, 0.40% IVT (January to early July 2023)Aggregate shares
7 August 2024DoubleVerifyExplicit pages; also the MFA and Forbes reportsScope (all tags shown to it were publisher services), settings (exception lists, monitoring); says all content shared was classified accurately, virtually all the MFA sites were already classified as MFA and the Forbes inventory as low-tier MFANo classifications or counts
7 August 2024IASExplicit pagesScope ("almost all" tags from publisher solutions), settings (monitoring, not blocking); no timeframe or sample in the report; silent on the endpoint's low-risk valuesNo
10 January 2025DoubleVerifyBot report, in draftBilling stage: under the industry standard, GIVT that reaches an ad is removed from reporting and not chargedNo
7 February 2025DoubleVerifyImage hostDenominator: 0.000047% of the media transactions it measured in 30 days; the archived copy adds that 83% served to the home page, a figure the live page dropped on 15 October 2025; classification only above an ad-traffic thresholdOne share
7 February 2025IASImage hostDenominator: 88,000 of nearly 1.9 billion webpages in 60 days; domains made ineligible for monetizationPage counts
13 February 2025DoubleVerifyImage host (remedial)New default-on "Highly Illicit: Do Not Monetize" category; about 100 of over 300 domains in public data from the National Center for Missing & Exploited Children (NCMEC) already coveredDomain counts
14 February 2025IASImage host (remedial)Over 350 NCMEC-notified sites already covered; a category for unmoderated user-generated content sold since 2023Site and campaign counts
21 March 2025IASBot report, before publicationMethod: publishers can target away from impressions labelled IVTNo
28 March 2025DoubleVerifyBot reportEvery example shared before publication correctly identified; billing stageNo
28 March and 3 April 2025IASBot reportScope (publisher tag, not advertiser tag), stage (no post-bid blocking by design), decoding (back-end adjudication); HTTP Archive flagged as GIVT, URLScan.io impressions as GIVT, sophisticated invalid traffic (SIVT) or reduced-value inventoryNo counts
7 April 2025DoubleVerifyBot reportScope: of 115 examples, 62 had no DoubleVerify measurement, 10 fell outside retention, all 43 matched were invalidated post-bid; labelling (URLScan undeclared); decoding (the "DV source code" is DSP code); its own test figuresExample counts; no per-impression data or method
11 April 2025MRCBot reportScope (IAS publisher tools and pre-bid IVT, DoubleVerify publisher tools and Scibids then unaudited); standard (post-serve GIVT filtration required, pre-bid not); labelling (URLScan.io's user agent undeclared); no insight into billingDid not test each observation
None found by 10 November 2026HUMANBot reportNo public response in the searches and sitemaps readn/a

The August 2024 answers correct scope and settings, but not the low-risk values IAS's publisher endpoint returned for explicit pages, which are classifications whatever a buyer's settings . In February 2025 both vendors supplied the denominators the report lacked, and DoubleVerify disclosed a coverage limit: a page is classified only once it reaches a certain level of ad traffic . Its statement as archived on 7 February 2025 adds, in its summary, "with over 80% served to the site's home page", and puts that share at 83% in the body; the live page, revised on 15 October 2025, carries neither . Neither vendor met the advertisers' account of 100% safe or suitable with data .

On the bot report, DoubleVerify's breakdown of the 115 examples is a scope correction H3 counts, since more than half carried no DoubleVerify measurement; its own detection figures come from the classifier being defended, without method or sample . IAS corrects scope, stage and decoding by its own assertion, and MRC supports it on scope and stage, but IAS classified the URLScan.io impressions as GIVT, SIVT or reduced-value inventory, a category it keeps apart from IVT, so its answer does not say every one was classified as invalid . MRC, which had not tested each observation, wrote: "It's very likely a material amount of SIVT goes undetected" .

Smith et al. 2026: agreement, not accuracy

Smith, Grossman, Torres-Agüero, Sen, Borcea and Chen, five authors at the New Jersey Institute of Technology and one at DeepSee.io, measured how often three providers agree about the same news article, not whether any is right . The paper was posted to arXiv on 3 January 2026 and published on 25 May 2026 in the proceedings of the International AAAI Conference on Web and Social Media, with funding from two National Science Foundation grants, the Martin Tuchman '62 Chair Endowment and the Leir Foundation . DeepSee.io also supplied the ANA study's MFA list, and this paper records the two as one interest .

The authors read the brand-safety key-values that publishers' own Google Ad Manager integrations exposed, client-side, for DoubleVerify, IAS and Oracle (Moat) on news articles collected in September 2024 . These are publisher-side labels, not scores obtained from the providers or the ratings advertisers buy, and the sample covers only pages whose publishers collect them . Of 5,241 URLs attempted, 4,352 returned tag data and 3,219 carried any classification; 1,416 of the 3,219 (43.99%) were unsafe for at least one provider under the authors' own rules . Those rules are partly the authors' construction: IAS gives no overall rating, so an article counted as unsafe for IAS when any flag was present or any category sat above "veryLow", and DoubleVerify's label is read from its "Any moderate content" code . A union across three providers is not a ground-truth unsafe rate. The data date from the month Oracle retired Moat, on 30 September 2024, so the Oracle comparisons are historical .

Providers comparedArticlesDisagreementKrippendorff's alphaEvidence
IAS and DoubleVerify94, all on forbes.com (89 also rated by Oracle)37.23%0.19
Oracle and IAS329, rated by just these two29.79%0.42
Oracle and DoubleVerify249, rated by just these two14.06%0.53
All three89, all on forbes.com41.57%0.35

The overlaps are small, and the IAS and DoubleVerify pair rests on 94 articles from one publisher . Of the 31 articles Oracle or IAS flagged for hate speech, the two disagreed on 27 . The authors found misclassifications by every provider by hand, without a rate, say no ground-truth dataset of brand-safety labels exists, and call for a public, human-labelled benchmark . The study measures consistency between vendors, not accuracy.

Other academic work and platform reports

Few other studies touch a verification product. Adalytics' analysis of December 2020, inside H3's window from 2019, found Comscore and Oracle Moat agreeing on safe or unsafe for 58.7% of 25,730 archived wsj.com articles; a companion post was edited after IAS said the data it described were not advertiser keyword blocklists . Searches of arXiv on 10 November 2026 found no other study testing a named verification product's outputs and none reporting a verification product's miss or false-positive rate on a labelled set . The labelled designs found test other products, such as FP-Inconsistent's evasion rates against the bot detectors DataDome and BotD, while Vekaria and colleagues document inventory pooling that evades brand-safety protections . Google Scholar, SSRN, the ACM Digital Library and IEEE Xplore were not queried . Platform transparency reports, YouTube's Violative View Rate and Meta's prevalence metric, estimate violating content from labelled samples, but they are first-party measures of each platform's own policy enforcement, not error rates of a third-party product .

Coding the record by estimand

Coded by the estimand each design supports, the record is mostly failure cases. Workstream B coded 15 rows: 7 failure-mode cases, 2 studies of prevalence in a sample, 2 of agreement between vendors, 2 candidate error rates and 2 platforms' first-party prevalence metrics . The bot report fills two rows, its served-ad cases as failure modes and its IAS shares as a candidate rate; the other candidate, FP-Inconsistent, tests bot detectors that are not verification products . The verifier pass added the July 2026 AI-slop analysis to prevalence and Smith et al.'s hand-found misclassifications to the failure cases .

INDEPENDENT AND PLATFORM STUDIES BY WHAT THEY CAN ESTIMATE0246Failure-mode cases7Prevalence in a sample2Agreement between vendors2Candidate error rate2First-party prevalence (platforms)2Of the two candidate error rates, one concerns IAS’s publisher-side invalid-traffic flag on declared bots in one crawl dataset (2025) and theother a non-verification fingerprinting tool. No labelled test of a scored product’s sold category was found.
Most of the independent record shows failure modes, not error rates. The 15 independent and platform studies coded in workstream B by the estimand each design supports: failure-mode cases without a denominator, prevalence in a sample, agreement between vendors, a candidate error rate with a labelled set and a denominator, and platforms’ own prevalence metrics.Units: count of studies · Scope: studies concerning verification products, 2019 to 10 November 2026 · Sources A count of designs, not of verdicts; one study can support only one estimand here.Data: data/figure-data.json#estimands

H3's registered support rule requires no independent error-rate study of a scored sold category on a labelled set with a denominator . The screen through 10 November 2026 meets it because IAS Publisher Optimization is outside the scored register . Adding that publisher tool before its session counts are public would invoke the unresolved branch for an incompletely documented candidate study . The screen must be repeated at the freeze.

An independent miss-rate, false-positive or precision test of a scored sold category, with labels and a denominator, would contradict H3 . The IAS candidate would matter if its session counts were published and the publisher tool entered the register . Smith et al. call for a public labelled benchmark, which offers another route .

Vendor and commissioned evidence, labelled as such

Vendor and commissioned numbers measure prevalence, compare protected with unprotected campaigns, including in connected TV (CTV), or model savings. In the publications screened, by or with DoubleVerify, IAS, HUMAN, Pixalate, Zefr, Peer39 and Channel Factory, plus studies by TAG, Juniper Research, CHEQ, the ANA with White Ops and Forrester, none reports a miss rate, false-positive rate or precision for a sold category on a labelled holdout with a denominator; Adloox, Protected by Mediaocean (formerly Protected Media), Fraudlogix, Escalated, TrafficGuard, Pixability and Apostra were not screened for such claims in that workstream, and illuma and Proximic in one query only .

Vendor or body (sponsor)FigureBase and periodDesignEvidence
DoubleVerify, release of 7 May 2026CTV fraud under 1% protected, nearly 9% unprotected"Billions of impressions" and controlled tests; no period, counts or assignmentProtected against unprotected, measured by DoubleVerify
IAS, 20th Media Quality ReportFraud 10.9% non-optimized, 0.7% optimizedSecond half of 2024; non-optimized campaigns 4.0% of those monitored (2024 in the text, the second half on the chart)Self-selected arms, measured by IAS
IAS, 21st Media Quality ReportCTV IVT 9.1% non-optimized, 0.1% optimizedSecond quarter of 2026; arm sizes not publishedSame
IAS, 21st Media Quality ReportIVT 1.1%2025; open-web desktop and mobile web; IAS's intent-based definitionPrevalence
Pixalate, releases of 29 April and 12 August 2026IVT 20% web, 39% mobile app, 25% CTV (first quarter); 20%, 41% and 26% (second quarter)Over 82 billion and over 80 billion impressions, predominantly open auction; Pixalate labels its first-quarter figures opinionsPrevalence
TAG, conducted by The 614 GroupIVT under 1% in TAG Certified ChannelsJanuary to June 2024; 745 billion certified-channel impressions from agency holding companies, measured by vendors TAG describes as MRC-accredited; USPrevalence after filtering

DoubleVerify and IAS publish the same design: fraud in protected campaigns set against unprotected ones, measured by the classifier that did the protecting, with no campaign or impression counts per arm . In IAS's comparison buyers chose their arm, and only 4.0% of the campaigns it monitored ran unprotected . A low protected rate shows that the controls and the classifier agree; it cannot show what both missed . The prevalence figures move with base and definition, so global rates from 0.1% to 41% sit in the table without contradicting one another; none is an error rate, and they cannot be pooled . IAS's brand-safety fail rates appear only as multiples of baselines it does not publish .

Vendors also state accuracy without a test. IAS says it is "42% more accurate than the next best provider at classifying online content", and its FY2024 10-K claims "up to 74% more accurate" brand-suitability reporting on social platforms, neither with method, sample, period or denominator . HUMAN says its brand-safety engine is benchmarked against over 400,000 expert-labelled pages with "high accuracy", and gives no figure . DoubleVerify's FY2025 10-K concedes that partners' data have caused errors in its technology, without a rate .

Zefr's August 2025 study tests six general-purpose multimodal models on 1,500 videos, with the best F1 score at 0.91 against reviewers' labels . It does not test Zefr's sold classifier . The Oracle-commissioned Forrester work combines two economic-impact studies for since-retired products . Zefr's Forrester survey and Channel Factory's MAGNA study measure opinions and adjacency effects .

Scale claims use different units and describe populations, not error-rate denominators: DoubleVerify files about 9.5 trillion Media Transactions Measured for 2025, its billing unit, IAS cites over 300 billion "digital interactions" a day, and Pixalate over 82 billion impressions in the first quarter of 2026 .

Industry "ad fraud" totals are sponsored estimates

The industry-wide "ad fraud" totals in circulation are projections or models, each produced with or distributed by a fraud-protection vendor, or by bodies scoring their own programmes; none is a measured market total .

EstimateValueYearBase (denominator)Producer and interestStatusEvidence
Juniper Research, ad-fraud whitepaper (September 2023)$84.2 billion; almost $100 billion; $172.3 billion2023; end of 2024; 2028Global; 2023 put at 22% of digital ad spend; no method stated; a broad and a click-based definition both givenCopy hosted by Fraud Blocker, a click-fraud vendor, with a Fraud Blocker chapter; sponsor not statedForecast
IAS, 20th Media Quality Report$100 billion2024Cites Juniper's forecastIAS, a verification vendorForecast cited as an estimate
CHEQ with Roberto Cavazos (University of Baltimore)$35 billion2020An assumed 10.5% "composite rate" times $332.84 billion of global digital ad spend (eMarketer)CHEQ, a bot and click-protection vendorForecast
ANA and White Ops (now HUMAN), Bot Baseline$5.8 billion; $14 billion without anti-fraud measures2019Bot fraud only; 50 ANA members' campaigns, August and September 2018, projected globallyWhite Ops analysed the dataProjection; counterfactual
TAG with the 4A's, ANA and IAB (Cunningham.Tech Consulting), US Ad Fraud Savings Report$10.8 billion saved2023$11.78 billion counterfactual (9.96% unfiltered IVT from DoubleVerify, IAS, Oracle Moat and Pixalate times $118.2 billion of US display and video revenue) minus $979 million (0.92% times $106.4 billion through TAG-certified and non-certified channels)The sponsors evaluate their own programmesCounterfactual model
TAG, European Ad Fraud Savings Report (release of 17 June 2025)EUR 3.45 billion saved2023Counterfactual from a 10.36% unfiltered IVT rate; subtracts losses in protected and unprotected channelsSame developer; TAG evaluates its own programmesCounterfactual model
DoubleVerify, release of 7 May 2026About $1.8 million per billion CTV impressionsNot statedUnprotected campaigns; price per thousand impressions (CPM), fraud rate and method not statedDoubleVerifyPer-unit estimate

They differ in definition, base and interest. Juniper's whitepaper gives a broad definition and a click-based one without saying which its forecast uses, CHEQ counts impressions, clicks, conversions and data events, and the ANA and White Ops counted bot fraud only . IAS sizes the problem with Juniper's figure in its own report . TAG's US saving subtracts actual losses only on the $106.4 billion that ran through certified and non-certified channels, so losses on the rest of the $118.2 billion, at the report's own unfiltered rate, count as saved; the European report, by the same developer, subtracts both .

Industry-wide ad fraud estimates, as published

Sponsored and commissioned estimates with their definitions and producers. Values are as published; none is the paper's own, and no two can be added or compared as if they measured one thing.

ProducerPeriodWhat is estimatedValueYearGeographyDefinition and scopeSponsor and interestCombinable?Src
Juniper Research2023ad fraud loss$84.2 billion2023globalad fraud defined as intentionally repeated clicking on PPC ads; 22% of digital ad spendwhitepaper copy hosted by Fraud Blocker with a Fraud Blocker chapter; sponsor not statedNo: each uses its own definition, scope and year
Juniper Research2028ad fraud loss$172.3 billion2028globalsamesameNo: each uses its own definition, scope and year
Juniper Research2024ad fraud lossalmost 100end of 2024globalsamesameNo: each uses its own definition, scope and year
Juniper Research2023spend recoverable by fraud mitigation platforms$23 billion2023globalsamesameNo: each uses its own definition, scope and year
Juniper Research2028spend recoverable by fraud mitigation platforms$47 billion2028globalsamesameNo: each uses its own definition, scope and year
IAS, MQR 20th edition2024ad fraud cost cited by IAS$100 billion2024globalcites Juniper (September 2023)IAS (verification vendor)No: each uses its own definition, scope and year
CHEQ with Prof. Roberto Cavazos2020direct cost of ad fraud$35 billion2020globalimpressions, clicks, conversions and data eventsCHEQ (bot and click protection vendor)No: each uses its own definition, scope and year
ANA and White Ops2019bot fraud losses$5.8 billion2019globalbot fraud; 50 ANA members' campaigns, August to September 2018White Ops (now HUMAN) analysed the dataNo: each uses its own definition, scope and year
ANA and White Ops2017bot fraud losses (previous study)$6.5 billion2017globalbot fraudsameNo: each uses its own definition, scope and year
ANA and White Ops2019bot fraud losses absent anti-fraud measures (counterfactual)$14 billionannual (2019 study)globalbot fraudsameNo: each uses its own definition, scope and year
TAG with 4A's, ANA, IAB (Cunningham.Tech)2023IVT losses absent anti-fraud programmes (counterfactual)$11.78 billion2023USdisplay and video; 9.96% x $118.2bnthe sponsors' own programmesNo: each uses its own definition, scope and year
same2023IVT losses in protected channels (modelled)$0.979 billion2023US0.92% x $106.4bnsameNo: each uses its own definition, scope and year
same2023savings attributed to anti-fraud programmes (counterfactual difference)$10.8 billion2023US11.78 - 0.979sameNo: each uses its own definition, scope and year
DoubleVerifynot statedfraud cost per billion unprotected CTV impressions$0.0018 billionnot statedglobalCTV; 'conservative estimates'; method not statedDV (verification vendor)No: each uses its own definition, scope and year
Machine-readable copy: data/estimates.csv.

Chapter 4

Whether a verdict decides money

A third-party report determines money in 2 of 16 readable entries on the fixed platform list, or 12.5%, in the 10 November 2026 coding . The cases are The Trade Desk's publisher terms and Microsoft Invest's route, whose end was announced for February 2026 . No paid settlement on a third-party verdict or matched price comparison was found . H2 is supported on that restricted reading, while coding all contracts in its registered population can cross the contradiction threshold .

In short

Of 17 platform entries fixed at registration, the initial screen read 16 . The Trade Desk makes accredited reports conclusive for publisher fees, and Microsoft Invest's refund route used an approved verifier's report . These give the platform-only count of 2/16, or 12.5%, rising to 25% with review-qualified clauses . Thirteen entries describe first-party validation . Publicly recorded settlements and matched price differences were not found in the dated screen . H2's broader registered population includes public contracts, whose inclusion and coding can move the share above 25% . The buyer contract RM6123 gives an unnamed tool billing force between client and agency but expired in June 2026 . The author retains H2 as supported under the registered listed-terms operational rule: 2/16 readable listed entries, or 12.5%. The broader population and treatment of review-qualified or trigger-only clauses remain reported sensitivities, not an unconditional claim across all public contracts .

The claim under test

Two of the author's essays set out the framing under test. "Where the Surplus Settles" (September 2026) says "A DoubleVerify verdict moves nothing": verification is "a procurement record rather than a price input", and an impression clears at the same cost per thousand impressions (CPM) verified or not. "The Risk You Can Price" (August 2026) says advertising has auditors and accreditation bodies but no adjudicator with published rules of evidence and the power to bind an escrow, a claim about disputes over outcomes and conversions. Both are rival framings under test, never evidence. At registration one public contract already contradicted the first: The Trade Desk's OpenPath terms .

H2's registered list covers 17 buying platforms and exchanges plus any public agency or advertiser contract found. It records a metric setting billable quantity, a third-party report deciding fees or remedies, an actual settlement on a verdict, and a matched price difference. First-party validation is recorded separately (Chapter 1). Third-party pre-bid targeting changes eligible supply without itself setting a refund . Both essay framings remain claims under test.

Sixteen of seventeen terms could be read

Sixteen of the seventeen entries could be read on 10 November 2026 . Index Exchange's exchange-quality and policy pages carry no buyer or seller billing terms . The verifier pass (a separate AI agent told to refute each finding the research agents made under the author's direction) later found Index's publisher-facing Marketplaces Terms of Use, last modified 13 May 2026, which pay publishers on Index's own records and give no third-party report a role . The count keeps Index unreadable, the cautious choice, since reading those terms in would lower the share; one entry in seventeen is far from the third that would leave H2 unresolved.

Readable terms almost all bill on the platform's own count. Google Ads advertisers buying viewable CPM (vCPM) on Display Network campaigns pay for impressions that Google's own Active View counts as viewable (a small portion of served impressions might also be charged), and DV360 bills outstream video only on viewable impressions; DV360's viewable CPM is otherwise a bid goal, and its pay-per-viewable insertion orders were archived on 1 August 2023 . Google Ads does not charge for invalid traffic (IVT) detected before billing and bases charges solely on Google's metrics . DV360's exchanges refund traffic Google considers invalid, with HUMAN (formerly White Ops) as an extra check feeding Google's decision . Amazon bases fees solely on its own measurements, and its traffic validation removes IVT detected within 30 days from invoices . Meta does not charge for impressions it determines invalid . Reddit's terms rule out charges based on third-party reporting, even from its approved measurement partners , and Magnite's own determination is dispositive once a buyer's dispute stays unresolved for 30 days . Yahoo's measurements are definitive , and Snap, Pinterest, X and LinkedIn bill on their own .

In 13 of the 16 readable entries the pages read describe the platform's own check of validity or viewability: first-party validation. For TikTok, Pinterest and X they describe none, though all three bill on their own data, and TikTok's terms make no commitment about invalid activity . Third parties' delivery counts do carry billing force in places, in Google's discrepancy credit and TikTok's 20% review, but those count delivery, not verdicts .

Where a third-party report decides

OpenPath clause 4.4 in v.10.31.24, unchanged from clause 4.5 in v.7.5.23, makes MRC-accredited reports used by The Trade Desk or its Ad Partners conclusive for fees owed to publishers . Invalid Inventory is unpaid or refunded within 30 days of notice, with a 1% monthly charge available for repeat sources . The supplier terms carry the same mechanism . Advertiser terms instead use The Trade Desk's own metrics and give no corresponding refund right . The screened records show neither invocation frequency nor named verifier reports .

Microsoft Invest's April 2025 documentation requires monthly claims backed by an approved partner's report, refunds the claimed quantity at Microsoft's average CPM, and adjusts seller accounts . No partner or binding policy text was found . Sellers retain a dispute route, which could lower the strict numerator by one . Microsoft announced the end of Invest buying for 28 February 2026, though pages still live on 10 November did not confirm retirement .

PubMatic needs a vendor's report but keeps the decision. Section 8.4 of its Demand Partner Master Services Agreement (page modified 18 August 2026) credits impressions deemed fraudulent on an MRC-accredited vendor's scoring, within the window and threshold tabled below, only if PubMatic "verifies in good faith" that they were suspect; no make-good applies to viewability or brand safety . The Trade Desk's SIVT Credit Program for advertisers worked the same way: claims needed a report from an MRC-accredited vendor such as Integral Ad Science (IAS) or DoubleVerify, the supplier's approval decided, and since late 2023, when its description was withdrawn, its status is unknown .

Yahoo's display terms, in the same document as its DSP terms, also govern managed Yahoo DSP buys placed by insertion order . They bill viewability-guaranteed vCPM line items on the viewable share measured by Yahoo or by an approved vendor the terms call "certified by the MRC", with Yahoo reserving a right to dispute vendor measurements and 10 days to reconcile; mobile and connected TV are among the exclusions. Some view-priced items may be invoiced on an approved vendor's count with no stated dispute right, which, folded in, would add one to the strict count .

Coded strictly, a third-party report has determining effect in 2 of the 16 readable listed entries, 12.5%; set aside Microsoft's route, whose end Microsoft announced, and only The Trade Desk's clause remains . Adding PubMatic's trigger-only credit and Yahoo's dispute-qualified vCPM clause, folded into the Yahoo DSP entry, takes the share to 25%, exactly a quarter . The figure codes every entry, and the templates beside them, by component.

DOES A VERIFICATION VERDICT DECIDE WHAT IS PAID? THE REGISTERED LIST AND THE PUBLIC TEMPLATESPlatformbills orcredits onits ownmeasurementA third-partycount setsthe billablequantity (a)A third-partyreportdecides a feeor credit (b)A settlementpaid on athird-partyverdict (c)Priceevidence,verified vsunverified(d)Google Adsyesnononone foundnone foundDisplay & Video 360yesnonoown detectionnone foundGoogle Ad Manageryesn/anonone foundnone foundAmazon Adsyesnononone foundnone foundMetayesnononone foundnone foundTikToknot statednononone foundnone foundSnapyesnononone foundnone foundPinterestnot statednononone foundnone foundReddityesnononone foundnone foundXnot statednononone foundnone foundLinkedInyesnononone foundnone foundThe Trade Deskyesoptionalyesnone foundnone foundMicrosoft Investyesnoyes (retired)none foundnone foundYahoo DSPyesnononone foundnone foundMagniteyesnononone foundnone foundPubMaticyesnotrigger onlynone foundnone foundIndex Exchangen/en/en/en/en/e◦ ANA media-buying template v3.0n/aoptionalyesn/an/a◦ IAB Terms, Direct Buy Addendumn/aoptionalnon/an/a◦ IAB Measurement Addendumn/aoptionalnon/an/a◦ 4A's/IAB Standard Terms v3.0n/aoptionalnon/an/a◦ Disney DAS IAB Addendumn/anoqualifiedn/an/a◦ Yahoo Display Termsyesoptionalqualifiedn/an/a◦ Spotify Master Termsyesnonon/an/aYes, in the terms readYes, in a route since retiredOptional or by agreementStarts a review; the platform decidesDecides subject to a dispute rightNoNone foundSettlement on the platform’s own detectionNot statedTerms not foundNot applicable (template)◦ marks public templates and contracts outside the list fixed at registration. The Trade Desk’s clause binds it andpublishers (supply side), not advertisers; Microsoft retired Invest buying on 28 February 2026. A trigger-onlyclause starts a review that the platform decides. Read between 9 and 10 November 2026; Index Exchange published nobilling terms that were found.
A third-party report decides money in two of sixteen readable terms, and no settlement was found. The platform and exchange terms on the list fixed at registration (hypothesis H2), with the public templates read alongside, coded for the four components: whether a third-party count sets the billable quantity, whether a third-party report decides a fee or credit, whether a settlement paid on a third-party verdict is documented, and whether verified and unverified impressions of the same inventory are shown to clear at different prices. Every large platform bills or credits on its own measurement.Units: coded state per term · Scope: 17 listed terms (16 readable) and 7 public templates and contracts, read 9 to 10 November 2026 · Sources The share moves with coding: 12.5% on the strict count, up to 25% if trigger-only and qualified clauses are counted. Executed contracts are private; this is what public terms say.Data: data/figure-data.json#money

Windows, thresholds and who decides

The terms also set claim windows and who rules. On the list, the platform rules everywhere except where a report is conclusive (The Trade Desk's supply-side clauses) or sets the refunded quantity (Microsoft Invest) . The windows start from different points and are not one series.

TermMechanismWindowThreshold or formulaWho decidesEvidence
Google AdsAdvertiser-requested IVT investigationActivity in the last 60 daysNone statedGoogle
Google Ads (US terms of 29 July 2025 and 1 July 2026)Third-party ad server discrepancy creditClaim within 60 days of the invoiceGap above 10%; credit of (90% of Google's count minus the ad server's) times the average CPM or vCPMGoogle
Google Ad ManagerAdvertiser credits behind publisher deductionsTraffic in the 60 days before the invoiceNone statedGoogle; no appeal
Amazon AdsIVT monitoring; fee claimsUp to 30 days; claims within 60 days of the chargeGap of up to 5% from third-party IVT rates expectedAmazon
TikTokThird-party count review (reservation display ads); payment disputesDisputes waived a year after the chargeReview only above a 20% difference; TikTok's data final if no error is foundTikTok, in a joint review
Pinterest; XCharge disputes60 daysNone statedThe platform
LinkedInFee disputes and invalid-activity credit claims90 daysCredit at LinkedIn's sole discretionLinkedIn
RedditAds Credits claim period30 daysCredits at Reddit's discretionReddit
YahooCharge disputes; vCPM vendor reconciliation30 days of the invoice; 10 daysViewability minimum set per orderYahoo; the parties for vCPM
MagniteNotice of invalid activity10 days after the invoice; Magnite decides after 30 moreReasonable evidenceMagnite
PubMaticNoncompliant Impression report30 days after month-endOver US$500 a month; capped at winning bidsPubMatic
The Trade Desk (OpenPath, supplier terms)Seller refund of paid Invalid Inventory30 days after The Trade Desk's notice1% of the month's payable for repeat sourcesThe reports, deemed conclusive
The Trade Desk (advertiser terms)Supply discrepancy30 days of the invoiceThe Trade Desk bears gaps above 10%The Trade Desk
The Trade Desk (SIVT Credit Program, withdrawn 2023)Client SIVT claim30 days of the incidentOver US$1,000 of media spendSupplier approval
Microsoft Invest (end announced for 28 February 2026)IVT refund claim10 days after month closeNone statedThe partner's report sets the quantity

Templates and contracts outside the list

The population also takes in public contracts: templates from the ANA (Association of National Advertisers), the IAB (Interactive Advertising Bureau) and the 4A's (American Association of Advertising Agencies), sellers' own terms and one government framework. The paper reads them beside the list; the verifier pass argued they belong in the test, and the verdict below reports both readings .

DocumentDateWhat it does with a verification verdictCoded (b)Evidence
ANA Master Media Buying Services Agreement Template v3.0June 2023The agency must make media owners accept the advertiser's ad-verification tool as the basis of the billable metric, and seek terms under which the advertiser does not pay for IVT, brand-unsafe or non-viewable placements the tool identifiesYes, in template language
IAB General Terms v1.1, Direct Buy Addendum v1.0, Appendix 1 v1.1Released 18 May 2026Controlling measurement set per order; an ad verification company is mutually agreed; remedies mutually chosenNo default
IAB Measurement Services Addendum v1.0Posted as final 28 October 2026; the file reads "DRAFT"Provides for a provider's outputs designated as the controlling measurement under another agreement; sets no defaultNo default
4A's/IAB Standard Terms v3.02010A guideline-compliant third-party ad server's delivery count controls invoices; adjacency remedy on written request, with no named decider; no verification termNo
OpenX addendum to the 4A's/IAB termsUndatedOn an unresolved discrepancy, pays on the average of the two measurements; no verification termNo
Disney Advertising addendum, US media buys28 March 2023Verifier reports decide adjacency compliance; Disney reviews the incident list before any remedy; its own ad server controls billingQualified
Yahoo Display Advertising Program Terms, clause 10.11 January 2026vCPM billed on Yahoo's or an approved vendor's viewable share, subject to Yahoo's dispute rightQualified
Spotify Master Advertising Terms1 July 2026Spotify's metrics control over third-party statistics unless an order says otherwiseNo
RM6123 Media Services framework (UK public sector)14 December 2021 to 13 June 2026 (expired)Where ad fraud tools are used, the client pays only for what the tool verifies as human trafficYes, between client and agency

The ANA template is the strongest buyer ask in writing, but no member has to use it, its non-payment clauses bind the agency only to commercially reasonable efforts, and whether media owners accept them is not public .

No settlement and no price difference found

In the sources screened by 10 November 2026, no settlement paid on a third-party verdict was found . The one advertiser settlement over invalid-traffic credits found, in AdTrader v. Google, was given final approval on 1 November 2022: a non-reversionary US$7 million fund for US AdWords (now Google Ads) advertisers who were not refunded for ads on Google AdX publishers' sites although Google withheld those publishers' payments for invalid activity or other breach. The money moved on Google's own determinations, and a settlement is not a finding of liability . Uber's US$6 million settlement with Phunware (9 October 2020), over fraud claims, names no verification vendor . The screen was thin, with no docket search by vendor name, and credits paid privately would not appear in public records .

Parties say verdicts move money; none documents a payment. DoubleVerify's May 2025 complaint against Adalytics, a party's pleading, says general invalid traffic (GIVT) is deducted from the billable count before a customer is billed and SIVT is flagged so the advertiser can seek compensation from the seller; it cites no contract, platform term or settlement, and MRC says it has no insight into billing arrangements . A 2019 release from the ANA and White Ops (now HUMAN) said most fraudulent impressions were by then invalidated by platforms, filtered before payment or recovered through claw backs, but it recorded no claw back . Pixalate has sold a signed Clawback Report for refund negotiations since 4 November 2025, but its announcements name no counterparty or amount refunded . The verifiers' own terms promise no more: IAS's and DoubleVerify's make their statistics determinative, or potentially so, for billing their own fees, and Pixalate's bar clients from using its output in litigation or advertising claims without its consent .

Component (d), the essay's "same CPM", is neither shown nor refuted: no public evidence found by 10 November 2026 compares the prices of verified and unverified impressions of the same inventory . The closest are publishers' statements to Digiday in December 2017, one reporting only a 13% CPM gap between its most and least viewable units, another, anonymous, roughly estimating CPMs about 20% higher per 30-point gain in viewability; those compare viewability levels, not verification status, and billing on viewable impressions makes a verdict the unit paid for, not a different price for the same impression . The ANA's TrueCPM, $7.46 for higher-performing and $19.04 for lower-performing advertisers per thousand quality-tested impressions (27 May 2026), compares advertisers, not inventory .

The verdict on H2

H2 requires third-party determining effect in no more than 25% of readable listed terms, with no documented paid settlement or matched price difference. Its registered population also includes public agency or advertiser contracts. The platform-only reading meets the support test at 2/16, or 12.5%, and 25% on lenient coding . Across all contracts found, strict coding gives 4/26 to 4/23, while including review-qualified or trigger-only clauses can exceed 25% . The rule counts the listed terms, so H2 is supported on its registered test; the wider contract population and the qualified clauses are reported as sensitivity, not as the test .

Three qualifications travel with it. The rule names no tier for clauses that keep a review or dispute right, a gap the verifier pass flagged along with the treatment of ended terms; on the list the verdict holds at every tier, and dropping Microsoft's route only lowers the share . Across every public contract found as well, RM6123, the OpenX addendum and Index's publisher terms among them, the share stays under a quarter on strict coding, falls either side of it when review-qualified clauses are counted, and passes it when verification-gated triggers are added; on terms in force at reading it stays under a quarter at every coding . And the party statements above document no payment: DoubleVerify's pleading describes billing mechanics, which H2 counts as neither (b) nor (c), and the 2019 release asserts recoveries without a verdict source, payer or amount; were it read as evidence of (c), the rule would leave (c) unresolved .

What buyers write down

Buyers document verification mostly as demands, guidance and certification duties . The one public buyer contract found that gives a tool's verdict determining effect is the UK government's RM6123 Media Services framework (14 December 2021 to 13 June 2026). Where ad fraud tools are employed, the public-sector client pays only for impressions, clicks or conversions the tool verifies as human traffic, and the agency must use reasonable endeavours to recover the value of fraudulent, non-viewable or unsafe ads it paid for . The clause binds client and agency, not the seller; it names no vendor, and the tool need not be independent or accredited . No record shows money withheld or recovered under it, and "human" is the contract's word: an IVT tool classifies an impression it does not filter as valid . On 10 November 2026 the government's upcoming-agreements page listed no media-buying successor .

The rest is demand. The Global Media Charter of the World Federation of Advertisers (WFA), dated 21 June 2018, asked for refunds of all media investment associated with invalid traffic, fees and commissions included . The ANA's Complete Report (December 2023) tells advertisers not to pay for invalid, unmeasurable or non-viewable impressions and to settle refund processes with DSPs in advance . ISBA, the UK advertisers' body, tells advertisers to ask whether they pay for blocked impressions . Member-only templates could not be read, so their clauses are unknown, not absent .

The ANA's study used vendors' classifications for measurability, viewability and IVT (Chapter 3) . Its Q3 2025 release called the new risk measure an AI-powered standard without naming the classifier, while the benchmark report identified Mobian and reported 99.1% of spend in low-risk environments . Those are measured outcomes without billing force.

Who pays, and what the seals certify

Advertisers pay DoubleVerify per thousand impressions, often through agencies or platforms . IAS pays DSPs a share of revenue earned through their platforms . The ANA's 2023 member survey reports 78% holding direct verification contracts but publishes no response count . No verification rate card was found in that screen . Chapter 7 covers seller revenue and interests.

Every holding company in scope shows TAG seals for a buying entity, but a seal certifies process, not a verifier. TAG's Brand Safety Certified Guidelines v5.0 (July 2026) require certified buyers, agencies included, to define make-good and clawback procedures and to screen all transactions with content verification services that have passed independent validation or with inclusion and exclusion lists, so lists alone can meet the rule . TAG's Certified Against Fraud rules add a duty to filter all transactions for invalid traffic . The registry read on 10 November 2026 shows:

Holding companyRegistry entityBrand Safety CertifiedCertified Against FraudEvidence
WPPWPP MediaIndependent validation; Asia Pacific, UK, US, IrelandSelf-attested; North America
PublicisPublicis Groupe/MMS USA HoldingsIndependent validation; globalIndependent validation; global
OmnicomOmnicom Media GroupIndependent validation; UKSelf-attested; North America, Asia Pacific, Africa, Central and South America, Middle East, Oceania, Russia
Omnicom (IPG since 26 November 2025)IPG MediabrandsIndependent validation; globalIndependent validation; global
dentsuDentsu Aegis NetworkIndependent validation; Ireland, UKIndependent validation; global
HavasHavas MediaIndependent validation; UK, IrelandIndependent validation; UK

The registry's names lag corporate change (IPG Mediabrands, Dentsu Aegis Network) . Omnicom's and WPP's FY2025 annual reports say nothing about brand safety, viewability, invalid traffic or ad fraud .

Disputes on the record

Both public verifiers' filings acknowledge billing disputes without naming any, and on 10 November 2026 CourtListener's RECAP index showed no suit by an advertiser or agency against either by case name, a search that misses sealed matters, arbitrations and state courts . The register's other items record no payment on a third-party verdict (Chapter 9):

DateItemWhat it isPostureEvidence
3 June 2020Sizmek DSP v. DoubleVerify (Bankr. S.D.N.Y., Adv. No. 20-01119)Adversary complaint by a buy-side platform in bankruptcyTerminated 1 March 2021; no document public
3 June 2020Sizmek DSP v. Integral Ad Science (Adv. No. 20-01134)Adversary complaint by a buy-side platform in bankruptcyClosed by so-ordered stipulation on 12 April 2021; no document public
6 August 2024Rumble v. WFA, Diageo, WPP plc and GroupM Worldwide (N.D. Tex.)Antitrust complaint by a publisherDismissed without prejudice on 13 August 2025 for lack of personal jurisdiction and improper venue, a procedural ruling
7 August 2024Adalytics report on brand-safety technologyIndependent report relaying statementsRelays that a brand was told its campaigns were 100% brand safe; records no contract, refund or dispute outcome
7 February 2025Adalytics report on an image hostIndependent report relaying statementsAn intention to seek refunds stated; none documented
26 September 2025Federal Trade Commission Decision and Order, Omnicom and IPG (Docket C-4823)Consent order, a settlement and not a findingBars the merged agency, in the US, from steering spend or relying on exclusion lists on "Covered Bases" such as political viewpoint; client-directed lists allowed; compliance monitor

What the finding means for the framing

The public terms do not bear out "moves nothing" as a general claim: a third-party report decides fees in The Trade Desk's supply-side terms, decided refunds in Microsoft Invest's route, whose end was announced, sets vCPM billing in Yahoo's display terms subject to dispute, and is the billing basis in the ANA's template and the UK's expired RM6123 framework . As a description of the largest platforms' own billing terms, the line holds: Google, Amazon, Meta, TikTok, Snap, Pinterest, Reddit, X and LinkedIn bill on their own counts, and any verdict that moves money there is their own . The "same CPM" half is neither shown nor refuted . On who decides, the terms read name the platform, the parties or a report deemed conclusive, and none names an outside body that hears disputes under published rules of evidence; that fits the adjudicator essay as far as these terms go, though the essay concerns conversion disputes . All of this rests on public terms read on 10 November 2026, to be rechecked at the freeze; executed contracts, negotiated orders and privately paid credits were outside reach.

Chapter 5

Inside the platforms

Where the nine platforms document a route, third-party verification mostly uses platform-supplied data . Meta supplies filtered impressions, YouTube routes partners through Ads Data Hub, and Amazon runs vendor detection inside its environment . TikTok's documented OM SDK route contradicts H4 under the registered comparator . That conclusion carries medium confidence because the route was described in 2021–2022 and TikTok controls the integration .

In short

Meta, YouTube and Amazon document platform-supplied routes, with different data limits . Snap and LinkedIn document trackers without establishing where they fire, while X, Pinterest and Reddit leave the route unstated . Four of nine platforms have accredited third-party measurement, one has third-party content-level measurement and six have accredited first-party measurement, on MRC's 10 November 2026 list . Meta and YouTube third-party viewability is accredited at GIVT level . Documented pre-bid controls are executed by the platform . Post-bid blocking remains undocumented outside Amazon's unresolved tag scope . H4 is contradicted on TikTok under the literal tag comparator, at medium confidence . Platform-reported pass rates remain bounded by the impressions and content made measurable .

How a route is coded

H4, registered on 9 November 2026, predicts that on each of the nine platforms (Meta, YouTube, TikTok, Amazon, X, Snap, Pinterest, Reddit and LinkedIn), documented third-party brand-safety, suitability and viewability measurement of the platform's own placements runs on data the platform supplies, "not on the verifier's own code or crawler observing the placement" . Its comparator is "the verifier's own tag or SDK running in the placement, or its own crawl of the content adjacent to the ad" .

Data access limits what partners can check. MRC says IAS lacks user- and device-level data on Meta and cannot independently filter IVT there . DoubleVerify's FY2025 filing says integration partners control its platform solutions and have supplied inaccurate signals . It says it uses its own tags when possible, while IAS says methods vary by platform .

AI research agents working under the author's direction read the platforms' help, policy and developer pages, the verifiers' integration pages and MRC's records on 9 and 10 November 2026, and a separate AI verifier, told to refute each finding, re-checked it against its sources in what this paper calls the verifier pass (Chapter 13). Each platform, metric and verifier was coded with one of four routes: platform-supplied (a feed, interface, log, clean room or server-to-server processing inside the platform's environment), verifier-executed (the verifier's own tag, SDK or crawler), a vendor tracker the platform accepts, or not stated, which is the code for a partner named without a route . Off-platform supply that a platform sells, such as LinkedIn's Audience Network, is outside H4 . Every entry in the table is bounded to the pages screened.

PlatformPartners documentedViewability and invalid trafficBrand safety and suitabilityEvent-level or aggregateEvidence
MetaDoubleVerify, IAS, Zefr and Adloox, on Meta's pagesPlatform-supplied: impressions Meta measures, with Meta's invalid-traffic removalPlatform-supplied on Feed and in-stream video, inferred from MRC's description of Meta's pipeline (under review); not stated for Reels proper, Stories or ThreadsImpression-level without user or device identifiers; per-content impression estimates, hourly
YouTubeDoubleVerify, IAS and Zefr, on Google's pagesPlatform-supplied through Ads Data Hub; no third-party pixels acceptedExposure through Ads Data Hub; how vendors obtain the content not statedPrivacy-checked aggregates of roughly 10 to 50 users a row (inference)
TikTokDoubleVerify, IAS and Zefr, on TikTok's pagesVerifier-executed: the vendor's tag through TikTok's OM SDK (2021, 2022); not restated in 2026IAS: the same integration (2022), content route not stated; DoubleVerify and Zefr: not statedNot stated
AmazonDoubleVerify, IAS and Adloox, in Amazon's policyPlatform-supplied: server to server, detection inside Amazon's environment; monitoring tags also listed for supply not namedServer to server per Amazon's policy; scope not statedNot stated
XIAS and DoubleVerify, on X's pagesNot statedNot stated (DoubleVerify says it classifies the posts above and below ads)Not stated
SnapIAS and DoubleVerify, and Zefr for brand safety, on Snap's pagesApproved vendors' 1x1 trackers, fired when the ad renders; where they fire from not statedIAS trackers required; content route not statedNot stated
PinterestIAS and DoubleVerify, on Pinterest's and the vendors' pagesNot statedNot statedNot stated
RedditIAS on Reddit's pages; DoubleVerify in its own releaseNot statedNot statedNot stated
LinkedInDoubleVerify, and IAS for video ads only, on LinkedIn's pagesImage or tracking tags; JavaScript and iFrame tags refusedNo third-party measurement of the feed documentedNot stated

The figure shows the coded routes beside MRC's status for third-party and first-party measurement on each platform .

INSIDE THE NINE LARGE PLATFORMS: WHERE THIRD-PARTY MEASUREMENT GETS ITS DATAViewability andIVTBrand safety andsuitabilityMRC: third-partymeasurementMRC: theplatform's ownmeasurementMetaplatform dataplatform dataaccreditedaccreditedYouTubeplatform dataplatform dataaccreditedaccreditedTikTokverifier tagverifier tagaccreditedAmazonplatform dataplatform dataaccreditedaccreditedXnot statednot statedSnapplatform-fired tagnot statedaccreditedPinterestnot statednot statedaccreditedRedditnot statednot statedLinkedInplatform-fired tagnot statedaccreditedRuns on data the platform suppliesVendor pixel the platform firesVerifier’s own code in the placementRoute not documentedMRC-accreditedNo MRC accreditationTikTok’s route rests on DoubleVerify’s (2021) and IAS’s (2022) documents describing measurement through TikTok’sOpen Measurement SDK; no 2026 page restates it, and TikTok controls the SDK (medium confidence). Snap’s andLinkedIn’s first-party accreditations cover the platforms’ own measurement, not a verifier’s. Read 10 November2026.
On eight of nine platforms, third-party verification checks the platform’s own data or cannot say how it gets it. For each of the nine large platforms, the documented route by which third-party verifiers measure viewability, invalid traffic and brand safety on the platform’s own placements, and whether MRC accredits third-party measurement there or only the platform’s own. Only on TikTok is the verifier’s own code documented as running in the placement.Units: coded route per platform and metric · Scope: Meta, YouTube, TikTok, Amazon, X, Snap, Pinterest, Reddit, LinkedIn; documents read 10 November 2026 · Sources A partner’s name alone is coded "not stated". Platform pages change often; the routes are status-sensitive.Data: data/figure-data.json#platforms

Meta, YouTube and Amazon supply the data

Meta's partners measure what Meta passes them. MRC accredits DoubleVerify's and IAS's reporting of Meta impressions and viewability only on placements that Meta itself measures and has accredited, net of GIVT, and IAS's letter of 9 July 2026 records that, because of limits Meta imposes, IAS cannot independently identify and filter invalid traffic there and has no user- or device-level data . The French competition authority's interim-measures decision of 4 May 2023, an account of the parties' statements rather than a ruling on the merits, describes Meta supplying the data through a programming interface, removing the invalid traffic it identifies before passing impression-level data to viewability partners, and not letting partners collect raw data with their own technology .

For suitability, Meta's announcement page dates its partners from Zefr (30 March 2023) to Adloox (24 October 2024) without saying what data it supplies . MRC's under-review list, read on 10 November 2026, carries a Meta Third-Party Brand Suitability Pipeline, audit reviewed in August 2026: defined data fields that Meta delivers to partners for the text, image, audio and video in Facebook in-stream video and the Facebook and Instagram feeds . Meta's Content Delivery Reports interface gives partners, for Facebook in-stream and Reels overlay ads, each public content item's ID with estimated impressions, hourly, over a 15-day lookback, and the IDs can be used to retrieve the video file; neither page names a verifier . How partners obtain content on Reels proper, Stories and Threads is not documented in the pages opened .

Google accepts no third-party tracking pixels for YouTube measurement, and enabling a vendor in Google Ads means choosing to share campaign data with it . DoubleVerify's, IAS's and Zefr's accredited YouTube viewability is processing of Google's Ads Data Hub (ADH) data at GIVT level, while Google's own ADH accreditation covers its metrics including SIVT and the passthrough reporting that vendors generate by querying ADH, which MRC's report for the first quarter of 2026 calls templated vendor queries . Google says ADH's privacy checks need roughly 10 to 50 unique users per result row and are designed to help prevent the transmission of data about individual users, so verifiers query event data Google holds without receiving event-level records; that is an inference, though Google's page names measurement partners among ADH's users .

For content, exposure reaches Zefr through ADH, but how the vendors obtain the video they classify is not documented on the pages opened; MRC's chart calls Zefr's accredited method independent URL classification . YouTube's derived-metrics policy, last updated on 14 September 2026, reserves the impression-level data needed for safety claims about specific impressions to Brand Safety Reporting Partners, whom it does not name . Channel Factory says it uses YouTube's interface and its own engine to scan video, audio and metadata and build suitability lists, which this paper reads as a pre-bid list product . Adalytics argued in a report public by 27 June 2023 that YouTube's measurement vendors rely on aggregated server-to-server transfers (Chapter 3) .

Amazon says its server-to-server approach lets DoubleVerify and IAS apply their detection methods inside its secure environment, and its policy, last updated on 1 June 2026, offers DoubleVerify, IAS and Adloox on its demand-side platform (DSP) through server-to-server integrations and refuses JavaScript tags for third-party tracking in its mobile app . The same policy lists JavaScript monitoring tags on creatives hosted by Amazon DSP, and monitoring-and-blocking tags on creatives from third-party ad servers, without naming the supply, so verifier code on some Amazon placements is not excluded . Which data fields Amazon passes is not documented on the pages opened . Amazon says its own systems, using signals from HUMAN (White Ops until March 2021), exclude identified invalid traffic from reports and invoices: a first-party verdict (Chapter 4) .

TikTok runs the verifier's tag

TikTok is where the record differs. DoubleVerify (28 September 2021, read in a syndicated Business Wire copy) and IAS (9 May 2022) announced that they measure viewability and invalid traffic on TikTok's placements through TikTok's integration of IAB Tech Lab's OM SDK, and IAS said the same on 17 October 2022 of post-bid brand safety and suitability, in markets it did not name . In the OM SDK the app sends measurement signals and the provider's own tag collects them, which for viewability and invalid traffic is verifier-executed observation under H4's comparator .

No 2026 source opened restates the OM SDK route, and TikTok controls which scripts load and what signals they receive . Its current IAS-tag requirement supports continuity, but MRC separates DoubleVerify's TikTok traffic from its tag-based reporting . The impression tag also leaves IAS's content-acquisition route unstated .

TikTok's help pages offer DoubleVerify and IAS invalid-traffic and viewability measurement on the For You Feed and TikTok Ad Network (formerly Pangle); DoubleVerify's metrics there include Total Calls and Prerender Calls, names consistent with code calling home from the device but no proof of it . MRC granted DoubleVerify initial accreditation for TikTok in-app video impressions and viewability, filtered for GIVT and SIVT, on 13 April 2026, and in December 2025 removed TikTok itself from in-process status because it did not intend to take part in a third-party audit . None of TikTok's pages opened says what content data it gives its brand-safety partners .

Snap and LinkedIn accept trackers

Snap documents approved tracking URLs, including a 1×1 impression tracker and IAS tags for suitability . LinkedIn accepts image and tracking tags but rejects JavaScript and iFrames, with DoubleVerify's display 1×1 URL and an undescribed video tag . Its suitability offer applies only to the off-platform Audience Network .

Neither platform establishes whether the request fires from the device or its servers . The verifier pass therefore treated the viewability route as unresolved, rather than accepting the research's platform-mediated inference .

X, Pinterest and Reddit name partners only

X (formerly Twitter) names IAS and DoubleVerify on help pages read in a browser on 10 November 2026, for viewability and brand-safety measurement and for post-bid adjacency measurement in the Home Timeline, defining adjacency as one organic post above the ad and one below . DoubleVerify said in January 2023 that it classifies those posts, but no page opened says how they reach the verifiers; that X must supply them is inference . Pinterest, DoubleVerify and IAS document viewability and invalid-traffic measurement since 2020 and brand-safety reporting since 2024, with no route on any page opened .

Reddit offers a Limited inventory tier, powered by IAS, to managed accounts, and DoubleVerify and IAS describe measurement on its feed and conversation placements . Reddit's tracker page names ten approved partners, as some of those it works with, and neither verifier . A trade report called DoubleVerify's integration server to server; no primary page opened confirms it . Zefr announced a Reddit partnership on 30 September 2026 that no Reddit page opened names .

What MRC accredits on the platforms

Accreditation on the platforms is MRC's, set letter by letter; it certifies a scoped measurement process against MRC's standards; the public accreditation letter does not supply the sold-category error-rate estimate defined in H3 (Chapter 2). On the list read on 10 November 2026, third-party measurement of a platform's own placements is accredited on four of the nine platforms (Meta, YouTube, TikTok and Amazon), content-level measurement only for Zefr on YouTube, and the platforms' own measurement on six . A membership or a pre-assessment is not accreditation: X says it is an MRC member that completed a pre-assessment in December 2022 without a full audit, and MRC's chart of 22 October 2026 shows X as not in MRC audit . Reddit appears on neither MRC list nor the chart . Each row's filtration level, GIVT or SIVT, is the invalid-traffic standard its letter applies .

PlatformThird-party measurement accreditedThe platform's own accredited measurementEvidence
MetaDoubleVerify: display and video impressions and viewability, desktop, mobile web and in-app, net of GIVT, Facebook and Instagram feeds only (13 April 2026). IAS: the same metrics net of GIVT on Facebook Feed in all three environments and, in-app, on Instagram Feed and Stories and Facebook in-stream video; Reels and Audience Network excluded (9 July 2026)Facebook and Instagram impressions and video plays on desktop, mobile web and in-app, and first-party viewability report metrics (in-app video on MRC's list), by placement, including SIVT filtration (2 October 2026); content-level controls revoked in September 2025 on Meta's withdrawal
YouTubeDoubleVerify (13 April 2026), IAS (9 July 2026) and Zefr (6 November 2025): Ads Data Hub video impressions and viewability, desktop, mobile web and in-app, GIVT level (IAS's higher level expressly not SIVT), in-stream and bumper; Shorts, in-feed and masthead excluded. Zefr: content-level labelling and reporting of English-language in-stream video and music video, desktop, mobile web, in-app and connected TV (CTV); Shorts, YouTube TV, Google TV, masthead, Google Video Partner videos and volumetric data excluded (20 January 2026)Ads Data Hub metrics for in-stream YouTube and Google Video Partner inventory, desktop, mobile web and in-app, including SIVT, with third-party passthrough reporting (20 January 2026); Inventory Modes content-level controls on in-stream video and Shorts sold through Google Ads and Display & Video 360, desktop, mobile web and in-app; masthead, YouTube Kids and live streams excluded (28 May 2026)
TikTokDoubleVerify: video impressions and viewability, in-app, filtered for GIVT and SIVT (initial grant, 13 April 2026)None; removed from in-process status in December 2025
AmazonIAS, server to server: display and video impressions, viewability, tracked ads and invalid traffic, desktop, mobile web and in-app, for Amazon Retail, IMDb and third-party supply; Fire TV and Twitch excluded; SIVT in the letter's body, net of GIVT in its enclosure (7 November 2025)Amazon DSP display impressions and click-throughs for Amazon Retail, IMDb, Amazon Publisher Services and third-party exchanges, net of GIVT, and a first viewability grant across those sources and Alexa devices, net of GIVT (29 May 2026); Twitch and Prime Video video ads neither accredited nor in audit
XNone; platform not in MRC auditNone; pre-assessment in December 2022 and no full audit, by X's account
SnapNone accredited or in processImpressions, viewability, swipe-ups, reach and video duration, in-app, GIVT (29 July 2026); desktop web denied in September 2025
PinterestNone accredited or in processDisplay impressions and pin clicks (outbound clicks excluded), desktop, mobile web and in-app, GIVT (20 January 2026)
RedditNo entryNo entry
LinkedInNoneImpressions and clicks on LinkedIn.com and its app, desktop, mobile web and in-app, net of GIVT; Audience Network excluded (28 May 2026)

On Meta and YouTube, where the verifiers process data the platform supplies, their accredited viewability stops at GIVT while the platforms' own measurement includes SIVT, and MRC's IAS letter gives the reason for Meta . The two third-party accreditations on the nine platforms that include SIVT are DoubleVerify on TikTok, where the route documented in 2021 is the verifier's tag in the app, and IAS on Amazon, where detection runs inside Amazon's environment . A withdrawal such as Meta's is not a finding on accuracy .

MRC's own records also disagree. Its chart of 22 October 2026 footnotes Amazon DSP third-party viewability as accredited for DoubleVerify and in process for IAS, the reverse of the list and letters read on 10 November 2026, which show IAS accredited and no DoubleVerify Amazon entry . The chart also shows DoubleVerify accredited on Instagram Stories and Facebook in-stream Reels video, beyond its feeds-only letter; MRC says the chart does not reflect each platform's full scope, so the letters and list are treated as the operative record .

MRC's 2018 content-level supplement makes platform cooperation a condition of third-party content-level accreditation: the platform allows direct third-party content-level measurement or provides validated first-party data . Two platform data services were under MRC review on 10 November 2026: Meta's suitability pipeline and Google's YouTube Relay Log, whose aggregated views and watch time for measurement partners are audience data rather than verification . IAS's YouTube brand-safety classification has been in audit since April 2023 .

Lists before the bid, little after it

Every third-party pre-bid control whose mechanism is documented is a verifier-supplied list, classification or tier that the platform executes, such as TikTok's partner-managed video exclusion lists, which take up to 200,000 videos at a time, and Meta's partner block lists . Meta names Scope3 (now Apostra), which it says acquired Adloox's measurement operations, as a block-list partner; Peer39 has owned Adloox since June 2026 . Pixalate documents pre-bid filtering inside Amazon DSP that Amazon configures and runs; Amazon's own pages opened do not name Pixalate .

No screened page documents verifier post-bid blocking on platform-owned placements outside Amazon . Amazon accepts blocking tags without identifying supply, while IAS's letter includes blocked-display metrics for some Amazon sources . Sampling rates are unstated, and only Meta, Snap, IAS-on-Snap and IAS-on-Pinterest document reporting freshness . Billing remains a separate question (Chapter 4).

PlatformThird-party pre-bid controlPost-bid blocking by a verifierEvidence
MetaPre-screen and block lists from DoubleVerify, IAS, Zefr and Scope3, applied by MetaPeer39 describes Adloox blocking; no mechanism stated
YouTubeNot stated for YouTube; Channel Factory's lists, built through YouTube's interface (a pre-bid reading)None documented
TikTokVideo exclusion lists managed by DoubleVerify, IAS or Zefr and executed by TikTok, for Standard and Limited inventory; IAS pre-bid in unnamed markets (2022)None documented
AmazonPixalate's pre-bid invalid-traffic filter, configured and run by Amazon, by Pixalate's accountOpen: blocking tags accepted on supply not named; blocked-ad metrics in IAS's letter
XIAS pre-bid on the vertical video surface only, for US advertisers who opt in; mechanism not statedNone documented
SnapIAS floor classifications sent to Snap, which acts on themNone documented
PinterestNot statedNone documented
RedditLimited inventory tier powered by IAS, managed accounts onlyNone documented
LinkedInNone on the feed; DoubleVerify pre-bid filtering on the Audience Network, which is off-platform supplyNone documented

The platforms' own controls and numbers

Each platform also offers its own controls. Among the nine, only Google's YouTube Inventory Modes hold MRC content-level accreditation, for in-stream video and Shorts; Meta's was revoked in September 2025 on its withdrawal, and the other platforms' tiers, such as TikTok's Inventory Filter, Snap's Brand Suitability Suite and X's Sensitivity Settings, are unaccredited . TikTok calls its invalid-traffic handling certified under the Trustworthy Accountability Group's (TAG) Certified Against Fraud programme and powered by Protected by Mediaocean (formerly Protected Media), whose MRC accreditation is the vendor's own and does not extend to TikTok; the page does not say which TikTok traffic the vendor filters . Which framework edition each tier follows is Chapter 8's question.

The table's pass rates come from inventory sellers and apply to the content and impressions made measurable . IAS defines its denominator in two ways, all served ads and measured impressions . Figures invoking GARM use a framework whose owner announced discontinuation on 9 August 2024 .

PlatformWhat the platform statesBase and period, as statedMeasured by, per the platformEvidence
X99.99% brand safety and 97% brand suitability, as averages2025; base not statedIAS and DoubleVerify
XMore than 99% compliance with industry-defined brand-safety floorsPeriod and base not statedIAS and DoubleVerify
XMore than 99% efficacy of its keyword and account adjacency controlsPeriod and base not statedX itself
Snap99% of Spotlight and creator content brand safe against the GARM floorFour beta advertisers, two weeks of live campaigns, US and EU, May 2024IAS
Snap99% of Spotlight and Creator Stories content brand safe on average30 days before 24 February 2025; base not statedIAS
MetaUnder 1% of Facebook Feed content in GARM's high-risk categoryEarly testing, before 30 March 2023; base not statedZefr
MetaMore than 99% of content next to ads in Feed and Reels brand safe30 days before 13 February 2025; brand safety not definedMeta's brand safety and suitability partners
TikTokMore than 26 billion impressions measured, in over 20 languagesThe year before 19 May 2023DoubleVerify, IAS and Zefr
TikTokMisinformation adjacency below 0.1%May to August 2024Zefr

H4 under its rule

H4's contradiction rule requires one documented verifier-executed route on a platform's own placements . TikTok meets the literal test on DoubleVerify's 2021 and IAS's 2022 OM SDK releases for viewability and IVT . Confidence is medium because the current route is not restated and TikTok controls the integration . A platform-control interpretation would change that coding, so it remains an explicit alternative for the author .

On the other eight platforms, the pages screened document no verifier code running in the platform's own placements, Amazon's tags on unnamed supply aside, and no verifier crawl of the content next to ads . Platform-supplied routes are documented for Meta's feeds and in-stream video, for YouTube and for Amazon; Snap and LinkedIn document trackers without saying where they fire; X, Pinterest and Reddit name partners without a route, which the rule codes unresolved . Outside TikTok, the brand-safety and suitability route is platform-supplied where documented (Meta, YouTube's exposure data and Amazon) and unresolved on X, Snap, Pinterest, Reddit and LinkedIn's own feed . One platform of the nine has a documented verifier-executed route, on documents from 2021 and 2022, and every route here is due to be rechecked at the freeze on 11 December 2026 .

Chapter 6

CTV and in-app

On MRC's list read on 10 November 2026, five scored products hold CTV SIVT accreditation, two hold certified-app video viewability, and Zefr holds English-language YouTube in-stream content-level accreditation . Streaming programme signals outside YouTube lack content-level accreditation in that screen . TV-off detection is not mandatory, and OM SDK adoption is not measured by its certified-integration count . The fraud-scheme disclosures below come from detection sellers and publish no error rates .

In short

Of 41 MRC services in the 10 November 2026 screen,18 carry a CTV metric and eight hold CTV SIVT filtration . Five SIVT holders are scored products, while only DoubleVerify and IAS hold certified-app CTV video viewability . Zefr's English-language YouTube in-stream reporting is the sole CTV content-level grant . The preliminary six-product tally of 14 accredited units among 17 requires reconciliation with the full sold-unit grid . MRC's 2021 guidance makes TV-off detection optional and requires unknown treatment for SSAI traffic that cannot be discretely evaluated for IVT . The OM SDK register held 26 active CTV integrations from 12 companies, giving no impression-level adoption share . All 15 schemes below were disclosed by detection sellers, with no error rate in the 14 disclosures opened . The BADBOX family also has government and court records, whose allegations and procedural stages remain separate .

Note on a declared interest

The author works at Samba TV, which competes with Nielsen in TV measurement. DoubleVerify agreed on 6 August 2026 to be acquired by Neptune BidCo US Inc., Nielsen's parent, and no closing was on record on 10 November 2026 (Chapter 7) . DoubleVerify is treated under the same published rules as every other vendor in this paper.

What MRC accredits on CTV

CTV SIVT is the widest-accredited family in this screen, with eight holders, five scored . MRC treats CTV as a separate environment and does not accredit OTT discretely . Google Ads and DV360 are first-party validation, while other holders include ad servers, publishers and measurement services . DoubleVerify estimates CTV at 12% of its 2025 post-campaign measurement . Chapter 2 explains what the letters certify.

DoubleVerify's 13 April 2026 letter covers CTV impressions with SIVT, video viewability, property-level verification in ten languages and specified pre-bid segments . Certified apps limit viewable impressions and Fully On-Screen, while other tested apps may enter other viewability metrics . Pre-bid suitability scope remains narrower than its 2024 release might imply (Chapter 2) .

IAS's letter of 9 July 2026 accredits CTV video tracked ads, impressions, viewability and SIVT filtration for certified applications only, and lists its counts and rates of general invalid traffic (GIVT) and SIVT for desktop, mobile web and in-app only . Its CTV invalid-traffic accreditation is therefore filtration applied to certified-application metrics, first granted in April 2024 . IAS's ad-fraud page drops the certified-applications limit and claims a priority on CTV SIVT that MRC's archived lists contradict (Chapter 2) .

Protected Media's and Pixalate's CTV accreditations cover impressions and invalid traffic, not viewability, and so does HUMAN's . HUMAN holds CTV detection before and after the bid, and no HUMAN CTV viewability product was found in the pages read . Pixalate's letter of 8 April 2025 is still the one MRC links, and Pixalate's documentation says it reports no CTV viewability . Protected by Mediaocean, which MRC still lists as Protected Media, holds CTV accreditation for video impressions only, with GIVT and SIVT metrics, by what its letter calls non-SDK measurement . Pre-bid detection of invalid traffic is accredited on CTV for DoubleVerify and HUMAN only, and MRC does not require pre-bid filtration . Adloox, owned by Peer39 since June 2026, has no CTV row .

In-app accreditation is broader. DoubleVerify, IAS and HUMAN hold impressions with SIVT plus display and video viewability . Pixalate holds impressions and display viewability, and Adloox's in-app metrics are net of GIVT only .

On the large platforms' own CTV inventory, third-party accreditation is thinner (Chapter 5). The verifiers' accredited YouTube viewability covers desktop, mobile web and in-app only, and IAS's accredited reporting for Amazon's demand-side platform excludes Fire TV and Twitch . MRC's platform chart of 22 October 2026 shows third-party viewability neither accredited nor in process on Fire TV, Twitch and Prime Video .

Service on MRC's listCTV metric familyStatus on 10 November 2026Scope limit or noteLetterEvidence
DoubleVerify Quality Analytics PlatformDisplay and video impressionsAccreditedInclusive of SIVT filtration13 Apr 2026
DoubleVerifyGIVT and SIVT filtrationAccreditedTrustworthy Accountability Group (TAG) addendum: IVT 2.0 GIVT and SIVT provisions for CTV metrics13 Apr 2026
DoubleVerifyVideo viewability, Fully On-Screen, completion and quartilesAccreditedViewable impressions and Fully On-Screen from certified apps only; other viewability metrics may include tested, non-certified apps; completion and quartiles need no app certification13 Apr 2026
DoubleVerifyProperty-level ad verificationAccreditedTen CTV languages (Chinese, English, French, German, Hindi, Japanese, Korean, Malay, Spanish, Vietnamese; French added April 2026); DoubleVerify's own page says nine13 Apr 2026
DoubleVerifyPre-bid segmentsAccredited for fraud and IVT, Fully On-Screen, app-store categories and site and app listsPre-bid brand safety, suitability and contextual segments on desktop, mobile web and in-app only13 Apr 2026
DoubleVerifyContent-level brand safety and suitabilityNot accredited, not under reviewNoneNone
Integral Ad ScienceVideo tracked ads, impressions and viewabilityAccreditedCertified applications only; video only; the enclosure has no CTV tracked-ads or Total Net Impressions row; the 2024 letter named five operating systems or devices9 Jul 2026
Integral Ad ScienceGIVT and SIVT filtrationAccredited; CTV SIVT first granted April 2024Certified applications only; counts and rates listed for desktop, mobile web and in-app only; standards paragraph names only IVT 2.0's general provisions, the TAG addendum general and sophisticated9 Jul 2026
Integral Ad ScienceSSAI Noncompliant AdsAccreditedCertified traffic only (list note)9 Jul 2026
Integral Ad ScienceProperty-level ad verificationNo CTV rowDesktop and mobile web on the list; letter basis unresolvedNone
IAS YouTube Brand SafetyContent-level brand safety and suitability, English and Spanish YouTube user-generated videoAudit listed as in process as of April 2023Desktop, mobile web, in-app and CTVNone
HUMANTracked ads and impressions (Ad Fraud Sensor)AccreditedNone stated14 Apr 2026
HUMANGIVT and SIVT detection after the bid; GIVT and SIVT requests before it (Ad Fraud Defense)AccreditedList: SIVT applies to backend detection and IVT predictions; MRC (April 2025): pre-bid suggestions go to platforms, which make the bid decisions14 Apr 2026
HUMANViewabilityNot accredited on CTVGranted on desktop, mobile web and in-app14 Apr 2026
PixalateDisplay and video ad counts and impressionsAccreditedThe letter includes CTV display impressions; the list omits them8 Apr 2025
PixalateGIVT and SIVT filtrationAccreditedStandards: MRC's Minimum Standards and IVT 2.0's GIVT and SIVT provisions; no later Pixalate decision in MRC's quarterly reports, third quarter of 2025 to third quarter of 20268 Apr 2025
PixalateSSAI video metricsAccreditedNone stated8 Apr 2025
PixalateViewabilityNeither accredited nor under review on CTVVideo viewability audit listed as in process as of September 2025, for desktop, mobile web and in-app onlyNone
Protected Media (Protected by Mediaocean)Video impressionsAccreditedNon-SDK measurement; inclusive of SIVT filtration22 Sep 2026
Protected MediaVideo GIVT and SIVT metricsAccreditedNone stated22 Sep 2026
Protected MediaViewabilityNot accredited on CTVNoneNone
Zefr Third-Party Reporting of Google YouTube Content-Level Brand SafetyContent-level brand safety and suitabilityAccredited, initial grantEnglish-language in-stream video and music video; Shorts, YouTube TV, Google TV, Masthead, Google Video Partner videos and volumetric data excluded20 Jan 2026
Adloox (Peer39)Any CTV metricNo CTV rowNoneNone
Google YouTube Brand Safety (first-party, not scored)Content-level brand safety and suitabilityNo CTV on the listDesktop, mobile web and in-app28 May 2026
GumGum Contextual Platform API (listed, not scored)Content-level brand safety, open webRevoked September 2025 on withdrawalEnglish; had covered desktop, mobile web and CTVContinued May 2025

Content-level and program-level suitability

Zefr's 20 January 2026 grant covers English-language YouTube in-stream video and music video on CTV . It excludes Shorts, YouTube TV, Google TV, Masthead, Google Video Partners and volumetric data . Google's first-party row lists no CTV, IAS's YouTube audit remains in process, and GumGum's former open-web grant was revoked on withdrawal in September 2025 .

Vendors sell or describe program-level signals for streaming without that accreditation . On 6 November 2025 DoubleVerify launched Verified Streaming TV, a check of premium streaming-player environments, and Do Not Air automation, which curates streaming inventory by genre and maturity rating and was offered in early access through The Trade Desk . Pixalate's OpenEPG resolves the show in a CTV bid request, and IAS sells CTV Brand Safety for YouTube while its YouTube audit is listed as in process . Channel Factory added genre and mood classification for premium CTV on 28 September 2026 . Peer39 says roughly 60% of CTV bid requests lack usable program-level signal, with no sample, period or method .

Program metadata often comes from the seller: IAB Tech Lab notes that genre travels as free text in OpenRTB bid requests, and DoubleVerify's Certified Transparent Streaming programme takes content data that publishers contribute through clean rooms . Classifying a program from its genre, rating or title differs from the content-level measurement MRC accredits, which no CTV service outside YouTube held on 10 November 2026 .

CTV units in the final grid

The final H1a grid (Chapter 2) records 34 CTV units sold by 14 of the 17 scored products, of which 13 are MRC-accredited and none is in audit . The grid counts one unit per product, metric family and environment, so pre-bid and post-bid delivery of one family on CTV is one unit. YouTube is one of the nine large platforms and has its own units, so YouTube served on a TV screen, such as Zefr's content-level accreditation, is counted under YouTube, not CTV. The stream's earlier six-product working tally (14 of 17) is superseded by this grid .

Server-side insertion and the switched-off screen

SSAI ad-stitching servers send tracking requests on devices' behalf . VAST4.3 requires device IP and user-agent forwarding to distinguish them from fraud, but White Ops says these fields are easy to falsify . MRC's guidance governs their validity treatment below .

MRC's SSAI and OTT guidance of August 2021 makes vendors verify that providers declaring SSAI are SSAI providers, and treat their traffic as unknown rather than valid unless it is discretely measurable for IVT; proxy and data-centre traffic may be filtered only when direct signals show it is invalid . The guidance says CTV may be more vulnerable to ad fraud, partly because measurers' telemetry there may be more limited . It encourages ads.cert, IAB Tech Lab's protocol for signing requests between servers, and proposes an SSAI-provider certification that no MRC or Tech Lab page searched on 10 November 2026 shows established . MRC's IVT 2.0 addendum (June 2020) has SIVT vendors consider in-app and OTT traffic discretely where it is a material share of what they measure, and its April 2024 interim update says non-CTV apps misrepresented as CTV should be considered SIVT .

Pixalate's SSAI video metrics and IAS's count of SSAI Noncompliant Ads on CTV are accredited . Pixalate says it judges SSAI traffic per transaction from more than 30 signals about the SSAI server, and publishes no signal list, threshold or validation result . TAG's Certified Against Fraud 11.0 (July 2026) dropped its CTV SSAI billing-notification requirement .

MRC's 2021 guidance does not require TV-off detection, but requires removal of confirmed TV-off events and disclosure of the limitation . Unflagged plays on a switched-off screen can remain in an accredited count . DoubleVerify's certified-app Fully On-Screen metric adds its TV-on condition . Its 2024 estimate of more than one in three impressions in TV-off-capable environments had no published method or base .

Open Measurement on CTV and in-app

On CTV, what is public about the OM SDK is its releases and its certified integrations . The kit lets a measurement provider's script collect signals inside an app or player that integrates it . Version 1.4 (2022) first supported CTV, on Android TV and Apple's tvOS, with a signal for a display that is off . Version 1.5 (announced 30 May 2024) added Samsung and LG smart-TV web apps, and device attestation for Fire TV and Apple devices, announced on 23 October 2025, reached the JavaScript client as 1.6.0 on 29 October . On 10 November 2026 IAB Tech Lab announced impression measurement for pause, menu, screensaver and overlay ads on CTV, naming no version . No Tech Lab page opened names Roku or Vizio support .

On 10 November 2026 IAB Tech Lab's compliance register listed 341 active certified integrations from 134 companies, plus 125 expired records; most active records are for Android and iOS apps, and 26 active records from 12 companies carried a CTV label . The records are mostly SDKs, players and ad-insertion services, with some individual apps . Thirty-two active records were last certified before 2023, and DIRECTV's CTV-Android record, which a May 2024 Tech Lab release named, shows as expired . That release's unsourced claim that the OM SDK covered 40% of the CTV market measures platform support, not adoption . Companies can hold records on several platforms, so the rows below do not sum to 134 .

Platform labelActive recordsCompaniesv1.2v1.3v1.4v1.5v1.6Evidence
Android145127417332764
iOS136120319183066
Web34300341710
CTV-Android111000362
CTV-iOS (tvOS)10900235
CTV-Web (Samsung, LG)4400022
CTV, platform not specified1100001
All active3411347396085150
All active CTV26120051110

MRC's letters tie OM SDK measurement to DoubleVerify's non-CTV viewability and IAS's in-app measurement, with MRAID metrics excluded . DoubleVerify separately claimed CTV OM SDK support in December 2025 . TAG's 2026 rules require verifiers under the device-attestation guidance to use 1.6 or higher, while impression-level adoption remains unreported in the screened sources .

The answers to device spoofing are proprietary or unmeasured . App-ads.txt extends authorised-seller files to CTV app stores but authenticates no device, and IAB Tech Lab's CTV Programmatic Guide, released for public comment in December 2025, calls device information in bid requests a text string that is easily manipulated . Roku's Advertising Watermark is designed to let only genuine Roku devices display authenticated ads, and OM SDK attestation covers Apple devices and Fire TV . The ads.cert protocol for authenticating devices is still marked future, its repository's last commit is dated 8 March 2023, and no adoption figure was found in the sources opened . IAB Tech Lab's release names DoubleVerify, HUMAN and IAS, which sell detection, among the companies in the group behind attestation . No source opened reports how much CTV inventory carries any of these signals .

Schemes and who found them

Each scheme in the register is its finder's finding, and every finder sold the detection it publicised at the time . Of the 15 schemes, disclosed from 2020 to 2026, White Ops and HUMAN named nine and DoubleVerify four; Oracle, then owner of the Moat verification service, and Pixalate named one each . Every figure in the table is the finder's, in its own unit, and the units cannot be added .

SchemeFinder, and what it soldDisclosedEnvironment and techniqueScale as the finder stated itEvidence
MonarchPixalate (IVT analytics and blocking)25 Mar 2020CTV (Roku apps): app spoofing and falsified measurement eventsHundreds of thousands of IVT impressions a day for over six months across over a dozen apps, on "conservative estimates"; asserts no culpability
ICEBUCKETWhite Ops, now HUMAN (bot mitigation and pre-bid blocking)16 Apr 2020CTV and mobile SSAI: data-centre traffic with forged device detailsAbout 1.9 billion ad requests a day at the January 2020 peak; nearly 28% of the programmatic CTV traffic White Ops could see and 66% of the programmatic CTV-related SSAI traffic it protected that month; more than 2 million people impersonated
StreamScamOracle (Moat verification, retired 30 September 2024)17 Dec 2020CTV SSAI: servers sent forged impression eventsMore than 28.8 million US household IP addresses; about 3,600 apps; 3,400 CTV device models
CopyCatzWhite Ops (bot and ad-fraud detection)Jan 2021Mobile in-app (Android): copycat apps164 apps; more than 10 million downloads
OctoBot, with MultiTerra and SneakyTerraDoubleVerify (fraud protection)20 Apr 2021CTV: seven related spoofing variants since November 2019MultiTerra about $1 million and SneakyTerra about $5 million a month, method not stated; not re-opened in the verifier pass
PARETOHUMAN (ad-fraud detection)21 Apr 2021CTV, spoofed from infected Android devicesNearly 1 million infected mobile Android devices; more than 6,000 CTV apps spoofed; an average 650 million ad requests a day
SmokeScreenDoubleVerify (fraud protection)12 Aug 2021CTV screensaver apps, including with the TV offAbout 10,000 devices and up to 10 million requests a day; over $6 million a month by pricing ad requests at $20 a thousand; not re-opened in the verifier pass
LeoTerra variantDoubleVerify (fraud protection)24 Aug 2022CTV SSAI: spoofed device lists, including connected appliancesUp to $10 million in 2022; more than 92 million devices spoofed in the first half; not re-opened in the verifier pass
ScyllaHUMAN (ad-fraud detection)2022, restated Jan 2023Mobile in-app (9 iOS and 80 Android apps)89 apps; more than 13 million downloads; known only from the VASTFLUX release
VASTFLUXHUMAN (ad-fraud detection)19 Jan 2023Mobile in-app, mainly iOS: stacked, invisible video adsMore than 1,700 apps spoofed; nearly 11 million devices; a peak of 12 billion ad requests a day
BADBOX and PEACHPITHUMAN (ad-fraud detection)4 Oct 2023Backdoored CTV boxes, phones and tablets; hidden adsMore than 74,000 infected devices; PEACHPIT's 39 apps averaging 4 billion ad requests a day
CycloneBotDoubleVerify, with Roku (fraud protection)28 Feb 2024CTV: bots spoofing sessions and devicesCapable of up to 250 million falsified ad requests and about 1.5 million spoofed devices a day; up to $7.5 million a month on industry price estimates
KonfetyHUMAN (ad-fraud detection)16 Jul 2024Mobile in-app (Android): evil-twin appsMore than 250 decoy apps; a peak of 10 billion bid requests a day
BADBOX 2.0HUMAN, with Google, Trend Micro and Shadowserver (ad-fraud detection)5 Mar 2025Backdoored CTV boxes and other Android devicesMore than 1 million infected devices; called by HUMAN the largest CTV botnet ever disclosed
NewsJunkieHUMAN (ad-fraud detection)7 Jul 2026CTV: device spoofingHundreds of millions to nearly 2 billion invalid CTV bid requests a day per seller at peak; number of sellers and period not given

None of the 14 disclosures opened reports a miss rate, a false-positive rate or precision, and 12 say the finder's customers were protected or its product detected the scheme . The research was done by AI agents under the author's direction; in the verifier pass a separate AI agent, told to refute each finding, re-opened ten of the 14 (Chapter 13) . The OctoBot, SmokeScreen and LeoTerra figures rest on the research agents' reads alone .

Several schemes attacked measurement itself . Oracle said StreamScam's servers sent Moat and advertisers impression events with forged IP addresses, app IDs and device IDs, and delivered no ad . ICEBUCKET hid bots in the limited signal of SSAI traffic, and VASTFLUX stacked invisible video ads where iOS apps give verification partners little signal .

Scale figures travel beyond their base. ICEBUCKET's 28% was a share of the programmatic CTV traffic White Ops could see, but the briefing TAG co-hosted on 30 April 2020 captioned it as global CTV traffic affected . Roku said PARETO affected less than 0.1% of its devices and that CycloneBot did not affect its platform .

Only the BADBOX family has government and court records, and they draw on the finders . A Federal Bureau of Investigation (FBI) announcement of 5 June 2025 counts millions of infected devices without mentioning advertising . Google's complaint in Google LLC v. Does 1-25 (Southern District of New York, 1:25-cv-04503) alleges more than ten million infected devices as of April 2025, against HUMAN's count of more than one million in March 2025 . A complaint is an allegation, and the preliminary injunction of 27 June 2025 is not a judgment . Google's statement that it refunds advertisers for fraudulent traffic describes first-party validation (Chapter 4) .

DoubleVerify reports 140% more detected CTV schemes and variants in Q1 2026 than Q1 2025, without raw counts or a detection-validation method . The period is stated, so the figure is a vendor trend within that period . None of this chapter's screened sources independently measures a verifier's CTV error rate . Chapter 3 covers prevalence comparisons.

Standards and specifications, 2019 to 2026

Each row is dated from the issuer's own document as read on 10 November 2026. A guidance date is not an accreditation date, and a release is not deployment .

DateIssuerDocumentWhat it says for CTV and in-app verificationEvidence
March 2019IAB Tech Labapp-ads.txt (ads.txt 1.0.2)Authorised-seller files for apps, including connected-TV app stores
June 2020MRCIVT 2.0 addendum, section 4.2.2In-app and OTT traffic considered discretely where material; OTT proxy and data-centre traffic filtered only when known to be invalid
March 2021IAB Tech Labads.txt 1.0.3INVENTORYDOMAIN variable for inventory shared between sellers
August 2021MRCServer-Side Ad Insertion and OTT GuidanceCTV accredited discretely, OTT not; TV-off detection not required; unverifiable SSAI traffic unknown; ads.cert encouraged
2022IAB Tech LabOM SDK 1.4First CTV release: Android TV and tvOS; display-connection status
August 2022IAB Tech Labads.txt 1.1OWNERDOMAIN and MANAGERDOMAIN variables
December 2022IAB Tech LabVAST 4.3Device IP address and user agent sent with ad-stitching tracking requests
14 February 2024IAB Tech Labads.cert 2.0 page, last updatedAuthenticated Connections ready for adoption; Authenticated Devices future
24 April 2024MRC2024 IVT Interim UpdatesCTV bundle IDs within app misrepresentation; non-CTV apps posing as CTV are SIVT; app-ads.txt telemetry encouraged
30 May 2024IAB Tech LabOM SDK 1.5 announcedSamsung Tizen and LG webOS through the Web Video SDK
2024IAB Tech LabVAST CTV Addendum 2024AdVerification and UniversalAdId fields extended to VAST 2 and 3
11 April 2025MRCStatement on pre-bid IVTPost-serve GIVT filtration required on every impression; pre-bid filtration not required
23 and 29 October 2025IAB Tech LabDevice attestation announced; JavaScript client 1.6.0Privacy Pass attestation for Apple devices and Fire TV
11 December 2025IAB Tech LabCTV Programmatic Guide, for public comment to 31 January 2026; page modified 30 January 2026Use cases for OM SDK, app-ads.txt, ads.cert, device attestation and content taxonomy
July 2026TAGCertified Against Fraud 11.0CTV SSAI billing-notification requirement removed; verifiers under its device-attestation guidance must implement OM SDK 1.6 or higher
2 November 2026IAB Tech LabOM SDK JavaScript client 1.6.12Version alignment; no change from 1.6.11
10 November 2026IAB Tech LabOM SDK support for emerging CTV ad formatsImpression measurement for pause, menu, screensaver and overlay ads; no version named

Chapter 7

Who pays the verifiers, and who owns them

Advertisers pay most of the verification bill, and sellers pay part of it. In FY2024, the last fiscal year both verifiers filed, DoubleVerify's Supply-side customers brought it 8.6% of its revenue, and the Publisher channel of Integral Ad Science (IAS), which includes Publica's connected-TV (CTV) ad serving, brought IAS 14.4%. Within eight months both companies changed hands or agreed to: Novacap's funds took IAS private on 23 December 2025, and DoubleVerify agreed on 6 August 2026 to be acquired by Neptune BidCo US Inc., Nielsen's parent, a sale still pending on 10 November 2026. H5, the prediction that no verifier with seller revenue or an interested owner publishes a safeguard, is contradicted, but narrowly, by one DoubleVerify page that describes a structural separation for its publisher clients and says nothing of the pending owner. Ownership is recorded as a governance attribute, never as evidence that a measurement is biased.

In short

In FY2024, DoubleVerify's Supply-side customers paid $56.8 million of $656.8 million revenue (8.6%), while IAS's Publisher channel, including Publica's CTV ad serving, brought $76.5 million of $530.1 million (14.4%) . IAS's last periodic filing at the 10 November 2026 check was its 10-Q for the quarter to 30 September 2025 . DoubleVerify's measured transactions rose from about 4.5 trillion in 2021 to 9.5 trillion in 2025, while its average fee fell 4% in FY2024 and 3% in FY2025, excluding one introductory deal . Platforms collect or facilitate substantial revenue for both vendors, on different measures . IAS closed its take-private on 23 December 2025; DoubleVerify's agreed sale to Neptune BidCo US Inc., Nielsen's parent, remained pending on 10 November 2026 . H5 is contradicted on DoubleVerify's June 2024 separation statement for publisher clients, although its scope, responsible owner and enforcement are unspecified and it does not address the pending owner . A stricter policy requirement would yield a different reading within the sources screened, as set out below .

Advertisers pay most of the bill

Advertisers and their agencies pay most of what the two verifiers with public filings earn; sellers pay a steady minority share. DoubleVerify files revenue by three customer types: Activation, from advertisers that evaluate inventory before buying; Measurement, from advertisers that measure quality and performance afterwards; and Supply-side, from publishers, retail media networks, ad networks and exchanges . In FY2025 the three brought $427.3 million, $249.7 million and $71.3 million of $748.3 million . Its annual report, the 10-K, rounds the advertiser share to 90%, against 91% in FY2024 . Supply-side customers sign monthly or annual contracts with minimum guarantees and tiered pricing above them .

IAS files revenue by channel: Optimization, Measurement and Publisher, which in FY2024 brought $242.6 million, $211.0 million and $76.5 million of $530.1 million; 86% of revenue came from advertiser customers . The Publisher channel mixes tools sold to publishers with the CTV ad serving of Publica, which IAS bought in August 2021, and IAS credited most of the channel's 2024 growth, $12.7 million or 20%, to Publica . Neither company splits its seller-side line into verification and other products, so the ad-serving part of IAS's line is unknown .

Seller-side line, as filedFY2022FY2023FY2024FY2025Evidence
DoubleVerify, Supply-side (customer type)$43.3m of $452.4m (9.6%)$45.6m of $572.5m (8.0%)$56.8m of $656.8m (8.6%)$71.3m of $748.3m (9.5%)
IAS, Publisher (channel; includes Publica's CTV ad serving)$62.8m of $408.3m (15.4%)$63.8m of $474.4m (13.5%)$76.5m of $530.1m (14.4%)Not filed

The comparison uses FY2024, the last year both companies filed, while keeping the customer-type and channel definitions apart . DoubleVerify's Supply-side share was 9.8% in the nine months to 30 September 2026, $56.8 million of $580.3 million . IAS's last filed period was the nine months to 30 September 2025, with Publisher revenue of $68.7 million of $437.6 million . Its FY2025 amount remained unfiled at the 10 November check and is unresolved under H5 . The figure plots each line against its own company's revenue.

SELLER-SIDE REVENUE AS A SHARE OF TOTAL REVENUE, AS FILED0%5%10%15%20%9.6%15.4%FY20228%13.5%FY20238.6%14.4%FY2024DoubleVerify, Supply-sideIAS, Publisher (includes Publica CTV ad serving)DoubleVerify FY2025: 9.5%; IAS filed no FY2025 figures
Both verifiers earn from sellers: about a tenth of DoubleVerify’s revenue, about a seventh of IAS’s. DoubleVerify’s Supply-side customer type and IAS’s Publisher channel as a share of total revenue in the 10-K filings, FY2022 to FY2024 (DoubleVerify FY2025: 9.5%). The two categories are defined differently: IAS’s Publisher line includes Publica’s CTV ad serving, and neither company says how much of the line is paid by platforms rather than publishers.Units: percent of total revenue · Scope: DoubleVerify and IAS 10-K filings, fiscal years as filed · Sources A revenue category is not a map of who pays for each verdict, and existence of seller revenue is not evidence that a verdict is biased (hypothesis H5).Data: data/figure-data.json#seller_revenue

The filed lines do not identify the payer for each verdict. Agencies pay DoubleVerify for multiple clients and platforms collect advertiser fees . DoubleVerify also sells platform-wide measurement subscriptions to supply-side partners . IAS's terms allow a media partner to pay an advertiser's verification fees, without disclosing the amount billed that way . Seller revenue alone establishes no measurement bias.

Volume doubled while the fee per thousand fell

DoubleVerify bills advertisers per thousand Media Transactions Measured, mostly at fixed prices, with caps where platforms charge a share of spend . The approximate volume rose from 4.5 trillion in 2021 to 5.5 trillion in 2022, 7.0 trillion in 2023, 8.3 trillion in 2024 and 9.5 trillion in 2025 . This company-defined operating metric counts measurements before and after purchase, rather than unique impressions or impressions classified as valid; the 2025 mix was an estimated 44% display and 56% video .

Average fees fell 4% in FY2024 and 3% in FY2025 while measured volume rose 19% and 15%, excluding an introductory fixed-fee deal with one unnamed customer . In the nine months to September 2026, fees fell 4% and volume rose 11%; the filings report changes without giving an average dollar fee .

DOUBLEVERIFY MEDIA TRANSACTIONS MEASURED, TRILLIONS A YEAR0T2T4T6T8T20214.5T20225.5T20237T20248.3T20259.5TThe measured-transaction fee per thousand fell 4% in FY2024 (volume +19%) and 3% in FY2025 (volume +15%), excluding an introductory-pricingeffect the filings name.
DoubleVerify measures twice the volume it measured in 2021 and charges less for each thousand. Media transactions measured by DoubleVerify, as reported in its 10-K filings, 2021 to 2025, with the change in its measured-transaction fee per thousand. The price of a verdict falls as its volume rises: the fee is set per thousand impressions measured, not by what the verdict changes.Units: trillions of media transactions measured per year · Scope: DoubleVerify, worldwide, as filed · Sources A media transaction is DoubleVerify’s unit (an impression measured, pre-bid or post-bid); IAS publishes no comparable annual count.Data: data/figure-data.json#dv_volume

DoubleVerify's partner-pay rate card, effective 21 May 2026 and applying only where a partner remits its fees, lists $0.135 per thousand impressions for DV Authentic Ad and $0.400 for DV Social Pre-Bid Controls; no direct-contract price list was found .

IAS's FY2024 10-K gives no comparable annual measurement count. Its more than 280 billion daily interactions and Publica's more than 10 billion monthly CTV ads use different units and activities . IAS's measurement impressions rose 23% with average costs per thousand (CPMs) down 7%; optimization impressions rose 4% with CPMs up 2% . Revenue follows measured volume and CPMs, with minimums, tiered overages or flat fees in some contracts . The two vendors' volume series cannot be added.

A few platforms collect much of the money

Customer concentration and platform dependence use different measures. No DoubleVerify customer exceeded 10% of revenue in 2023–2025, while IAS's 10-K gives only an accounts-receivable concentration test . Two unnamed platforms collected 24% and 12% of DoubleVerify's FY2025 revenue; DSP partnerships facilitated 46% of IAS's FY2024 revenue, the largest one 18% . These are collected and facilitated revenue, respectively.

Money also flows back. IAS pays DSPs, including Google and The Trade Desk, a share of the revenue it receives through its products, and says the platforms it integrates with produce no material revenue for it directly . The platforms that collect the fees are also where pre-bid segments are sold: a dependency, not a finding of conflict .

Large customers are counted in different units. DoubleVerify had 131 customers worth at least $1 million each in 2025, against 93 in 2023 . IAS had 237 advertisers spending at least $200,000 a year in 2024, and they produced 85% of its advertising revenue .

Who owns the verifiers now

The dated record includes acquisitions, financings, rebrands and retirements . DoubleVerify completed its initial public offering on 23 April 2021 at $27 a share; IAS closed its offering on 2 July 2021 at $18 . Within eight months, IAS closed a take-private and DoubleVerify agreed to one .

OWNERSHIP EVENTS IN AD VERIFICATION, 2021 TO 2026202120222023202420252026DoubleVerifyIASSpecialistsRetiredMeetrics boughtOpenSlate boughtSale to Nielsen's parent agreedStockholders approvePublica (CTV ad serving) boughtNovacap take-private agreedTaken privateFlashtalking buys Protected MediaWhite Ops renamed HUMANHUMAN merges with PerimeterXScope3 buys AdlooxPeer39 buys AdlooxOracle retires Moatannouncedcompletedrebrandlaunchretired
Both independent verifiers changed hands or agreed to within eight months. Dated ownership events from the companies’ filings and releases. IAS was taken private on 23 December 2025; DoubleVerify’s sale to Neptune BidCo US Inc., Nielsen’s parent, was agreed on 6 August 2026 and approved by stockholders on 29 October 2026, and had not closed when last checked (10 November 2026). The full dated table, with value bases, is in the data downloads.Units: dated events · Scope: verification companies and products, 2021 to 10 November 2026 · Sources Pending: DoubleVerify’s closing is rechecked at the freeze. A headline deal value is shown only with the basis its source states.Data: data/figure-data.json#consolidation

Q3, the paper's question on consolidation, is answered under a coding rule: an event is completed only when a filing or the parties' release records completion, and a pending transaction is dated to the latest primary record read. Acquisitions, mergers, financings, rebrands and retirements are different events, and none shows that a product or metric failed . The table lists the events other than IAS's take-private and DoubleVerify's sale, which follow.

DateCompanyEventClassValue as statedEvidence
23 Feb 2021Protected MediaFlashtalking announces it has acquired the companyAcquisition (completed per release)Not stated
30 Mar 2021White OpsRenamed HUMAN, three months after Goldman Sachs Merchant Banking Division, ClearSky Security and NightDragon bought itRebrandn/a
13 Jul 2021FlashtalkingMediaocean agrees to acquire it; no completion date found; part of Mediaocean by 12 May 2022Acquisition (agreement)Not disclosed
9 Aug 2021PublicaAcquired by IAS (CTV ad serving)Acquisition (completed)$171.4m cash plus shares valued at $49.6m
31 Aug 2021MeetricsAcquired by DoubleVerify (ad verification)Acquisition (completed)$24.3m net cash
2 Sep 2021PixalateGrowth capital from Western Technology Investment and Javelin Venture PartnersFinancing$18.1m ($22.7m in total)
22 Nov 2021OpenSlateAcquired by DoubleVerify (contextual targeting); agreement announced 9 Nov 2021Acquisition (completed)$148.2m net of cash, final
31 Dec 2021Context (Nobora SAS)Acquired by IAS (contextual classification)Acquisition (completed)$22.6m cash plus shares valued at $10.4m
2021 (day not stated)Amino PaymentsAcquired by IASAcquisition (completed)Not stated
4 Jan 2022HUMANGrowth round led by WestCapFinancing$100m raised
20 Jul 2022Adverif.aiAcquired by Zefr (misinformation detection)AcquisitionNot stated
27 Jul 2022HUMAN and PerimeterXMerger under the HUMAN name announced; approvals received; completion not statedMergerNot disclosed
14 Aug 2023ScibidsAcquired by DoubleVerify (campaign optimization)Acquisition (completed)$121.4m initially reported; $121.1m net of cash as adjusted in the FY2025 10-K
21 Nov 2023Kubient (KAI)Form 25 removes its securities from listing; a current report of 8 May 2024 announces job cuts tied to a potential wind-downDelistingn/a
30 Sep 2024Oracle MoatRetired with every Oracle Advertising product, after notices in the week of 17 Jun 2024Product retirementn/a
9 Oct 2024HUMANGrowth capital led by WestCapFinancing$50m-plus raised
4 Nov 2024AdlooxScope3 announces the acquisitionAcquisition (announced as done)Not disclosed
21 Nov 2024InnovidMediaocean, owner of Protected Media, agrees to acquire itAcquisition (agreement)$3.15 a share; no aggregate stated
13 Feb 2025InnovidMerger completed; combined with FlashtalkingAcquisition (completed)$3.15 a share
13 Mar 2025RockerboxAcquired by DoubleVerify (marketing attribution)Acquisition (completed)$82.3m net of cash
24 Apr 2025Channel FactoryTruelink Capital completes a significant investment; Star Mountain exits through it (announced 5 Jun 2025) and stays a minority ownerInvestmentNot disclosed
2 Jun 2026AdlooxPeer39 announces it has bought Adloox from Scope3; completion recorded in a UK filingAcquisition (completed)Not disclosed
23 Sep 2026Scope3Announces it is becoming Apostra; Adloox's owner unchangedRebrandn/a

DoubleVerify's filings call only one of its four purchases since 2021, Meetrics, an ad verification provider . OpenSlate was a pre-campaign contextual targeting platform which, by DoubleVerify's own announcement, helped brands match ads to suitable or relevant content in social video and CTV; that announcement valued the deal at $150 million, and the final price was $148.2 million net of cash . Neither brand survives: on 10 November 2026 openslate.com redirected to doubleverify.com and meetrics.com did not resolve .

IAS: agreed, consented, closed

IAS agreed on 24 September 2025 to merge with a subsidiary of Igloo Group Parent, Inc., an affiliate of funds managed by Novacap, for $10.30 a share . Vista and Atlas funds holding about 52.6% consented that day, so the deal used a Schedule 14C information statement . The funding commitments were up to $857.8 million in equity, a $1.0 billion term loan and a $150 million revolver, with termination fees of $52.5 million for IAS and $100 million for the buyer .

IAS became wholly owned by Igloo Group Parent on 23 December 2025 . The closing 8-K reports about $1.6 billion paid for outstanding shares, excluding its defined Owned Company Shares and Appraisal Shares . The parties' earlier $1.9 billion valuation states no basis, and the 19 December event on the 8-K cover concerns retention agreements .

DateStepRecordEvidence
24 Sep 2025Merger agreement at $10.30 a share; written consent of holders of about 52.6% of the voting power8-K; joint release
24 Oct 2025Hart-Scott-Rodino (HSR) antitrust notifications filed; waiting period scheduled to end on 24 Nov 2025 (actual expiry not documented in the sources opened)Information statement
7 Nov 2025Schedule 14C information statement dated and first mailedSchedule 14C
23 Dec 2025Merger completed, about $1.6 billion paid for the outstanding shares; Nasdaq files Form 25Closing 8-K; Form 25
2 Jan 2026Form 15 filed, certifying one holder of recordForm 15

DoubleVerify: agreed, cleared, approved, not closed

Note on a declared interest

The author works at Samba TV, which competes with Nielsen in TV measurement. Neptune BidCo US Inc., the parent of Nielsen, is DoubleVerify's pending buyer. DoubleVerify is treated under the same published rules as every other vendor in this paper.

On 6 August 2026 DoubleVerify agreed to become a subsidiary of Neptune BidCo US Inc. at $13.60 a share in cash . The definitive proxy identifies Neptune as Nielsen Company (US), LLC's parent and an affiliate of Elliott- and Brookfield-managed funds . The table separates the completed approvals from the uncompleted merger at the 10 November check .

The two value figures rest on different bases: an enterprise value of about $2.15 billion in the joint release, an implied equity value of about $2.332 billion in the proxy . Elliott's affiliates committed $200 million of equity and lenders about $1.8 billion of debt; the termination fees are $60 million for DoubleVerify and $144 million for the parent; and Providence's funds, holding about 11.8%, agreed to vote in favour .

DateStepRecordEvidence
6 Aug 2026Merger agreement with Neptune BidCo US Inc. and Wallace Merger Sub Inc.; the joint release forecasts closing by the end of the fourth quarter of 20268-K, Item 1.01 (filed 10 Aug); joint release
20 Aug 2026HSR notification filedDefinitive proxy
21 Sep 2026HSR waiting period expired at 11:59 p.m. ET; Australian notification lodgedDefinitive proxy; 8-K of 20 Oct
29 Sep 2026Definitive proxy; targets closing by the first quarter of 2027 (forecast)Schedule 14A
5 to 13 Oct 2026Clearances in Germany (5 Oct), Cyprus (7 Oct), Australia, by the Australian Competition and Consumer Commission (ACCC), and Morocco (13 Oct)8-K of 20 Oct; ACCC register
20 Oct 2026All four foreign reviews reported complete, subject to the 14-day window for applications to the Australian Competition Tribunal8-K, Item 8.01
27 Oct 2026Window closed (13 Oct plus 14 calendar days); no application on the Tribunal's list read on 10 NovDerived; Tribunal list
29 Oct 2026Stockholders adopt the agreement: 118,218,914 for, 7,592,263 against, 214,454 abstaining8-K, Item 5.07
30 Oct 2026Continues to expect closing by the first quarter of 2027 (forecast); guidance withdrawnResults release
9 Nov 2026Expects the parties to be in a position to close during the fourth quarter of 2026, subject to the remaining conditions (forecast)8-K, Item 7.01
10 Nov 2026No 8-K under Item 2.01, Form 25, Form 15 or completion release: pendingEDGAR; investor feed
6 May 2027Outside date, extended to 6 Aug 2027 only if every non-regulatory condition is satisfied by then8-K, Item 1.01

The expected closing moved twice, from the end of 2026 to the first quarter of 2027 and back to the fourth quarter of 2026, and every statement of it is a forecast . DoubleVerify withdrew its guidance and suspended earnings calls while the deal is pending . On 10 November 2026, last checked at about 12:44 UTC, EDGAR held no closing 8-K, Form 25 or Form 15, and the investor feed no completion release, so DoubleVerify remained a listed company not owned by Nielsen's parent . A closing before the freeze on 11 December 2026 will be dated from its own record. Stockholder complaints over the merger belong to Chapter 9 .

Specialists sold, merged, renamed or retired

Protected by Mediaocean, formerly Protected Media, sits inside an ad-serving group: Flashtalking bought it in 2021, Mediaocean agreed that July to buy Flashtalking, and no completion date for that purchase, or date for the rename, was found . Mediaocean announced in May 2022 that verification was fully integrated into Flashtalking's ad serving, and in February 2025 it completed its purchase of another ad server, Innovid . Innovid's merger proxy of January 2025 calls Mediaocean a portfolio company of funds managed by CVC Capital Partners and of TA Associates and Charlesbank Capital Partners, wording that does not state control .

HUMAN, formerly White Ops, renamed itself on 30 March 2021, three months after Goldman Sachs Merchant Banking Division, ClearSky Security and NightDragon bought it . It then raised two rounds led by WestCap and announced a merger with PerimeterX on 27 July 2022, with approvals received and no completion date stated . In a court filing docketed on 26 June 2026 HUMAN named Singularity Buyer LLC and Singularity TopCo LLC as its corporate parents; no source opened says who controls them .

Adloox passed from Scope3 to Peer39 in June 2026, after Scope3's November 2024 acquisition . Adloox Limited's September 2026 accounts name Peer39 Tech LLC as ultimate controlling party from June . MRC's 29 June letter continued the renamed service's post-serve accreditation, with the full metric and environment scope in Chapter 2 . Scope3's later rename to Apostra leaves Adloox's ownership unchanged . Peer39 describes itself as independent and employee owned, but the sources screened to 10 November do not settle its group ownership; the UK subsidiary record covers that entity only .

Oracle retired Moat instead of selling it: its end-of-life notice of 12 August 2024 ended every Oracle Advertising product and its support, Moat Analytics and Moat Reach among them, from 30 September 2024 . No sale was found in the notice or in the one search logged, and a retirement does not show that Moat's measurement failed . Pixalate, Zefr and Channel Factory recorded smaller events, listed in the table; Pixalate told a court in April 2025 that it has no corporate parent, and none of the three states who controls it .

H5 is contradicted, narrowly

H5 fails on one page, and only narrowly. It has two parts. The first says both public verifiers' last filings show seller-side revenue, and it holds . The second says no scored product with seller-side revenue, or with an owner that has an interest in the inventory or the currency it measures, publishes a conflict or separation policy covering that interest. The registered rule calls H5 contradicted if "at least one such policy is published by a scored product or its owner", and supported if "no published conflict or separation policy covering such an interest is found in the sources screened".

On 13 June 2024 DoubleVerify published a page on measuring publishers impartially. It says its media quality assessment teams work independently of its business development and client relations teams, calls that a structural separation that keeps its evaluations free of commercial influence, says it does not alter its methods or soften its assessments because of a publisher's status or relationship with DoubleVerify, and lets publishers raise site-classification concerns directly . Read as the rule is written, that is a published separation statement covering DoubleVerify's seller-side interest in its publisher clients, and the paper codes H5 as contradicted on it .

The separation page names no responsible policy owner or enforcement mechanism and does not address Neptune's prospective interests . Its general audit assertion identifies no report tied to the separation . DoubleVerify's Code of Ethics concerns personal conflicts and its 10-K discusses Providence's corporate opportunities; neither supplies the missing governance record . Its Impact Report's claim of no inherent conflict is vendor positioning alongside seller revenue . Requiring a policy with scope and enforcement would instead support H5 within the products screened .

The merger release promises independent operating structures without a mechanism or duration, while also describing an integrated platform . The definitive-proxy search found no conflict, firewall or separation policy . The declared Samba–Nielsen competition interest applies here. This records governance disclosure and establishes no bias in DoubleVerify's measurements.

For the other vendors screened, no conflict or separation policy was found . The table sets out what each discloses.

VendorSeller-side or platform customersSeller-side amountOwner and its interestsSafeguard foundEvidence
DoubleVerifySupply-side customer type: publishers, retail media networks, ad networks, exchanges$71.3m of $748.3m, FY2025 (9.5%)Listed; sale to Neptune BidCo US Inc., parent of the Nielsen Company (US), LLC, pending on 10 Nov 2026Separation page of 13 Jun 2024 (publisher clients; no scope, owner or enforcement); nothing on the pending owner
IASPublisher channel, including Publica's CTV ad serving; media partners can pay advertisers' fees$76.5m of $530.1m, FY2024 (14.4%); FY2025 not filedNovacap's funds since 23 Dec 2025; the portfolio includes Static Media (digital media) and Cadent (TV advertising)None found
Protected by MediaoceanSupply-side platforms (SSPs), DSPs, exchanges and publishers among its customers in 2021Not filedMediaocean, which also runs the Flashtalking and Innovid ad serversNone found
AdlooxPublishers (ad-server segments; Brut since 2023)Not filedPeer39, which sells pre-bid controls and publisher categorisationNone found
HUMANDSPs, SSPs and exchanges (platform suite announced 5 Oct 2026); Magnite since 2015Not filedPrivate; parents Singularity Buyer LLC and Singularity TopCo LLC; control not statedNone found
PixalateSSPs and publishers among its customersNot filedPrivate; no corporate parent (April 2025)Method rules only
ZefrNot established; an IRIS.TV revenue share reported in 2020Not filedPrivate; control not statedNone found
Channel Factory (ViewIQ)Not established; relationships with creators and channel ownersNot filedTruelink Capital investment (2025); Star Mountain a minority ownerNone found

IAS's supply-side business includes publisher tools, Publica, Quality Connect, pre-bid segment resale and Media Partner Fees . Novacap also lists Static Media and Cadent, but whether IAS measures their inventory is undocumented . Searches of IAS's site and filings on 10 November found no policy covering these interests .

Mediaocean owns two ad servers and the verifier whose data it integrates into ad serving, and frames its independence as freedom from media ownership . Peer39 sells pre-bid controls and says its post-buy reporting exists to confirm that those controls worked, with no separate fee for it . Zefr buys YouTube media for advertisers and in 2023 supplied video-level signals to TikTok's first-party inventory filter . Channel Factory builds suitability lists and is paid to buy media against them, an inference labelled as such in the evidence . Pixalate's published safeguards are method rules: its Seller Trust Index ranks on buy-side data only, its terms bar using a client's first-party data to rank that client, and its clawback labels cannot be changed by hand; none addresses Pixalate measuring a seller client's own traffic . For the private vendors the amounts are unresolved under the rule, since no disclosure means unknown .

These absences rest on the pages, filings and searches logged on 9 and 10 November 2026; vendors were not contacted, and a policy kept inside order forms or customer documentation would not be visible . The screen also missed some registered candidates. Among those not screened for a safeguard are Proximic by Comscore, illumaSentinel, mFilterIt, Fraudlogix, which sells to bidding platforms, and Escalated, so for them the absence is not established . That gap cannot reverse the verdict as coded, which one published policy settles; under the strict reading it would limit any support to the products screened.

Interest disclosure is scored in Chapter 11. Ownership itself earns or loses no points .

Chapter 8

After GARM

The brand-safety floor and suitability tiers of GARM, the Global Alliance for Responsible Media, outlived it. The World Federation of Advertisers (WFA) announced on 9 August 2024 that it was discontinuing GARM, the framework's last edition is dated June 2022, and the documents that still use a floor, from the Trustworthy Accountability Group's (TAG) certification guidelines to vendors' and Google's pages, point back to that edition, to undated copies of it or to older frameworks. One screened body publishes a later framework with both a floor and tiers: ISBA, the UK advertisers' trade body, whose Responsible Media Guide of March 2025 keeps ten of GARM's twelve floor categories under three risk tiers without naming GARM. On the rule's wording that contradicts H6, the hypothesis that no maintained successor exists, and the coding rests on one reading: that ISBA's pledge of regular review, which comes with no published change log or procedure, is the "stated change process" the rule asks for.

In short

The WFA discontinued GARM in August 2024 and said in July 2026 that it would not restart it . Its framework page's last update is June 2022; BSI's hosted copy has twelve floor categories and risk tiers but no date or attribution . TAG maintains a certification floor without tiers, while IAB Tech Lab maintains category labels . ISBA's March 2025 brand safety chapter publishes ten floor categories and high, medium and low tiers; its hub names an overseer and pledges review, without a public procedure . The author's reading of that pledge as a stated change process makes H6 contradicted . Requiring a documented procedure makes it unresolved because the May 2026 full guide is member-only . Vendor references to GARM do not establish which dated edition their products implement .

GARM closed, and its framework stopped in 2022

The WFA announced GARM's discontinuation on 9 August 2024, citing the drain from allegations; its framework page gives 8 August . X and Rumble had sued on 6 August, and Chapter 9 dates those cases . The WFA's July 2026 promise not to restart GARM appeared in its settlement announcement, whose terms were unpublished .

The WFA page records June 2022 as the framework's latest update, with the document members-only . That edition has twelve floor categories and three risk tiers . The BSI copy used for textual comparison carries no date, version or attribution, so its identification with that edition is the paper's inference .

On 15 April 2026 the Federal Trade Commission (FTC) and eight states alleged that Dentsu, WPP Media and Publicis used the APB and GARM to set a common "Brand Safety Floor" aimed at "misinformation"; the agencies settled the same day by consent orders, no verification vendor is a defendant, and the case is dated in Chapter 9 .

Who publishes a floor now

Of the ten bodies H6 names (the WFA, the Association of National Advertisers (ANA), the 4A's APB, the IAB, IAB Tech Lab, TAG, the Media Rating Council (MRC), ISBA, BSI and the European Advertising Standards Alliance (EASA)), only ISBA publishes a floor and tiers under a date later than June 2022 in the documents screened to 10 November 2026 . The others certify against the floor, keep categories without tiers, host undated copies, publish principles, or show no framework of their own dated after June 2022 .

BodyDocument readDated after June 2022FloorTiersChange log or processEvidence
WFA (GARM)Brand Safety Floor + Suitability Framework page; document members-onlyNo: "Last updated: June 2022"12 categoriesHigh, medium, lowNone; GARM discontinued; no restart (29 July 2026)
ISBAResponsible Media Guide, brand safety chapterV1, March 2025; guide "Updated May 2026" (member-only)10 categoriesHigh, medium, lowNo log; pledge of regular review and update
TAGBrand Safety Certified Guidelines v5.0, Appendix BJuly 202612 categories, plus 11 IAB Tech Lab Sensitive Topics at "Floor"NoneChange log
4A's APBBrand Safety Floor and Brand Suitability FrameworkNo: 21 September 201813 categoriesRisk levels (2018)None found; the 4A's link leads to an unrelated members-only page
4A's, with the APBMedia Responsibility Principles, updated11 December 2023None (eleven principles)NoneUpdate announced; no log
BSIHosted floor-and-tiers tableUndated12 categoriesHigh, medium, lowNone; no owner or version
IAB Tech LabContent Taxonomy 3.1; 3.0 Descriptive Vectors3.1 posted 9 December 2024, last revised 22 July 2025; tier labels June 202211 Sensitive TopicsLabels only, undefinedVersioning with public comment
IABBrand Safety and Suitability GuideNo: December 2020As GARM and the APB thenAs GARM and the APB thenNone found
ANAQ3 2025 benchmark releaseNo framework in the releaseNot published"Low-risk" share onlyNot applicable
MRC2018 content-level supplementNo frameworkRefers to the APB's floorNone of its ownNot applicable
EASAPublication pages 1 to 5No framework; a December 2023 repository describes GARMNot applicableNot applicableNot applicable
UK Government Communication Service (outside the list)SAFER framework for digital brand safety2 March 2026, with a summary of changesNone (five principles)NoneSummary of changes
AgenticAdvertising.org (outside the list)Common Sense Brand Standards 1.0.0Effective 1 January 2026Exclusion floor and graduated categoriesNone definedNo revisions recorded

TAG keeps the floor inside a certification standard. Appendix B of its Brand Safety Certified Guidelines v5.0 lists the twelve "Brand Safety Floor" categories, linking to BSI's copy, and the eleven Sensitive Topics of IAB Tech Lab's Content Taxonomy version 2.2 at the "Floor" risk-rating, with no tiers and no mention of GARM . Version 2.2 dates from December 2020; IAB Tech Lab posted version 3.1 for comment in December 2024 . The guidelines keep a change log back to version 1.0 of September 2020 and certify companies, not products . H6 as registered counts a certification standard that reproduces the floor as certification maintenance unless it also publishes the tiers, and TAG's does not . In TAG's registry on 10 November 2026, DoubleVerify, IAS, Adloox, Channel Factory and illuma held the Brand Safety Certified seal, and Zefr was enrolled with no seal .

TAG links to BSI's undated, unattributed floor-and-tiers table. BSI also hosts the APB's 2018 framework and the 4A's July 2020 playbook .

The accreditor's standard points to a framework whose publisher's own link now leads elsewhere. MRC's 2018 supplement on content-level brand safety, the standard behind its content-level accreditation, sets no suitability tiers of its own: it asks for reporting discrete enough for buyers to set their own risk tolerance and refers to the APB's floor and Brand Suitability Framework . The APB released that framework on 21 September 2018, with thirteen categories . On 10 November 2026 the 4A's link labelled "Brand Safety Floor and Suitability Framework" led to an unrelated members-only page, and the only dated APB-related output found since June 2022 was the Media Responsibility Principles of 11 December 2023, updated by the 4A's Media, Tech and Data practice with the APB: eleven principles, with no floor categories or tiers .

The latest brand safety and suitability guide found on iab.com, from the IAB, the trade body of digital media sellers and platforms, is dated December 2020 . IAB Tech Lab maintains categories without tiers: its Content Taxonomy 3.1, posted for public comment on 9 December 2024 and last revised on 22 July 2025, keeps the eleven Sensitive Topics derived from GARM under versioning with public comment but has no tier rows, and the tier labels Floor, High, Medium and Low Risk appear, undefined, in older files: version 2.2 of December 2020 and the 3.0 Descriptive Vectors of June 2022 .

The ANA reports suitability results without publishing, in the releases read, the framework behind them. Its programmatic benchmark, built with TAG TrustNet and Fiducia on log-level data, added brand safety and suitability metrics in Q3 2025 from what it calls "an AI-powered standard" and reported 99.1% of the benchmark's programmatic spend in low-risk environments, without saying who classifies "low-risk" or against which framework or tiers; the benchmark report itself was not opened .

Two dated documents outside H6's list stop short of a floor-and-tiers framework. The UK Government Communication Service's SAFER framework of 2 March 2026 updates its 2020 SAFE framework with a summary of changes, but it is one advertiser's five principles, with no floor or tiers . AgenticAdvertising.org's policy registry serves Common Sense Brand Standards 1.0.0, a standard that originated at Scope3 (now Apostra) and took effect on 1 January 2026, with an exclusion floor and "graduated assessment" categories but no defined tiers; its history showed no revisions on 10 November 2026 . The author co-leads that organisation's Signals & Measurement working group, a declared interest; agent buying is the subject of Chapter 10 .

JICWEBS went into TAG

The UK's Joint Industry Committee for Web Standards (JICWEBS), which once ran UK digital trading standards, now sits inside TAG, and the one floor-and-tiers framework found with a date after June 2022 comes from ISBA, one of its four former controllers . JICWEBS was controlled by the UK trade bodies of online publishers (AOP), digital advertising (IAB UK), agencies (IPA) and advertisers (ISBA); its Digital Trading Standards Group (DTSG), set up in 2012, asked buyers and sellers to state in contract where ads should or should not appear, using independently certified content verification tools or schedules (lists) .

DateEventEvidence
2012JICWEBS's Digital Trading Standards Group set up
31 August 2016JICWEBS Ltd incorporated, controlled by AOP, IAB UK, IPA and ISBA
March 2017JICWEBS and TAG announce a formal partnership
January 2019TAG's Certified Against Fraud replaces JICWEBS's anti-fraud principles as the UK's main anti-fraud standard
21 January 2020Content Verification Guide issued by AOP, IAB UK, IPA, ISBA, JICWEBS and Newsworks
29 July 2020TAG and JICWEBS announce their merger, for completion in early September
1 September 2020TAG recorded at Companies House as JICWEBS Ltd's controller (75% or more of votes); the four trade bodies cease
10 September 2020TAG Brand Safety Certified launches, version 1.0 merging the DTSG with TAG's Inventory Quality Guidelines; 112 companies certified at launch
March 2025ISBA's Responsible Media Guide, brand safety chapter V1
May 2026ISBA's page marks the guide "Updated May 2026" (member-only)
July 2026TAG Brand Safety Certified Guidelines v5.0
10 November 2026JICWEBS Ltd still "Active" at Companies House, its confirmation statement overdue; jicwebs.org redirects to tagtoday.net

The merger also drew the UK bodies toward TAG's governance: the 2020 announcement said AOP, IAB UK, IPA and ISBA would join TAG's Leadership Council, and ISBA's guide says ISBA sits on that council and on MRC's board of directors, which this paper records as an interest in the bodies that grade verifiers, not as evidence of bias . ISBA's guide advises members rather than setting a trading standard: it calls an ad verification partner "the principal way to manage brand safety", names Adloox, DoubleVerify, IAS, Mantis and Zefr as examples, and says its guidance often feeds ISBA's member-only Media Services Framework for agency contracts .

ISBA's guide against the rule

ISBA's guide meets the rule's date, floor and tier elements on ISBA's own documents; only the change-process element needs a reading . The brand safety chapter, produced with The Aperto Partnership, an advisory firm to advertisers, carries "V1 March 2025" on its cover and calls the guidance "entirely voluntary" . It sets a floor of ten content categories with descriptions, a suitability framework of high, medium and low risk "against the same ten categories", and an adjacency framework . Its descriptions match the unattributed, undated floor-and-tiers text BSI hosts, which this paper identifies as GARM's June 2022 framework; the WFA's own copy is members-only and was not compared . The guide does not name GARM, drops two of GARM's twelve categories (debated sensitive social issues and misinformation), presents the floor as "industry-accepted" and ties the tiers to the inventory filters platforms already offer .

Element of the ruleWhat ISBA's documents showReadingEvidence
Published by a screened bodyISBA is one of the ten bodies H6 namesMet
A dated edition later than June 2022Chapter cover "V1 March 2025"; hub page dated 20 March 2025 and marked "Updated May 2026"Met
Floor categoriesTen, with descriptionsMet
Suitability tiersHigh, medium and low against the same tenMet
A change log or a stated change processNo change log, decision procedure or review cycle published; a pledge that the guide will be "regularly reviewed and updated", oversight by ISBA's Media Leaders group, and a promise to update after further Ofcom guidanceThe judgement

The remaining judgment concerns what the rule means by a change process. ISBA names oversight by its Media Leaders group, pledges regular review and identifies further Ofcom guidance as an update trigger . Its hub records a May 2026 update to the member-only full guide, whose effect on the public March 2025 floor and tiers was not readable .

The author treats the pledge, overseer and trigger as the rule's stated change process and codes H6 contradicted . A requirement for decision rights and a documented procedure instead yields unresolved . ISBA is a screened body under the rule, although its voluntary UK guide leaves the hypothesis statement's broader cross-industry wording open to challenge .

What GARM alignment means now

Where a scored product names a framework on the pages read, it names GARM, an IAB category list or an unnamed industry standard, and none names an edition . A House Judiciary Committee staff report of July 2024, a characterisation rather than a finding, said committee documents suggested GARM had been building its framework into the products of the ad-tech providers it invited, among them DoubleVerify, IAS, Peer39, Channel Factory, Pixability and Zefr (Chapter 9) .

Product (cohort)Framework namedEdition namedEvidence
IAS advertiser verification (A)Low, Medium or High "across the supported 12 GARM content categories"; social-platform models "trained against the GARM Framework"; a separate open-web taxonomy of 8 categories and 5 risk thresholdsNone
DoubleVerify advertiser verification (A)Its own Brand Safety Floor and Suitability Tiers, "mapped to the GARM framework" (January 2024 release); over 100 avoidance categories; no product page read names GARMNone
Adloox verification (A; owned by Peer39 since June 2026)"Based on the GARM framework" in Google's DV360 help; 12 exclusion segments on Adloox's DV360 page; "industry-standard BS&S framework" in its Meta releaseNone
mFilterIt Ad Traffic Validation (A)Brand-safety categories "as per the GARM guidelines"None
HUMAN Ad Integrity Suite (B)None on its brand-safety page (captured 4 February 2026)None
Zefr brand safety and suitability (C)GARM by name through 2023; twelve "industry-aligned" categories from September 2024; "GARM-aligned" on its Media Services pageNone
Peer39 brand safety and suitability (C)"The industry standard brand safety and suitability framework, formerly known as GARM" (30 October 2026)None
illumaSentinel (C)None named on illuma's pages read on 10 November 2026None
ViewIQ, Channel Factory (C)"IAB standard category lists"; "GARM+ strategies", undefinedNone

IAS's brand safety and suitability overview, published on 6 March 2024 and still online when read on 10 November 2026, scores suitability "across the supported 12 GARM content categories" and says its social-platform models are "trained against the GARM Framework"; its help-centre articles on GARM sit behind a login and were not read . Google's pages carry the name too. DV360's help, read on 10 November 2026, still offers Adloox's pre-bid filters under "Scope3 (formerly Adloox)", "based on the GARM framework", although Peer39 has owned Adloox since June 2026 . Google's Advertising Technology Partners directory describes its brand-safety partners' verification as "based on global industry standards defined by" GARM . Peer39, writing on 30 October 2026, calls its basis "the industry standard brand safety and suitability framework, formerly known as GARM", and names no maintainer .

Other vendors have dropped the name and kept the count. Zefr's releases mapped content to eleven GARM categories at its TikTok launch in January 2022 and to every GARM category, misinformation included, from February 2023; its September 2024 YouTube and October 2024 Snap releases describe twelve "industry-aligned" categories without naming GARM . DoubleVerify called its own floor and tiers "mapped to the GARM framework" in January 2024, and none of its product pages read on 10 November 2026 names GARM or publishes category definitions . Channel Factory's "GARM+" cannot be mapped to the June 2022 edition from its pages .

The WFA page documents no official GARM edition later than June 2022 . Counts and labels do not show which definitions a vendor implements or whether it has adapted them . The public floors also differ, with ten categories at ISBA, twelve at TAG and BSI, eleven IAB Tech Lab Sensitive Topics, and thirteen in the APB's 2018 floor. A framework name alone cannot identify the categories used .

H6 under its rule

The primary H6 verdict remains contradicted at medium confidence on the author's interpretation of ISBA's review pledge . The ten-body screen and the table above document the date, floor and tiers; treating a process as requiring published decision rights instead yields unresolved .

The other nine screened bodies supplied no qualifying edition in the documents read to 10 November 2026 . H6's separate implementation clause cannot be settled from the product pages, which name no dated edition; the apparent June 2022 mapping remains an inference and GARM+ is unmapped . Both the framework record and the primary judgment are due for recheck on 11 December .

Chapter 9

The record, 2023 to 2026

From January 2023 to 10 November 2026 the public record on the scored products is letters, case reports, complaints, pleading-stage rulings, voluntary dismissals, consent orders against advertising agencies and the parties' own statements. In the sources screened, no court, regulator or accreditor made a finding on the merits about any scored product's accuracy; the only regulator decision on the merits in the registers searched, Australia's clearance of DoubleVerify's sale, concerns competition. Of the matters still open, only DoubleVerify's suit against Adalytics turns on how a verification product performs, and its one ruling took DoubleVerify's allegations as true. The record sits outside the Verdict Quotient (VQ), the paper's score: nothing in it scores. Its absences are bounded to the public sources searched on 10 November 2026, beyond which lie nonpublic inquiries, sealed filings and state dockets.

In short

Senators sent eight letters in two rounds before DoubleVerify sued Adalytics in May 2025; no action on the requests was found in the screened agency sources . The April 2026 ruling in that suit accepted the complaint's allegations for the motion and made no factual finding about the report's accuracy . DoubleVerify's securities class action was voluntarily dismissed without prejudice; IAS's was dismissed with prejudice and appealed . The FTC withdrew its investigative demand to a third party in April 2026 and settled with three agency groups, with no verifier named as a defendant . X's and Rumble's GARM cases ended in pleading-stage dismissals; after the WFA settled, X's appeal continued against other defendants . Australia's clearance of DoubleVerify's proposed sale decides competition and establishes no measurement accuracy .

What the record is

Q1 records each letter, complaint, ruling and response by its legal posture and primary date, outside the VQ. Its fixed rule keeps allegations, pleading decisions and voluntary dismissals apart from factual findings. A request for an investigation does not show that one opened. Unread documents are identified, including Kaszirer's dismissal term and the Delaware IAS-share case . The window starts on 1 January 2023; pending matters were checked on 10 November 2026 and are due for recheck on 11 December .

The register

The table lists, oldest first, each record found that concerns a scored product or GARM, the WFA's Global Alliance for Responsible Media.

DateForumDocumentWhat it isStatus at 10 November 2026Evidence
4 May 2023Autorité de la concurrence (France)Decision 23-MC-01 on Adloox's request for interim measures against MetaRuling (interim measures; merits not decided)Final, not appealed; no merits decision listed
21 September 2023C.D. Cal. 2:23-cv-07914Montgomery Venkersammy v. Human Security, removed to federal courtComplaint (subject not established)Reported settled by mediation; terminated 22 May 2024
5 February 2024D. Minn. 0:24-cv-00310Thor v. DoubleVerifyComplaint (subject not established)Stayed 23 February 2024; dismissed 27 January 2025
18 March 2024 (public by)AdalyticsReport on ads on made-for-advertising sites in January 2024Independent report (cases)Published
4 April 2024 (public by)AdalyticsReport on the Forbes www3 subdomainIndependent report (one case)Published
10 May 2024 (first capture)AdalyticsReport on user-ID rotation at Colossus, naming HUMAN only as the supply-side platform's declared partnerIndependent report (one case)Published
10 July 2024House Judiciary Committee, majority staff"GARM's Harm", interim staff reportCongressional staff reportFinal at publication
6 August 2024N.D. Tex. 7:24-cv-00114X Corp. v. World Federation of Advertisers et al.ComplaintDismissed 26 March 2026
6 August 2024N.D. Tex. 7:24-cv-00115Rumble v. World Federation of Advertisers et al.ComplaintDismissed 13 August 2025
7 August 2024Adalytics"AI" brand safety technologyIndependent report (cases)Published
9 August 2024WFAStatement discontinuing GARMParty statementGARM discontinued
11 October 2024Marketing BrewReport of questions from Justice Department and Naval Criminal Investigative Service officials to advertising executives about IAS and DoubleVerifyPress report of a nonpublic inquiryUnverified; no primary record
29 January 2025S.D.N.Y. 1:25-cv-00847Oklahoma Firefighters Pension and Retirement System v. Integral Ad ScienceSecurities class action complaintDismissed 2 March 2026; on appeal
7 February 2025Senators Blackburn and BlumenthalLetter to DoubleVerifyLetter (five questions)Answers asked by 14 February 2025; none found in the pages opened
7 February 2025Senators Blackburn and BlumenthalLetter to IASLetter (questions)IAS says it answered; the answer was not found
7 February 2025Senators Blackburn and BlumenthalLetter to AmazonLetter (dated from the release; text not read)Answer not searched for
7 February 2025Senators Blackburn and BlumenthalLetter to GoogleLetter (dated from the release; text not read)Answer not searched for
7 February 2025Senators Blackburn and BlumenthalLetter to the Media Rating CouncilLetter (dated from the release; text not read)Answer not searched for
7 February 2025Senators Blackburn and BlumenthalLetter to the Trustworthy Accountability GroupLetter (dated from the release; text not read)Answer not searched for
7 February 2025AdalyticsReport on ads on an image host (imgbb.com, ibb.co)Independent report (cases; advertisers' statements relayed)Published
11 March 2025S.D.N.Y. 1:25-cv-02039Tauber derivative complaint against IAS directors, consolidated with Newman (filed 2 June 2025)Derivative complaintsVoluntarily dismissed 15 January 2026
24 March 2025Senator WarnerLetter to the FTC chairman, released 28 MarchLetter (request to investigate; answers by 28 April 2025)No FTC action found
24 March 2025Senator WarnerLetter to the Attorney General, released 28 MarchLetter (request to investigate; no deadline)No Justice Department action found; its releases not searched separately
28 March 2025AdalyticsReport on pre-bid bot detection and filtrationIndependent report (cases)Subject of DoubleVerify v. Adalytics
31 March 2025S.D.N.Y. 1:25-cv-02715Martin v. Zagorski, pro se, against DoubleVerify, Providence Equity and four individualsComplaint (whistleblower retaliation, civil rights)Dismissed 16 September 2026
30 April 2025S.D.N.Y. 1:25-cv-03614Pixalate v. Criteo, Pixalate as plaintiffContract complaint (subject not established)Voluntarily dismissed 18 June 2025
12 May 2025D. Md. 8:25-cv-01535DoubleVerify v. AdalyticsComplaint (Lanham Act, defamation, injurious falsehood, tortious interference, unfair competition)Partly dismissed 27 April 2026; continuing
20 May 2025FTC, File No. 251-0061Civil investigative demand to the Global Disinformation Index (GDI), listing IAS and DoubleVerify among thirteen rating entitiesInvestigative demand to a third partyWithdrawn 16 April 2026
22 May 2025S.D.N.Y. 1:25-cv-04332Securities class action against DoubleVerifySecurities class action complaintVoluntarily dismissed 22 December 2025
13 August 2025N.D. Tex. 7:24-cv-00115Rumble v. WFA, opinionRuling (personal jurisdiction and venue; merits not reached)Closed; no appeal or refiling found at the 9 November check
9 December 2025S.D.N.Y. 1:25-cv-10200Kaszirer v. Zagorski, against ten DoubleVerify officers and directorsDerivative complaintNotice of voluntary dismissal filed 4 February 2026, entered 6 February; prejudice term not read
22 December 2025S.D.N.Y. 1:25-cv-04332Lead plaintiffs' notice in the DoubleVerify class actionVoluntary dismissal without prejudiceClosed 23 December 2025
3 January 2026arXiv (New Jersey Institute of Technology; DeepSee.io)Smith et al., consistency of brand-safety classificationsIndependent study (agreement between vendors)Published
15 January 2026S.D.N.Y. 1:25-cv-02039Stipulation and order in the IAS derivative litigationVoluntary dismissal (terms not read)Closed
2 February 2026D.D.C. 1:25-cv-04137GDI's amended complaint against the FTC, alleging IAS received a demandComplaint (pleaded on information and belief)Closed 4 May 2026
February 2026C.D. Cal. 2:26-cv-01108Singer v. Comscore, privacy class action (not shown to concern Proximic)ComplaintDismissed with leave to amend 30 June 2026
26 February 2026Delaware Court of Chancery, per Law360A Glazer Capital fund's suit for payment for its IAS sharesPress report of a complaintUnverified
2 March 2026S.D.N.Y. 1:25-cv-00847Oklahoma Firefighters v. IAS, opinionRuling (pleading stage; dismissal with prejudice)Judgment 5 March; appeal noticed 26 March; briefing per the plaintiffs' firm (undated)
26 March 2026N.D. Tex. 7:24-cv-00114X Corp. v. WFA, opinionRuling (no antitrust injury, with prejudice for the WFA and domestic defendants; personal jurisdiction, without prejudice, for four foreign defendants)On appeal, 5th Cir. 26-10394
15 April 2026FTC and eight states; N.D. Tex. 4:26-cv-00469FTC v. Dentsu, WPP Media and PublicisComplaint; consent orders (settlement, no admission)Orders entered; Dentsu's runs ten years
16 April 2026FTCLetter withdrawing GDI's demand; notice that several other demands were withdrawnWithdrawalGDI's suit closed 4 May 2026
27 April 2026D. Md. 8:25-cv-01535DoubleVerify v. Adalytics, opinionRuling (pleading stage; tortious interference dismissed without prejudice)Answer filed; status report due 9 February 2027
5 May 2026S.D.N.Y. 1:26-cv-03727Chavez v. Human Security, privacy class actionComplaintVoluntarily dismissed with prejudice, so ordered 21 July 2026
11 May 2026D. Md. 8:25-cv-01535Adalytics' answerAnswer (pleads truth; calls the suit a strategic lawsuit against public participation)Case continues
29 July 2026WFARelease on the settlement with XSettlement (terms not public); party statementX's appeal continues against the other defendants
5 August 20265th Cir. 26-10394Stipulation dismissing the appeal and cross-appeal between X and the WFA (Doc. 119)Stipulation (settlement)Granted 11 August 2026, per the 9 November check
12 August 20265th Cir. 26-10394United States amicus brief supporting X (Doc. 126)Amicus briefBriefing
12 August 20265th Cir. 26-10394West Virginia and 11 other states' amicus brief supporting reversal (Doc. 123)Amicus briefBriefing
16 September 2026S.D.N.Y. 1:25-cv-02715Martin v. Zagorski, opinionRuling (dismissal with prejudice, failure to state a claim)Judgment 17 September; no notice of appeal on the docket
8 October 2026N.Y. County Supreme CourtKent v. DoubleVerify Holdings et al., seen only through an aggregatorSummons and complaint (not read)Open per the aggregator; not confirmed as one of DoubleVerify's two complaints
13 October 2026ACCC, MN-55036Phase 1 approval of Neptune BidCo's acquisition of DoubleVerifyCompetition determinationAssessment completed
20 October 2026DoubleVerify, 8-K filingFourteen demand letters and two unserved stockholder complaints over the mergerIssuer filing reporting allegationsUnserved at 30 October 2026; DoubleVerify disputes them

The dated record, 2023 to 2026

Every regulator action, congressional letter, court filing, issuer filing and independent report in the register, dated from its primary document and coded by what it is. Letters are not findings, and a ruling is procedural unless it says otherwise. Nothing here scores in the VQ, and nothing here is legal advice.

Forum or issuerRecordHow the paper codes it, and its postureDateScored products namedDocumentSrc
C.D. Cal.Montgomery Venkersammy v. Human Security, Inc. (2:23-cv-07914)allegation: terminated 22 May 2024 (docket mirror)2023-09-21HUMANcomplaint (subject not established)
D. Minn.Thor v. DoubleVerify Inc. (0:24-cv-00310)allegation; procedural ruling (stay 23 Feb 2024; dismissal 27 Jan 2025): closed 27 Jan 20252024-02-05DoubleVerifycomplaint (subject not established)
AdalyticsAdalytics report on made-for-advertising sites, January 2024 (adalytics.io/blog/ads-observed-on-made-for-advertising-sites-in-january-2024)allegation: published2024-03-18IAS; DoubleVerify; HUMAN; Pixalate; Oracle Moatindependent report (case observations)
AdalyticsAdalytics report on the Forbes www3 subdomain (adalytics.io/blog/ads-observed-on-www3-forbes-subdomain)allegation: published; revised after first publication2024-04-04DoubleVerify; IAS; HUMAN; Oracle Moat (mentions)independent report
AdalyticsAdalytics report on user-ID rotation (Colossus SSP) (adalytics.io/blog/user-id-rotation)allegation: published2024-05-10HUMAN (as the SSP's declared partner)independent report
D. Md.Colossus Media, LLC v. Adalytics Research, LLC (8:24-cv-01402)allegation: not established (CourtListener blocked)2024-05-14none (researcher sued by an SSP)complaint (Lanham Act, defamation)
House Judiciary CommitteeGARM's Harm (interim staff report) (staff report, 39 pp.)allegation: final at publication2024-07-10DoubleVerify; IAS; Peer39; Channel Factory; Pixability; Zefr; Oracle Moatcongressional staff report
N.D. Tex.X Corp. v. World Federation of Advertisers et al. (complaint) (7:24-cv-00114)allegation: see 2026-03-262024-08-06none (GARM)complaint
AdalyticsAdalytics '"AI" brand safety technology' (adalytics.io/blog/ai-brand-safety)allegation: published2024-08-07DoubleVerify; IASindependent report
World Federation of AdvertisersWFA discontinues GARM (about-garm statement)party statement: GARM discontinued2024-08-09noneparty statement
Marketing Brew (anonymous sources)Reported DOJ and NCIS questions to ad executives about IAS and DoubleVerifyunverified (outside reach): no primary record2024-10-11IAS; DoubleVerifypress report of a nonpublic inquiry
DoubleVerifyDoubleVerify pre-emptive statement on the forthcoming bot report (Setting the Record Straight)party statement: live (modified 15 Oct 2025)2025-01-10DoubleVerifyvendor statement
S.D.N.Y.Oklahoma Firefighters Pension and Retirement System v. IAS (complaint) (1:25-cv-00847)allegation: see 2026-03-022025-01-29IAS (company)securities class action complaint
Senators Blackburn and BlumenthalSix Senate letters (Amazon, Google, DoubleVerify, IAS, MRC, TAG) (letters dated 7 Feb 2025)allegation (questions): answers requested by 14 Feb 2025; none public found2025-02-07DoubleVerify; IAScongressional letters
AdalyticsAdalytics image-host (imgbb.com, ibb.co) report (adalytics.io/blog/adtech-vendors-csam-full-report)allegation: published2025-02-07DoubleVerify; IASindependent report
IAS (SEC filing)IAS FY2024 10-K, Item 3: no material proceeding (10-K FY2024)party statement: superseded by 10-Q disclosure2025-02-28IASissuer filing
S.D.N.Y.Tauber derivative complaint (IAS directors) (1:25-cv-02039)allegation: see 2026-01-152025-03-11IAS (company)derivative complaint
Senator Mark R. WarnerWarner letters to the FTC and DOJ (two signed letters)allegation (requests to investigate): no agency action found2025-03-24IAS; DoubleVerify; HUMANcongressional letters
AdalyticsAdalytics pre-bid bot report (adalytics.io/blog/prebid-bot-filtration)allegation: subject of DoubleVerify v. Adalytics2025-03-28DoubleVerify; IAS; HUMANindependent report
S.D.N.Y.Martin v. Zagorski (complaint) (1:25-cv-02715)allegation: see 2026-09-162025-03-31DoubleVerify (company and officers)pro se complaint
S.D.N.Y.Pixalate, Inc. v. Criteo S.A. (Pixalate as plaintiff) (1:25-cv-03614)allegation: voluntarily dismissed 18 Jun 2025; terminated 20 Jun 20252025-04-30Pixalate (plaintiff)contract complaint
D. Md.DoubleVerify v. Adalytics (complaint) (8:25-cv-01535)allegation: see 2026-04-272025-05-12DoubleVerify (plaintiff)complaint (Lanham Act, defamation and others)
FTC (File No. 251-0061)FTC civil investigative demand to GDI naming IAS and DoubleVerify among rating entities (CID specification 10)regulator action (investigative step): withdrawn 16 Apr 20262025-05-20IAS; DoubleVerify (named, not addressed)regulator investigative demand to a third party
S.D.N.Y.DoubleVerify securities class action (complaint) (1:25-cv-04332)allegation: see 2025-12-222025-05-22DoubleVerify (company)securities class action complaint
S.D.N.Y.Newman v. Utzschneider (derivative, IAS directors) (1:25-cv-04622)allegation: consolidated 12 Jun 20252025-06-02IAS (company)derivative complaint
S.D.N.Y.Amended class complaint against IAS adds Vista fund entities (1:25-cv-00847 ECF 30)allegation: see 2026-03-022025-06-16IAS (company)amended complaint
N.D. Tex.Rumble v. WFA dismissed without prejudice (jurisdiction and venue) (7:24-cv-0115 ECF 69)procedural ruling: closed; no appeal found (9 Nov check)2025-08-13none (GARM)court opinion
IAS (SEC filing)IAS Q3 2025 10-Q discloses the class action (10-Q Q3 2025, Note 14)party statement: IAS's last periodic report2025-11-04IASissuer filing
S.D.N.Y.Kaszirer v. Zagorski (derivative complaint) (1:25-cv-10200)allegation: see 2026-02-062025-12-09DoubleVerify (company)derivative complaint
IASIAS drops GDI input from Context Control Avoidance (news release)party statement: live2025-12-18IASvendor statement (method change)
S.D.N.Y.DoubleVerify class action voluntarily dismissed without prejudice (1:25-cv-04332 ECF 35)voluntary dismissal (neither vindication nor liability): closed 23 Dec 20252025-12-22DoubleVerifynotice of voluntary dismissal
NJIT; DeepSee.ioSmith et al., brand-safety classification consistency study (arXiv:2601.01303)independent report (agreement estimand): published (ICWSM 2026 per the brief)2026-01-03DoubleVerify; IAS; Oracleindependent academic study
S.D.N.Y.IAS derivative litigation dismissed by stipulation and order (1:25-cv-02039 ECF 19)voluntary dismissal (terms not read): closed2026-01-15IASstipulation and order of voluntary dismissal
D.D.C.GDI amended complaint alleges IAS received an FTC CID (1:25-cv-04137 ECF 29)allegation (information and belief): case closed 4 May 20262026-02-02IAScomplaint against the FTC
S.D.N.Y.Kaszirer notice of voluntary dismissal entered as signed (1:25-cv-10200 ECF 5)voluntary dismissal (prejudice term not read): closed2026-02-06DoubleVerifynotice of voluntary dismissal
C.D. Cal.Singer et al. v. Comscore, Inc. et al. (privacy class action) (2:26-cv-01108)allegation: pending at 26 Mar 2026 (10-K)2026-02Comscore (company; Proximic not shown)putative class action complaint
Law360 (secondary)Glazer Capital fund sues for payment for IAS shares (Delaware Chancery)allegation (unverified): not established2026-02-26IASpress report of a complaint
S.D.N.Y.Oklahoma Firefighters v. IAS dismissed with prejudice (1:25-cv-00847 ECF 62)procedural ruling (pleading stage): judgment 5 Mar 2026; appeal noticed 26 Mar 20262026-03-02IAScourt opinion
N.D. Tex.X Corp. v. WFA dismissed (with prejudice for WFA and domestic defendants) (7:24-cv-00114 ECF 255)procedural ruling (failure to state a claim): on appeal, 5th Cir. 26-103942026-03-26none (GARM)court opinion
S.D.N.Y. to 2d Cir.Notice of appeal in Oklahoma Firefighters v. IAS (1:25-cv-00847 ECF 64)procedural: being briefed (plaintiffs' firm, undated)2026-03-26IASnotice of appeal
FTC; N.D. Tex.FTC and eight states v. Dentsu, WPP Media and Publicis; consent orders (4:26-cv-00469)allegation; settlement (consent order, no admission): orders entered; 10-year term2026-04-15none (agencies; brand-safety floor)regulator complaint and consent orders
FTCFTC withdraws GDI's CID; 'several other CIDs' withdrawn (ECF 37 and 37-1, D.D.C. 1:25-cv-04137)regulator action (withdrawal): GDI case closed 4 May 20262026-04-16none namedregulator letter and notice
D. Md.DoubleVerify v. Adalytics: motion to dismiss granted only as to tortious interference (8:25-cv-01535 ECF 30)procedural ruling (pleading stage): answer filed; status report due 9 Feb 20272026-04-27DoubleVerifycourt opinion
S.D.N.Y.Chavez v. Human Security, Inc. (privacy class action) (1:26-cv-03727)allegation: see 2026-07-212026-05-05HUMANputative class action complaint
D. Md.Adalytics answer (truth; anti-SLAPP) (8:25-cv-01535 ECF 32)party statement: case continues2026-05-11DoubleVerifyanswer
S.D.N.Y.Chavez v. HUMAN voluntarily dismissed with prejudice (1:26-cv-03727 ECF 18)voluntary dismissal (with prejudice; no finding): closed2026-07-21HUMANnotice of voluntary dismissal, so ordered
World Federation of AdvertisersWFA and X settle; WFA will not restart GARM (release)settlement (terms not public) and party statement: appeal continues against other defendants2026-07-29noneparty statement
5th Cir.X and WFA stipulate to dismissal of their appeal and cross-appeal (26-10394 Doc. 119)settlement (terms not public): granted 11 Aug 2026 (9 Nov check)2026-08-05nonestipulation
5th Cir.United States amicus supporting X (26-10394 Doc. 126)party-side position: briefing2026-08-12noneamicus brief
5th Cir.West Virginia and 11 states amicus supporting reversal (26-10394 Doc. 123)party-side position: briefing2026-08-12noneamicus brief (state AGs)
D. Md.DoubleVerify v. Adalytics: next status report due 9 Feb 2027 (8:25-cv-01535 ECF 40)procedural ruling: pending2026-09-02DoubleVerifyscheduling order
S.D.N.Y.Martin v. Zagorski dismissed with prejudice (1:25-cv-02715 ECF 61)procedural ruling (pleading stage; judgment for defendants): judgment 17 Sep 2026; no appeal on docket at 10 Nov 20262026-09-16DoubleVerify (company and officers)court opinion
N.Y. County Supreme CourtMichael Kent v. DoubleVerify Holdings, Inc. et al. (index number masked)allegation (unverified caption): open per aggregator (14 Oct 2026)2026-10-08DoubleVerify (company and directors)summons and complaint (not read)
ACCCACCC approves Neptune BidCo acquisition of DoubleVerify (Phase 1) (MN-55036)finding on the merits of a competition review only: assessment completed2026-10-13DoubleVerifyregulator determination (competition)
DoubleVerify (SEC filing)Fourteen demand letters and two unserved stockholder complaints over the merger (8-K Item 8.01)party statement (reporting allegations): complaints unserved2026-10-20DoubleVerifyissuer filing
DoubleVerify (SEC filing)Two merger complaints still unserved (10-Q Q3 2026)party statement: unserved2026-10-30DoubleVerifyissuer filing
Machine-readable copy: data/rules.csv.

Eight letters and a staff report

Congress asked questions and characterised documents; no agency action on its requests was found .

On 7 February 2025, Senators Marsha Blackburn and Richard Blumenthal sent six letters about Adalytics' image-host report to Amazon, Google, DoubleVerify, IAS, MRC and TAG . DoubleVerify's letter asked five questions with answers due on 14 February; the other four recipients' letters were dated from the release without their texts being read . IAS said it answered , but no public answer from either verifier was found in the pages opened, without a dedicated search . These letters convey questions and allegations.

Senator Mark Warner's two letters were dated 24 March and released on 28 March 2025 . One asked the FTC to investigate the vendors' bot-filtering claims, with answers due on 28 April; the other sought Justice Department investigation of alleged misrepresentation to federal clients . The FTC letter distinguished general invalid traffic (GIVT) from fraud and used a vendor-linked fraud estimate . The screen found no action on the requests, although Justice Department releases were not separately searched .

Both rounds cite a Marketing Brew report of 11 October 2024, from anonymous sources, that Justice Department and Naval Criminal Investigative Service (NCIS) officials had asked advertising executives about IAS and DoubleVerify; with no primary record found, the item stays unverified .

The House Judiciary Committee's majority staff published "GARM's Harm", an interim report, on 10 July 2024 . It names DoubleVerify, IAS, Oracle Moat, Peer39, Channel Factory, Pixability and Zefr among eleven ad-tech providers GARM invited into its community, and says committee documents suggest GARM was building its framework into their products . It is an investigating committee's staff characterisation, not a finding, and accuses no vendor of unlawful conduct in the passages read; no response was found on Zefr's or Channel Factory's pages .

What regulators did

No regulatory act found makes a finding about a scored product's measurement .

The FTC's 20 May 2025 civil investigative demand to the Global Disinformation Index (GDI) named IAS and DoubleVerify among thirteen rating entities . GDI alleged on information and belief that IAS also received a demand; DoubleVerify's proxy repeated news accounts whose description matches IAS only by inference . The FTC withdrew GDI's demand on 16 April 2026 and reported withdrawing other demands . The record does not establish whether either verifier received one.

On 15 April 2026 the FTC and eight states sued Dentsu, WPP Media and Publicis over alleged boycott-related use of the APB and GARM and settled the same day . Dentsu's ten-year order, entered without admission, bars specified political, ideological, journalistic or diversity-related rating criteria unless an advertiser expressly directs their use . No verifier is a defendant. The order governs the agency's reliance on ratings .

In France, Adloox, owned by Peer39 since June 2026, complained in October 2022 that Meta had refused it access to Meta's viewability and brand-safety partner programmes since 2016 . On 4 May 2023 the Autorité de la concurrence, without deciding the merits, ordered Meta to publish objective, transparent, non-discriminatory and proportionate access criteria; the decision was not appealed, its measures bind Meta and its findings are provisional, and no decision on the merits was listed on 10 November 2026 . The authority sits outside the registers fixed for Q1, as an additional source .

The ACCC's Phase 1 approval of Neptune BidCo US Inc.'s acquisition of DoubleVerify, published on 13 October 2026, is a finding on the merits of a competition review only; the UK Competition and Markets Authority (CMA) and the European Commission returned no case for the vendors queried . The sale, pending on 10 November 2026, belongs to Chapter 7 .

DoubleVerify v. Adalytics

Of the matters open on 10 November 2026, only this suit turns on how a verification product performs; it has one pleading-stage ruling and no judgment . DoubleVerify sued in the District of Maryland on 12 May 2025 over the bot report of 28 March 2025, pleading false advertising under the Lanham Act, defamation, injurious falsehood, tortious interference and unfair competition .

On 27 April 2026 the court dismissed the tortious-interference count without prejudice and let the rest proceed . It held that DoubleVerify had not alleged facts showing that the report's statement that some ads reached bots before the bid was false, but had plausibly alleged that the report falsely implies its pre-bid bot detection does not work, while calling one of DoubleVerify's supports weak . The ruling took DoubleVerify's allegations as true and is not a finding about the report; there is no judgment. Adalytics answered on 11 May 2026, pleading that the statements are true or substantially true and that the case is a SLAPP suit (a strategic lawsuit against public participation) under Maryland law, with no counterclaim . Discovery is under way, and the next status report is due on 9 February 2027 .

The complaint is also DoubleVerify's account of money and accuracy. It says GIVT impressions are deducted from the billable count and sophisticated invalid traffic (SIVT) flagged so advertisers can seek compensation, citing no contract, platform term or settlement . Its detection figures, such as pre-bid filtering that keeps customers from bidding on or paying for 99% of unwanted GIVT impressions, come with no published test set, denominator or period . Chapter 4 and Chapter 3 test those accounts.

Shareholder suits and filings

Shareholders sued both public verifiers in 2025. The recorded dispositions established no product error rate or factual finding about measurement accuracy .

Electrical Workers Pension Fund, Local 103, I.B.E.W., filed the class action against DoubleVerify in the Southern District of New York on 22 May 2025, for buyers of its stock from 10 November 2023 to 27 February 2025 . Among other things it alleged that DoubleVerify systematically overbilled customers for impressions served to declared bots . The lead plaintiffs dismissed it voluntarily and without prejudice on 22 December 2025, before filing any amended complaint, and no refiled federal action naming DoubleVerify appeared in the party-name search of 10 November 2026 . A voluntary dismissal is neither vindication nor liability. A derivative suit, Kaszirer v. Zagorski, filed against ten officers and directors on 9 December 2025, ended with a notice of voluntary dismissal filed on 4 February 2026 whose prejudice term could not be read .

IAS's class action is about pricing disclosures. Oklahoma Firefighters Pension and Retirement System sued IAS, two officers and Vista Equity Partners on 29 January 2025 over undisclosed price competition from 2 March 2023 to 27 February 2024, after which the stock fell about 41% . On 2 March 2026 the court dismissed it with prejudice for failure to plead a material misrepresentation or omission, without reaching scienter or loss causation; the plaintiff noticed an appeal to the Second Circuit on 26 March, and that it is being briefed rests on the plaintiffs' firm's undated page . Two derivative suits ended in a stipulated voluntary dismissal on 15 January 2026, its terms unread .

IAS's February and May 2025 filings applied a material-adverse-effect test without naming the served class action; it first named the suit in August 2025 . Its later reporting suspension left no periodic filing covering the dismissal or appeal . The 10 November EDGAR screen found no filing containing Adalytics, bounded to that search's coverage . DoubleVerify's FY2025 10-K names the dismissed class action and records $3.5 million in unusual litigation and regulatory costs without allocating them . Both acknowledge billing disputes, and the federal case-name search found no advertiser or agency suit against either .

By 20 October 2026 DoubleVerify had received fourteen demand letters over its pending sale, and two stockholder complaints alleging deficient proxy disclosures had been filed, none served; it disputes them and names no court or caption . The only caption found, Kent v. DoubleVerify Holdings, filed in New York County Supreme Court on 8 October 2026, is visible only through a commercial aggregator and is not confirmed as one of the two . The buyer is Neptune BidCo US Inc., Nielsen's parent . The author works at Samba TV, which competes with Nielsen in TV measurement.

Other suits

The other suits found concern something other than a verification product, or could not be read . Martin v. Zagorski, filed pro se on 31 March 2025 against DoubleVerify, Providence Equity and four individuals, is a whistleblower-retaliation and civil-rights conspiracy suit, not a product claim; the court dismissed it with prejudice for failure to state a claim on 16 September 2026 . No notice of appeal was on the docket on 10 November, but a motion to extend the time to appeal remains possible until about 18 November .

Chavez v. Human Security, a putative class action filed on 5 May 2026, alleged that HUMAN's website bot-detection code covertly collects visitors' personal data; it names none of HUMAN's ad-fraud products, and the plaintiffs voluntarily dismissed it with prejudice, so ordered on 21 July 2026 . Two older dockets, one against HUMAN reported settled by mediation in May 2024 and one against DoubleVerify dismissed in January 2025, carry no public documents, so their subjects are unknown . A privacy class action against Comscore, filed in February 2026 and dismissed with leave to amend on 30 June 2026, is not shown to concern its Proximic suitability filters .

GARM in court

The WFA announced GARM's discontinuation three days after X and Rumble sued on 6 August 2024, citing allegations' resource drain . The announcement gave no court finding of conduct . X's complaint attached a WFA email chain about Meta's Zefr reporting partnership, but the later ruling does not mention Zefr .

The court dismissed X's claims on 26 March 2026, with prejudice for the WFA and domestic defendants for failure to state an antitrust claim, and without prejudice for four foreign defendants on jurisdiction . X appealed; after the WFA's July settlement, their appeal and cross-appeal were dismissed on 11 August . The United States and twelve states filed amicus briefs supporting X's appeal against the others . At the 9 November check, no argument was calendared .

Rumble's parallel suit against the WFA, Diageo, WPP plc and GroupM, filed the same day, was dismissed without prejudice on 13 August 2025 for lack of personal jurisdiction and improper venue; the court did not reach the merits, and no appeal or refiled case had been found by 9 November . Whether anyone now maintains GARM's floor and tiers is H6, tested in Chapter 8.

Reports and answers

The reports and responses establish different propositions, as Chapter 3 explains . The Adalytics sitemap screen found six reports naming scored vendors, among fourteen URLs modified since 2023; HUMAN appears in the user-ID report only as a declared partner . Adalytics sells advertiser analytics and states that publishers and ad-tech vendors do not fund its research .

The made-for-advertising (MFA) report, public by 18 March 2024, listed Oracle Moat, IAS, DoubleVerify, HUMAN, Pixalate and Fou Analytics among platforms observed monitoring ads on MFA sites in January 2024 . The report of 7 August 2024 showed IAS's publisher-side endpoint returning low-risk values for explicit pages, and ads carrying DoubleVerify or IAS code there, from a convenience sample . The report of 7 February 2025 relayed advertisers' statements that their vendors had marked all measured impressions on an image host safe or suitable, a second-hand, domain-level account; it showed ads on explicit adult pages there and records, without showing it, one archived instance beside putative abuse material . The report of 28 March 2025 showed ads served to bots in crawl datasets, disclaiming magnitude and fault; its statement that advertisers were billed rests on advertisers' non-public log files and invoices . Smith and co-authors, at the New Jersey Institute of Technology and DeepSee.io, posted a study on 3 January 2026 measuring how far the publisher-side labels of DoubleVerify, IAS and Oracle agreed, not whether they were right . Chapter 3 codes what each establishes.

DateResponderAnswersWhat it says, paraphrasedEvidence
7 August 2024DoubleVerify"AI" brand safety report; also the MFA and Forbes reportsThe report lacked clients' campaign settings and its results are manufactured; it had already classified the MFA sites and the Forbes subdomain as MFA
7 August 2024IAS"AI" brand safety reportAlmost all tags shown came from its publisher products; the report gave no timeframe or sample
10 January 2025DoubleVerifyBot report, before publicationAnswer in advance on GIVT handling and pre-bid settings
7 February 2025DoubleVerifyImage-host reportIts customers' impressions on the site were 0.000047% of the media transactions it measured in 30 days; pages are classified only above a level of ad traffic; the copy archived that day also put 83% of those impressions on the home page, a figure the live text dropped on 15 October 2025
7 February 2025IASImage-host reportThe two domains were 88,000 of nearly 1.9 billion webpages it observed in 60 days; it labelled them ineligible for monetisation
13 February 2025DoubleVerifyImage-host reportLaunched a default-on "Highly Illicit: Do Not Monetize" category
14 February 2025IASImage-host report and the senators' letterSays it answered the senators, and that more than 350 of the sites the National Center for Missing & Exploited Children had reported were already covered by its brand-safety products
21 March 2025IASBot report, a week before publicationMethodology post on its publisher tools
28 March and 3 April 2025IASBot reportThe report tested its publisher tag, not its advertiser tag; it classified URLScan.io impressions as GIVT, SIVT or reduced-value inventory
28 March and 7 April 2025DoubleVerifyBot reportGIVT is removed post-bid from billable counts; URLScan.io is not a declared bot; of the report's 115 examples, 62 had no DoubleVerify measurement, 10 fell outside its data retention and all 43 it could match were identified; no per-impression data
11 April 2025MRCBot reportPost-serve GIVT filtration is required and pre-bid filtration is not; it has no insight into billing; URLScan.io does not declare itself; it did not test each observation
12 May 2025DoubleVerify, in its complaintBot reportGIVT is deducted from billable counts and SIVT flagged so the advertiser can seek compensation; no contract cited
18 December 2025IASNo stated trigger; GDI links it to the FTC investigationStops using GDI's limited input in Context Control Avoidance, for product reasons
11 May 2026Adalytics, in its answerDoubleVerify's complaintDenies liability; pleads truth or substantial truth and Maryland's anti-SLAPP statute

Some answers were not found: no public HUMAN response to the bot report in the searches of 10 November 2026, and no vendor response to Smith and co-authors in the vendor pages opened, which was not a dedicated search . The MRC's answer states its standards and audit scope and says it did not test each observation; the MRC accredits the vendors the report names and had received the senators' February letter .

Where the search ran

Every absence here is bounded to named registers queried on 10 November 2026, and a zero means only that a register returned nothing for that name . The sources were fixed at registration: SEC EDGAR; CourtListener and Justia for federal courts; FTC, Justice Department and state attorney-general releases; senate.gov and house.gov; the ACCC, CMA and European Commission registers; the vendors' newsrooms; and Adalytics' and other researchers' publications, searched by vendor and product name. AI research agents ran the searches under the author's direction, and in the verifier pass a separate AI verifier, told to refute each finding, re-ran part of them the same day .

RegisterQueryResult on 10 November 2026Evidence
CourtListener, federal party namesDoubleVerify9 dockets, 5 filed since 1 January 2023
CourtListener, federal party namesIntegral Ad Science7 dockets, 3 since 2023
CourtListener, federal party namesHUMAN Security3 dockets, 2 since 2023
CourtListener, federal party namesPixalate5 dockets, 2 since 2023: the Criteo suit and an unreviewed trademark case that may not name Pixalate
CourtListener, federal party namesZefrNone in any year; in a full-text search, the X Corp. exhibits were the only verification matter on the first page
CourtListener, federal party namesPeer39, Adloox, Scope3, Protected Media (now Protected by Mediaocean), Channel Factory, Fraudlogix, TrafficGuard, Illuma, PixabilityNone filed since 2023
CourtListener, federal party namesAdalytics4 entries since 2023, among them Colossus Media's 2024 suit and DoubleVerify's
CourtListener, federal party namesComscoreNot obtained (human-verification page, not bypassed)
SEC EDGAR, full text"Adalytics", all filersNo filing
UK CMA case finderDoubleVerify, IAS, HUMAN Security, Pixalate, Zefr, Peer39, Adloox, Protected Media, Channel Factory, Scope3, "brand safety"No case
European Commission case searchDoubleVerify, Pixalate, ZefrNo case; the IAS query did not render
ACCC acquisitions registerNielsen and DoubleVerifyMN-55036, approved; determination published 13 October 2026
Autorité de la concurrence decisions"adloox"; "vérification publicitaire"Decision 23-MC-01; no later decision on the merits
Adalytics sitemapURLs modified in 2023 or later14 URLs; six reports name scored vendors

CourtListener's access failures prevented completion of the Comscore query and repeat checks of zero results; Justia's Kaszirer docket also remained unread . The screen omitted separate Justice Department and state attorney-general searches, party-name queries for mFilterIt, Sightly and FouAnalytics, and publications by FouAnalytics and Check My Ads . State cases such as Kent and the Delaware IAS-share suit rest on an aggregator or reporting . Sealed and nonpublic matters remain outside the screen. An access failure establishes no absence.

Chapter 10

Verdicts in agent buying

The 10 November 2026 screen found verification fields, implemented connectors and interoperability tests in agent buying, while production claims lacked campaign or transaction records tying a verdict to an agent's decision. H7 remains unresolved under the stated reading of its rule. No screened agreement gives such a verdict financial force. The author co-leads the Signals & Measurement working group of AgenticAdvertising.org, which maintains the Ad Context Protocol (AdCP), a declared interest.

In short

AdCP represents verification thresholds with a named measuring vendor and terms for whose count governs an invoice . It supplies no measurement accreditation or dispute task in version 3.2 . IAB Tech Lab's OpenProposal draft represents a viewability guarantee and remedy, without invalid-traffic (IVT) or brand-safety guarantee fields . AAO's 10 November registry held 35 public agents and no measurement-type agent; no agent passed the accepted-measurement-terms test . DoubleVerify's Butler/Till workflow claim and Zefr's unnamed closed alpha lack a transaction record connecting a verification verdict to an agent's decision . Trade reporting provides partial corroboration of Butler/Till's quality-analysis work, without settling that link . H7 remains unresolved under the paper's recorded interpretation, and no identified agreement was found giving a verdict financial force .

Interests

The author co-leads AAO's Signals & Measurement working group; findings on AdCP are a declared interest . No Fluff Advisory, the author's company, operates two agents in AAO's registry, and its sales agent is one of the two that pass the billing-window test counted below . The author filed four public comments on OpenProposal on 27 October 2026, none on its measurement fields . DoubleVerify agreed on 6 August 2026 to be acquired by Neptune BidCo US Inc., the parent of Nielsen, a deal not closed when checked on 10 November 2026, and the author works at Samba TV, which competes with Nielsen in TV measurement . AdCP's lead architect co-founded Scope3, now Apostra, which founded AAO and lent it seed funding . None of this earns or costs a protocol state.

The test

H7, registered on 9 November 2026, predicts that, by the freeze, verification verdicts appear in agent-buying protocols "at most as schema representation", with no documented production use or demonstrated transaction in an agent-driven buy and none bound to a financial decision by an identified agreement . It covers AdCP, IAB Tech Lab's Agentic Advertising Management Protocols (AAMP) and the scored products' agent interfaces, including servers on the Model Context Protocol (MCP), through which an AI assistant calls outside tools .

Shared rule 12 records five states separately: schema representation, which proves only that a verdict can be expressed; an implemented connector; a demonstrated interoperability test; documented production use; and a demonstrated transaction (Chapter 13). Membership and schema fields earn no deployment credit, and financial force by an identified agreement is a separate field . H7 is supported if "no source screened documents state 4 or state 5 for a verification verdict", contradicted if "a source documents production use (state 4) or a demonstrated transaction (state 5) for a verification verdict", and unresolved if "a production claim rests on a vendor statement with no named buyer, campaign or transaction record (recorded as a vendor assertion)" .

An AI research agent under the author's direction read the sources on 10 November 2026; in the verifier pass, a separate AI agent told to refute each finding re-opened them that afternoon. Neither used a web search engine, so every absence below is bounded to the sources logged . In agentic sources "verification" often means agent or seller identity (AdCP's seller verification, IAB Tech Lab's Agent Registry, HUMAN's Verified AI Agent), which earns no state for an ad-verification verdict .

What AdCP represents

AdCP 3.0.0 introduced vendor-named performance thresholds . Version 3.2.9's subset comprises viewability, IVT, completion rate, brand safety and attention, with IVT a ceiling and the others floors . Measurement terms name whose count governs invoicing, tolerance, billing window and remedies such as credit or extra delivery . These are schema capabilities. The sources screened to 10 November contained no public agreement applying them with a verifier .

AdCP's taxonomy assigns verification to a third party while declining to adjudicate competing vendors . Its limitations page disclaims MRC accreditation and leaves filtration and measurement to the delivery stack or a vendor . The parties can nevertheless choose whose count governs an invoice in the schema .

AdCP 3.2's documentation also stops short of settlement: a structured dispute task "does not exist in 3.2 and is not scheduled for it", a verification vendor's own integration is "out of AdCP scope", and measurement agent interfaces are "a follow-up"; issue #7064, which asks what dispute lifecycle should cover invalid traffic, credits and make-goods, was routed to the Spec Backlog on 21 September 2026 . In 3.2 an ivt or brand_safety standard names the vendor but not which of its metrics is judged: Apostra (formerly Scope3) worked around that with a private extension on 22 October 2026, and the 3.3.0-beta.5 prerelease of 7 November adds an optional metric_id, still a beta on 10 November .

AAO's hosted grader defers measurement-verification assessment until the task surface is frozen . Experimental accreditation fields remain vendor asserted, and the production-evidence program was not selected for 3.3 in the roadmap record read on 10 November . No protocol-body production matrix for verification verdicts was found there .

What AAMP represents

IAB Tech Lab develops all its agentic work under AAMP, and AAMP represents less than AdCP . OpenProposal, published as AAMP 3.0 draft 1 for comment on 22 September 2026 and not a ratified standard, represents a committed viewable_rate on an MRC basis, a selectable measurement_source vendor (its example offers IAS and DoubleVerify) and a shortfall_remedy of make-good or credit, with no invalid-traffic or brand-safety guarantee field . Tech Lab's release page extends comment to 10 November 2026, while the repository still shows 22 October . None of the three fields was on the main branches of the AAMP buyer-agent, seller-agent or primitives repositories on 10 November; an unmerged seller-agent branch of 19 October models them and maps Google Ad Manager's viewability partner to measurement_source .

The Agentic Real Time Framework (ARTF) 1.0 of 12 November 2025 names pre-impression fraud detection as a use case and lets an agent attach metric objects from OpenRTB, the real-time bidding specification, to an impression before the auction . Those are predictions used to select inventory, not post-delivery verdicts, and the reference implementation shows them in a sample with no verification vendor's connector . The AAMP buyer agent (2.4.2) models brand safety as excluded content categories and keywords, and its example brief lists IAS and DoubleVerify as required vendors in a field no code reads .

What the vendors have built

The scored vendors' agent interfaces are mostly their own tools, and where one is said to reach a buying protocol, the vendor is the only source .

DateVendorInterface or claimState reached, as codedEvidence
21 Jul 2025HUMANVerified AI Agent: open-source demonstration of agent identityNone for an IVT verdict (agent identity)
24 Jul 2025HUMANOpen-source MCP server for its security productsNone for an IVT verdict
4 Nov 2025DoubleVerifyDV AI Verification, aimed at AI-agent traffic and AI-generated contentNo protocol state
16 Dec 2025IASIAS Agent, an assistant inside IAS's interface; dashboard feature launched March 2026No MCP, AdCP or AAMP interface found
30 Apr 2026ZefrZain: AdCP-based YouTube buying integration and an MCP serverConnector, as the vendor states it
16 Jun 2026IASLaunch partner of Databricks CustomerLake, signals to agents through clean roomsNo protocol state
17 Jun 2026DoubleVerifyDV Neura: Insight Agent over MCP "today"; Activation Agent forecast for the third quarterConnector, as the vendor states it
24 Jul 2026DoubleVerifyBlog: Butler/Till buys through AdCP; PubMatic applies DoubleVerify protections before the auctionProduction use, vendor assertion, uncorroborated
14 Sep 2026ZefrClosed alpha with unnamed partners; Rapid Response Blocking AgentProduction use, vendor assertion, no partner named
17 Sep 2026DoubleVerifyActivation Agent "expected to become available in the months ahead"Forecast
5 Oct 2026IAS; DoubleVerifyChatGPT Ads pilots (measurement; suitability evaluation)No protocol state
Undated, read 10 Nov 2026PixalateAnalytics and Enrichment MCP connectors and IQ Agent for a client's own IVT data; OpenEPG MCP plannedConnector for retrieving data, as documented by the vendor
Read 10 Nov 2026Channel FactoryAERO agents for lists, bidding and reports; no protocol namedNo protocol state
Read 10 Nov 2026Peer39 and AdlooxNo agent interface found; statements of intentNone found

DoubleVerify's June release claimed an available Insight Agent over MCP and forecast an Activation Agent; its September release still forecast the latter . IAS Agent runs inside IAS's interface, and HUMAN's public MCP and identity tools establish no ad-verdict integration in the sources screened . Those records are kept apart from production use.

Zefr's Zain pages claim an AdCP buying integration, an MCP server and a closed alpha, without documenting a suitability verdict in an AdCP field . Pixalate documents connectors for querying client IVT data without an AdCP or AAMP implementation . Channel Factory names no protocol for AERO, while Peer39 and Adloox supplied statements of intent in the screened sources .

For Protected by Mediaocean (formerly Protected Media), Proximic by Comscore, Illuma, Fraudlogix and Escalated the screen went no further than AAO's member list, where none has a public profile, and, for Protected by Mediaocean, a brand.json check that found none . For mFilterIt, FouAnalytics, Pixability and Sightly the stream kept no record, so this chapter bounds no absence for them. IAS's closed pilot and DoubleVerify's pilot in development, both announced for ChatGPT Ads on 5 October 2026, are third-party measurement in a channel that AdCP's taxonomy models as a closed loop in which the seller is also the measurement vendor, first-party validation in this paper's terms (Chapter 1); neither is an agent-buying protocol integration .

The instrument below dates every release, announcement, pilot and claim, keeping announcement, availability and forecast apart.

The registry, as what it is

AAO's registry lists the agents its members choose to register, and its conformance tests run on AAO's hosted runner against fictional test kits: it records what registered agents passed in a test, not what anyone bought . On 10 November 2026 the public listing held 35 agents and none of type measurement or governance; AAO's member records add 24 members-only and 19 private agent entries, none a verification vendor's, and an unregistered agent would not appear at all .

Vastlint and TVIQ's agent each passed ten content-standards test steps, establishing implementations and tests for creative verdicts rather than a scored verifier . No agent passed accepted measurement terms, and one of the two billing-window passes belongs to the author's firm . The table retains every test denominator.

AAO's public records, read 10 November 2026ValueEvidence
Agents listed (sales / signals / creative / buying / unknown)35 (28 / 4 / 1 / 1 / 1)
Agents of type measurement or governance in the public listing0
Agents exposing content-standards tools; passing content_standards2; 2 (10 of 10 steps)
measurement_terms_accepted: passing / untested / no record0 / 11 / 24
measurement_terms_rejected: passing9
billing_finality_delivery: passing / partial / untested / no record (one pass is the author's company's agent)2 / 5 / 11 / 17
vendor_metric_accountability: passing / untested / no record (fictional attention vendor)3 / 15 / 17
Public member profiles159
Verification-vendor domains serving a brand.json with agents (12 checked)0
AdCP compliance specialisms at 3.3.0-beta.5; for measurement or verification32; 0

The two production claims

Two sources claim production use of a verification verdict in agent buying, and both are the vendors' own . DoubleVerify's blog of 24 July 2026 says that "Today, Butler/Till is using Claude as its buyer agent" to activate media buys through AdCP, connected to DoubleVerify's platform over MCP, while PubMatic's AgenticOS applies the selected DoubleVerify protections to inventory packages before the auction . It names the agency but no advertiser, campaign, date range, spend or transaction record, and the agency's own words, found only in DoubleVerify's June release, name no buy or campaign . Taken at its word, it describes pre-bid exclusion, which changes what is bought, not what is paid (Chapter 4), and it does not say DoubleVerify's criteria travel in an AdCP field .

The original source screen did not corroborate the specific DoubleVerify–AdCP workflow . An expanded search on 11 November found Digiday's 24 June report that Butler/Till was developing Claude agents with DoubleVerify to analyse inventory and steer campaigns away from poor-quality placements . Digiday's March report names Geloso's December 2025–January 2026 flights and compares quality with a DoubleVerify benchmark; Marketing Dive and MediaPost relay the parties' claim that less than 1% failed DoubleVerify standards . These reports add campaign and buyer context, but do not establish that an actual DoubleVerify verdict determined the agent's purchase, which metric was used, or whether it travelled in the protocol. The later DV workflow could describe a different buy. The sources are counterevidence to a blanket absence of other-party reporting, not a transaction record settling H7.

Zefr's alpha is the second claim: its page says partners are "activating YouTube campaigns through Zain", which suggests live campaigns, but it names no partner, campaign or transaction and does not say Zefr's suitability exclusions apply in them .

H7, coded by the rule

H7 remains unresolved under the paper's recorded interpretation . The source record establishes schemas, connectors and tests; the phrase in H7's statement that use stops at schema representation is already too broad . Production requires the missing connection between a verdict and a live decision. Newly found buyer reporting narrows the evidentiary gap but does not by itself supply that connection . No identified financial-authority agreement was found in the screened record .

The figure sets out the coded states for each protocol, verdict type and vendor interface, with the agreement field beside them .

VERIFICATION VERDICTS IN AGENT BUYING: THE FIVE PROTOCOL STATES AND FINANCIAL FORCESchemaConnectorInteroptestProductionuseTransactionFinancialforceAdCP: viewabilityyesyesdraft/schemaAdCP: invalid trafficyesyesdraft/schemaAdCP: brand safety rateyesyesdraft/schemaAdCP content standards (non-verifier tesyesyesyesAdCP: attention scoreyesdraft/schemaAAMP OpenProposal: viewability guaranteedraft/schemadraft/schemaAAMP ARTF: pre-impression fraud and viewyesyesAAMP buyer and seller agents: brand safedraft/schemaDoubleVerify DV Neura: pre-bid brand suin/avendor saysvendor saysIAS Agent: brand safety and suitability HUMAN: ad fraudZefr Zain: brand suitability exclusions n/avendor saysvendor saysApostra Interchange: IVT and brand safetyesyesPeer39/Adloox, Channel Factory: allPixalate: Analytics and Enrichmentvendor saysDocumentedVendor assertion onlyDraft or schema field onlyNone foundNot applicableThe author co-leads AdCP’s Signals & Measurement working group, a declared interest. AdCP’s own measurementtaxonomy says the protocol does not adjudicate between verification vendors. Read 10 November 2026.
Verdicts are schema fields in agent buying; no transaction and no binding agreement were found. For each protocol and vendor agent interface, the five protocol states of shared rule 12 (schema representation, implemented connector, interoperability test, production use, demonstrated transaction) and whether an identified agreement gives a verification verdict financial force. The only production claims are vendor assertions.Units: coded state · Scope: AdCP, AAMP and the scored vendors’ agent interfaces, to 10 November 2026 · Sources Membership and schema fields earn no deployment credit; the states are status-sensitive.Data: data/figure-data.json#protocols

The rule is ambiguous because DoubleVerify names an agency buyer, while its unresolved clause refers to a vendor claim without a named buyer, campaign or transaction record . The primary reading requires more than that agency name to establish a verdict in production. The author reads it that way: an agency name without a verdict-to-decision record is still a vendor assertion. Zefr still names no alpha partner .

What would settle it

H7 would be contradicted by a documented live buy showing which verification verdict changed the agent's action, with a named campaign or transaction and its dates . A delivered and billed AdCP package naming the governing verifier would be one route. Financial authority additionally requires the identified agreement making that count determine an invoice or remedy .

The same parties could settle it the other way. If Butler/Till, PubMatic or DoubleVerify described the agent-driven buys without a third-party verdict applied, or Zefr's alpha proved to carry no suitability verdict, and no other source documented state 4 or 5, H7 would be supported as registered . The recheck at the freeze, on 11 December 2026, covers AAO's registry counts, the brand.json files, AdCP 3.3's changes, the disposition of OpenProposal's comments and the availability of DoubleVerify's Activation Agent .

Chapter 11

The Verdict Quotient

The Verdict Quotient (VQ) summarises the public record of a product's documented capability within its buying task. Independent detection evidence is the lowest-scoring dimension in the suite and invalid-traffic cohorts; accreditation is lowest among suitability specialists. Those are gaps in the evidence a buyer can inspect. They do not measure which service makes the fewest errors.

In short

The matrix scores 17 products on 12 dimensions, subject to each cohort's applicability mask. The field VQ is 60 for suites, 53 for invalid-traffic specialists and 39 for suitability specialists; these are separate reference points, not a comparison between cohorts. Only cohort C meets the minimum size for bands. It has 0 products in Most documented, 3 in Well documented, 3 in Partly documented and 1 in Least documented. Of its banded rows, 4 retain their band under every tested reweighting. Missing-evidence scenarios change 6 bands. The quotient describes disclosure and documented capability, without establishing detection accuracy.

What the quotient does

The dimensions give each attribute one home. Depth covers method disclosure, independent detection evidence, invalid-traffic method, safety and suitability method, and verdict uses and authority. Reach covers environment coverage, platform access, measured scale and US availability and use. Standing covers accreditation, interest disclosure and safeguards, and the dispute and correction process. Each dimension has published anchors from 0 to 5; the applicable set depends on the product's cohort. Chapter 13 publishes the rules and their limits.

The group means use counted cells only. An n/e cell means evidence was not established; n/a means the dimension falls outside the cohort's mask. Neither becomes zero. Raw = 0.45 Depth + 0.25 Reach + 0.30 Standing, with weights renormalised if a group is absent. The separate coverage field shows how much evidence underlies that average. A fuller record can lower an average when the additional evidence supports a low anchor.

The calculation pulls Raw toward the cohort field mean: Shrunk = (k × Raw + K × F) / (k + K). Here k is the row's counted-cell total, F is the mean Raw of admitted cohort rows, and K is 4. VQ = 100 × Shrunk / 5. K is the author's heuristic for moderating thin records, not a parameter estimated from observed error rates. Shrinkage leaves missing cells missing; it supplies no evidence about them.

Bands use the unrounded offset 20 × (Shrunk − F). Most documented begins at +12, Well documented at +1, Partly documented at −8, and Least documented below −8. Equal offsets receive the same band. Display rounding does not decide membership. A row needs 6 counted cells with every applicable group represented; a cohort needs 6 eligible rows before any bands appear. Cohorts A and B each contain 5 rows and are therefore shown without bands. The cohort gate applies even when an individual row has a complete record.

The results by buying task

VERDICT QUOTIENT (0–100) BY COHORT, WITH THE RANGE UNDER FOUR REWEIGHTINGSA · field 60 · no bands (5 eligible)No vendor in this cohort is banded.B · field 53 · no bands (5 eligible)No vendor in this cohort is banded.C · field 39Most documentedWell documentedLeast documentedZefr — Zefr brand safety and suitability47 · 10/11 counted, band movesSightly — Sightly Brand Mentality41 · 8/11 counted, band moves, ± missingProximic by Comscore41 · 10/11 counted, ± missingPeer39 — Peer39 brand safety and suitabil…39 · 11/11 counted, ± missingChannel Factory — ViewIQ39 · 10/11 counted, ± missingPixability — Pixability brand suitability37 · 10/11 counted, ± missingilluma — illumaSentinel29 · 9/11 counted, band moves, ± missing2030405060VQ (dashed line: each cohort’s field score)Grey bar: range under the four reweightings. Not plotted, A to Z: Adloox — Adloox verification (cohort too small), DoubleVerify — DoubleVerifyadvertiser ve… (cohort too small), Integral Ad Science — IAS advertiser veri… (cohort too small), mFilterIt — mFilterIt Ad Traffic Validati…(cohort too small), Protected by Mediaocean (cohort too small), Escalated — Escalated IVT scoring (cohort too small), FouAnalytics (cohort toosmall), Fraudlogix — Fraudlogix IVT detection (cohort too small), HUMAN — HUMAN Ad Integrity Suite (cohort too small), Pixalate — Pixalate adfraud protection a… (cohort too small). Their scores are in data/vq.csv.
Read the band, not the rank: no product clears the most documented line, and ten rows are not banded. The Verdict Quotient for the products with enough evidence to band, one section per cohort. The Verdict Quotient uses the band offsets of the author’s Contextual Quotient, but its field mean is the cohort’s mean Raw score rather than a pooled group mean, so the two do not read identically; neither construction validates 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 1 places within a cohort, and 4 of the 7 banded products keep their band under every weighting. VQ measures publicly documented capability within a cohort, not product performance.Units: VQ points on a 0–100 scale, pulled toward the cohort field mean; shown rounded, plotted unrounded · Scope: 7 banded products of 17 in the capability matrix, scores from evidence read in November 2026; recheck on 11 December 2026Single 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/vq.csv

Cohort A contains full verification suites, cohort B impression-level invalid-traffic (IVT) specialists, and cohort C separately purchased suitability services. The field values are 60, 53 and 39, respectively. A field value is a reference computed from that cohort's admitted rows. It is neither a market average weighted by spend nor a benchmark for a different buying task. A suite's broader mask must not turn into a claim that it outperforms a specialist.

In cohort C, the displayed lines are 51 for Most documented, 40 for Well documented and 31.1 for Partly documented. The calculation applies the full unrounded lines. No row reaches Most documented under the primary settings. The table groups products by band and uses alphabetical order within each band, avoiding a league table.

Cohort C productPrimary bandSame band under all reweightingsMissing-evidence flag
Proximic by ComscoreWell documentedYes±
Sightly Brand MentalityWell documentedNo±
Zefr brand safety and suitabilityWell documentedNoNone
Channel Factory ViewIQPartly documentedYes±
Peer39 brand safety and suitabilityPartly documentedYes±
Pixability brand suitabilityPartly documentedYes±
illumaSentinelLeast documentedNo±

The flags answer different questions. Reweighting stability asks whether changing the importance of the groups moves a band. A published sensitivity flag must cover a band change or eligibility loss under any single-cell deletion, as well as a band change under either missing-evidence scenario. The table separates these tests; six of the seven cohort C rows are deletion-sensitive under the current scores (removing Peer39's detection-evidence cell moves it from 39.38 to 42.05, and Pixability's accreditation cell from 37.29 to 40.11, both into Well documented), four are scenario-sensitive, and Zefr is affected by neither. Neither test is a confidence interval. A row can keep its band under every weighting and still carry ±, as Proximic, Channel Factory, Peer39 and Pixability do; Sightly and illuma also move under at least one weighting. The matrix exposes the underlying cells and their reasons rather than asking a reader to rely on the label alone.

The capability matrix

Each cell scores one documented capability from 0 to 5 against fixed anchors; n/e means no evidence, n/a means the dimension does not apply to the cohort. Click a cell for its rationale, availability, evidence grade, date, sources and verification status. Rows are compared only within their cohort, and a score is documented capability, not accuracy.

5 documented in operational detail3 real but thin or narrow1 adjacent or incidental0 documented absencen/e insufficient evidencen/a outside the cohort’s maskunverified: shown, counted as n/eβ beta · ann announced · ? unconfirmed · ret retired · ± provisional band: it moves when missing evidence is filled
ProductVQMethodDetectionIVTSuitabilityAuthorityEnvironmentsPlatformsScaleUSAccreditedInterestsDisputesCounted
A · Full verification suites
no bands in this cohort, A to Z
AdlooxAdloox verification · Verification · operating; acquired by Peer39 (June 2026, completion recorded in a UK filing); brand and service retained; MRC accreditation continued on 29 June 202657Cohort too small12/12
DoubleVerify ◆DoubleVerify advertiser verification · Verification · active, listed; merger with a subsidiary of Neptune BidCo US Inc. (Nielsen's parent) pending at 10 November 202668Cohort too small12/12
Integral Ad ScienceIAS advertiser verification · Verification · active; privately held portfolio company of Novacap-managed funds; no longer files periodic reports66Cohort too small11/12
mFilterItmFilterIt Ad Traffic Validation · Verification49Cohort too small11/12
Protected by MediaoceanVerification61Cohort too small9/12
B · Invalid-traffic and fraud specialists
no bands in this cohort, A to Z
EscalatedEscalated IVT scoring · Fraud49Cohort too small9/10
FouAnalyticsFraud46Cohort too small9/10
FraudlogixFraudlogix IVT detection · Fraud46Cohort too small10/10
HUMANHUMAN Ad Integrity Suite · Fraud · active; private63Cohort too small9/10
PixalatePixalate ad fraud protection and analytics · Fraud · active; private62Cohort too small10/10
C · Content safety and suitability specialists
banded, by VQ
ZefrZefr brand safety and suitability · Contextual · active, private47Well documented10/11
SightlySightly Brand Mentality · Contextual41±Well documented8/11
Proximic by ComscoreContextual41±Well documented10/11
Peer39Peer39 brand safety and suitability · Contextual39±Partly documented11/11
Channel FactoryViewIQ · Contextual · active; private39±Partly documented10/11
PixabilityPixability brand suitability · Contextual37±Partly documented10/11
illumaillumaSentinel · Contextual29±Least documented9/11
How VQ is built. The Verdict Quotient uses the band offsets of the author’s Contextual Quotient, but its field mean is the cohort’s mean Raw score rather than a pooled group mean, so the two do not read identically; neither construction validates it. Three groups: Depth (method disclosure, independent detection evidence, invalid-traffic method, safety and suitability method, verdict uses and authority), Reach (environment coverage, platform access, measured scale, us availability and use) and Standing (accreditation, interest disclosure and safeguards, dispute and correction process). Only verified or qualified cells count; an unverified cell counts as n/e, and n/a comes only from the cohort’s applicability mask. Each group is the mean of the counted cells inside it, so an unscored dimension is not counted as a zero. Raw combines the groups at 45/25/30, renormalised over the groups present, and is shrunk toward the cohort’s field mean, the mean Raw score of the cohort’s admitted rows, thin rows included, with a prior strength of 4 against the row’s own count of counted cells. A cohort is a set of products doing a comparable job; each has its own field mean and band lines, and rows are never ranked across cohorts. The weights and the prior strength are author-defined heuristics. Bands come from the unrounded distance to the cohort’s field score (Most documented at +12 or more, Well documented at +1 or more, Partly documented at −8 or more, Least documented below that); only the displayed score is rounded. A row gets a band only with at least six counted cells and at least one in every applicable group, and only in a cohort with at least 6 such rows; the others show a score without a band and are listed apart, A to Z, with any missing group named. The three bars under each score are Depth, Reach and Standing in that order. VQ summarises publicly documented capability within a cohort, not product performance, and is not calibrated to buyer outcomes.

Read the band, not the rank. A · Full verification suites (5 rows, field VQ 60.2). Only 5 rows are band-eligible, under the 6 a cohort needs, so its scores and coverage are shown without bands. Deleting one counted cell at a time moves a score by up to 3 points up (Protected by Mediaocean, dispute and correction process) or 4 down (mFilterIt, interest disclosure and safeguards); 0 deletions would drop a row below admission and 0 would cost a row its band eligibility. B · Invalid-traffic and fraud specialists (5 rows, field VQ 53.2). Only 5 rows are band-eligible, under the 6 a cohort needs, so its scores and coverage are shown without bands. Deleting one counted cell at a time moves a score by up to 4 points up (Fraudlogix, accreditation) or 3 down (FouAnalytics, interest disclosure and safeguards); 0 deletions would drop a row below admission and 0 would cost a row its band eligibility. C · Content safety and suitability specialists (7 rows, field VQ 39.0). Band lines: Most documented 51.1, Well documented 40.1, Partly documented 31.1, Least documented below. Under four reweightings (equal thirds, depth-led, reach-led, standing-led), with Raw and the field mean recomputed, no product sits in the top band (Most documented) (Zefr joins it under reach-led weights), while positions move by up to 1 places and by 0.3 on average. Of the 7 banded products, 4 keep their band under every weighting (Channel Factory, Peer39, Pixability, Proximic by Comscore); illuma, Sightly, Zefr cross a band line under at least one. With a prior strength of 0 (Raw, no shrinkage), 2 and 8 instead of 4, positions move by up to 0, 0 and 0 places. Deleting one counted cell at a time moves a score by up to 3 points up (Sightly, environment coverage) or 3 down (Peer39, interest disclosure and safeguards); 0 deletions would drop a row below admission and 0 would cost a row its band eligibility. In the missing-evidence scenarios (every n/e cell of a banded row filled with 1, then with 4, with the field and every other row frozen), 6 of the 6 banded rows with n/e cells change band (Channel Factory, illuma, Peer39, Pixability, Proximic by Comscore, Sightly; marked ±). Missing evidence is not neutral: every deletion case is in data/vq-missing-evidence.json, and every row’s scenario scores, including a hypothetical 0-to-5 envelope reported separately and never used for the flag, are in data/vq.csv. These tests cover weights, prior strength and missing cells, not evidence grades, the scorer or the product sample. Every input is in the CSVs in the appendix, so the index can be recomputed on different weights.

Mantis Sentinel was listed rather than scored after the final review: the scored-set rule requires a product sold apart from media and live or in beta, and its advertiser page establishes neither. Dropping it raised the cohort C field mean and moved Channel Factory and Peer39 from Well documented to Partly documented.

Where the field is weak and strong

Computed across numeric cells in the supplied matrix, independent detection evidence has the lowest dimension mean in A and B. US availability and use has the highest in each. Within A, platform access and environment coverage are also relatively well documented; the dispute process is weak. Within B, accreditation follows detection evidence among the weakest dimensions, while environment coverage follows US use among the strongest. A missing cell is excluded from each dimension mean, so the dimensions can have different contributing populations.

In cohort C, accreditation has the lowest mean, followed by dispute and correction process, then independent detection evidence. Interest disclosure and safeguards has the highest mean, followed by US availability. The interest cells credit disclosure of sell-side relationships. They do not establish that those relationships are harmless or that the vendor has a complete separation policy. Each score means only what its anchor says, which prevents a familiar label such as trust from swallowing distinct questions.

The low detection cells fit the bounded study result: no independent test of a scored product's sold category with a labelled test set and published denominator was found in the sources screened to 10 November 2026 . The low accreditation cells have a different basis, the match between documented sold units and MRC's scoped record . An error-rate test and an accreditation letter answer separate questions. Neither the quotient nor its strongest dimension supplies the missing answer.

How stable the bands are

The build recomputes the index under equal group weights, Depth-led 60/20/20, Reach-led 30/45/25 and Standing-led 30/20/50, recalculating Raw and F. In cohort C, 4 of 7 rows keep their band under all variants. Zefr reaches Most documented only under the Reach-led alternative. Sightly falls to Partly documented under the Reach-led and Standing-led alternatives, and illuma rises to Partly documented under the Standing-led one. These are consequences of the chosen weights, not new observations of the products.

The supplied reweighting statistic reports maximum within-cohort position changes of 0 in A, 1 in B and 1 in C. That diagnostic tests numerical movement; the paper makes no product ranking from it. In A and B, zero band changes merely reflect the absence of bands. They cannot establish stability of a distinction those cohorts do not publish.

Prior strengthA band changesB band changesC band changes
K = 0000
K = 2000
K = 8001

K = 0 removes shrinkage and reports Raw. The other alternatives test weaker and stronger pulls toward F. None changes a source, anchor or applicability mask. The results show how much of a band depends on the aggregation choice rather than on a new fact about a service.

Single-cell deletions test a different dependency. Across 55 deletions in A and 47 in B, the supplied outputs record 0 and 0 admission losses, and 0 and 0 eligibility losses. Their band-change counts are 0 and 0, subject to the small-cohort caveat above. In C, 17 of 68 deletions change a band, 0 lose row eligibility and 0 lose admission. The summary does not separately report loss of the cohort band gate; no such total is inferred here.

Deleting a low cell can raise a quotient. In the supplied C example, deleting Sightly's environment coverage cell, scored 1, raises its displayed VQ from 41 to 44. That is an evidence-loss scenario, not an improvement in the service.

Missing-evidence scenarios instead fill a row's applicable n/e cells with 1, then 4, while holding F and every other row fixed. They change 0 bands in A, 0 in B and 6 in C. Imputation creates neither admission nor eligibility. These are scenarios, not bounds; the separate endpoint envelope uses 0 and 5. Lack of a band-change flag in an unbanded cohort is not evidence that missing cells have no numerical effect.

What the quotient is not

The VQ measures documented capability under the author's anchors. It supplies no estimate of accuracy, return on spend, legal safety or suitability for a particular contract. The authority ladder asks how far a verdict travels toward money. It does not imply that a reporting product has less practical value than a blocking product for every buyer. Legal allegations and incident records stay outside the index.

Claude agents scored the initial calibration rows with separate refute-mode checks. An OpenAI model scored the later rows, then self-checked them in a refute-mode pass; a supplemental consistency review also inspected those rows. The author checked the work and retained the primary scoring calls. There was no human second scorer, blind second score or inter-rater statistic. A second pass by a row's own scorer can correct a mistake while sharing the assumptions that caused it. Chapter 13 records that distinction.

The profiles

The profiles connect the purchased service to its entity, owner, history, method claims and sources. Read the profile with the matrix cell and the dated scope record before making a buying decision. A parent company's capabilities do not transfer automatically to a subsidiary's product, and a platform partnership cannot replace an unstated data route. The profiles also preserve contrary evidence and access limits, which a composite number cannot carry on its own.

Deep profiles — select a company

Each card shows the matrix scores as a bar strip. Open a company for its SWOT, what it sells, who buys it, how it earns, its verified capabilities with availability and evidence grade, performance evidence with sponsorship, limits, risks and fit. Pick a SWOT quadrant to isolate it and compare every profiled company on it. Each SWOT line is tagged record, vendor documentation, vendor claim or inference, and opportunities and threats are the author’s reading of dated facts.

SWOT

Chapter 12

What a buyer should ask for

Before relying on a verification verdict, ask for three separate records for the exact service bought. An accreditation letter establishes its audited scope, a labelled test establishes an error estimate, and the contract states any deciding role in payment. The screened record to 10 November 2026 supplied no independent error rate for a scored product. Two of sixteen readable platform and exchange terms gave a third-party report determining effect, within limited contractual scopes. A buyer's executed contract may differ.

In short

Media Rating Council (MRC) accreditation establishes standards conformance within scope; the independent studies screened supplied no sold-category error rate for a scored product . Two readable terms gave third-party reports determining effect, with no settlement found . Platform data routes and accredited connected-TV (CTV) metrics require separate checks . Seller revenue and ownership call for disclosed safeguards; the DoubleVerify separation statement found is limited . Product pages naming GARM identify no dated implementation . Agent-workflow reporting still leaves the verdict-to-decision link and financial agreement unestablished .

Eight questions, chapter by chapter

Statuses were read by 10 November 2026 and will be rechecked at the freeze on 11 December 2026.

First, ask for MRC's letter for the exact unit bought: product, metric, environment, format, placement, language and standard (Chapter 2). Several public terms that give a third-party report a role in payment, among them The Trade Desk's OpenPath clause and PubMatic's fraud credit, ask for an MRC-accredited service . MRC names the letter as the authoritative reference, and the list does not always match: IAS's property-level row has no letter on MRC's site . Letters also narrow scope: Zefr's content-level accreditation covers English-language in-stream video and music video on YouTube, not Shorts . Peer39's verification page implies accredited pre-bid viewability that MRC does not list .

Second, request the miss or false-positive rate for the bought category, the labelled test set, settings, period and denominator (Chapter 3). The independent screen found cases and agreement studies without a qualifying error rate . Protected-versus-unprotected comparisons measured by the protecting vendor cannot supply that test on their own . Ask who assigned the labels.

Third, read who decides, pays and triggers a fee adjustment (Chapter 4). OpenPath makes accredited reports conclusive for The Trade Desk's publisher payments, while Microsoft announced the end of Invest's partner-report refund route . PubMatic retains a good-faith decision on fraud credits . H2 is supported on its registered test, the listed platform terms; the wider contract population is reported as sensitivity (Chapter 4) . IAB's May 2026 terms leave controlling measurement to the order .

Fourth, on a large platform, ask whose data the verdict is made from (Chapter 5). A verifier given only the platform's counts cannot detect invalid traffic on devices it never sees, a limit MRC records for IAS on Meta . Google accepts no third-party pixels for YouTube measurement, and Amazon says the verifiers' detection runs inside its environment . On Meta and YouTube, MRC accredits the verifiers' viewability net of general invalid traffic (GIVT) only and the platforms' own measurement including SIVT . TikTok is the exception: DoubleVerify and IAS said in 2021 and 2022 that their tag ran there through a measurement kit TikTok controls; no 2026 page restates it .

Fifth, on CTV, ask whether MRC accredits the metric bought there, and for what share of the buy (Chapter 6). CTV SIVT filtration is accredited for five scored products, but CTV video viewability only for DoubleVerify and IAS, whose viewable impressions count certified apps only, and neither letter says what share of the vendor's CTV traffic those apps carry . Content-level accreditation on CTV is Zefr's alone, for YouTube, and the program-level signals vendors sell there carry none . MRC's 2021 guidance does not require detection of ads played while the TV is off, though it asks vendors to disclose the limit .

Sixth, ask who else pays the verifier, and for the written policy that keeps those payers apart from the verdict (Chapter 7). Sellers brought DoubleVerify 8.6% and IAS 14.4% of FY2024 revenue, IAS's line including Publica's CTV ad serving . IAS's terms let a media partner pay an advertiser's verification fees, so a seller can pay for the check on its own inventory . DoubleVerify's separation page of June 2024 names no scope, owner or enforcement and does not address its agreed sale to Neptune BidCo US Inc., Nielsen's parent, which had no closing on record on 10 November 2026 . The author works at Samba TV, which competes with Nielsen in TV measurement. Seller revenue and ownership are governance attributes, not evidence of bias .

Seventh, request the implemented framework, dated edition and maintainer (Chapter 8). The WFA records June 2022 as GARM's latest update, while the product pages screened name no implemented edition . MRC's content-level supplement lets buyers choose risk tolerance and defines no tiers of its own .

Last, ask which protocol state is documented and which agreement gives the verdict financial force (Chapter 10). AdCP's thresholds can name the verifier, with the author's working-group interest declared . DoubleVerify and Zefr make production claims; new reporting partly corroborates Butler/Till's quality-analysis workflow without proving that a verdict drove the agent's purchase . The screened record establishes no such financial-authority agreement .

The decision tool pairs each question with the evidence that would answer it.

Before a verdict settles anything: what to ask for

Pick the question a buyer is asking. The panel shows the evidence that would answer it, what that evidence assumes, where it fails and what it still cannot tell you.

What would settle the open questions

Chapter 13

Method

The paper was registered before it was researched: its hypotheses, scoring rules and anchors were fixed on 9 November 2026 after an outside AI review of the brief whose factual points were checked at their primary sources. Claude agents gathered the workstreams and profiles and scored the calibration rows with separate refute-mode verifiers. An OpenAI model completed the later sweep checks, scored and self-checked the added rows, and fact-checked the chapters. The author checked the work and made every inclusion, cohort and verdict call. Sources were read on 9–11 November 2026; status-sensitive facts are due for recheck at the planned freeze on 11 December. The process supplies no human second score or audit.

In short

The brief went through the author's own fact check on 2 November 2026 and an outside review by an OpenAI model on 4 November (144 entries); the hypotheses and scoring rules were fixed on 9 November, before any research agent ran. Claude research agents produced 10 workstreams, a vendor-universe sweep, 7 company profiles and 7 analyses of strengths, weaknesses, opportunities and threats (SWOTs). Separate Claude verifiers, told to refute, checked the workstreams, initial sweep and deep profiles; an OpenAI model checked the later sweep files. Across the verifier records, of 1125 checks, 696 found the item supported as stated, 420 supported it with a qualification, 9 contradicted it and 0 left it unresolved. The research registered 1458 source records, 59 of them readable only in part, and 804 claims. Products are scored on 12 dimensions: Claude agents scored the calibration rows with separate refute-mode checks; an OpenAI model scored the added rows and then self-checked them in refute mode, and the Verdict Quotient (VQ) summarises documented capability within a cohort in bands, never ranks. Executed contracts, log-level data and vendors' test sets are private, and no vendor or third party was contacted . The author works at Samba, a Nielsen competitor in TV measurement, and co-leads a working group of AgenticAdvertising.org, which maintains the Ad Context Protocol (AdCP).

Fixed before the research

Brief v1 was drafted on 2 November 2026 from the author's published record. The same day, nine checker agents in the author's AI-assisted process, not independent or external reviewers, each opened the primary sources behind one group of its dated facts, and v2 corrected what they found.

On 4 November 2026 the author ran an outside review of v2 by an OpenAI model, which returned 144 entries. Before the brief changed, nine more checker agents re-opened the primary sources behind 87 of them: 71 were confirmed, 15 were right only in part and one was wrong, a challenge to the date White Ops renamed itself HUMAN, 30 March 2021, which the original release confirms . The checkers also caught errors the reviewer had missed: "100% brand safe" is the phrase of Adalytics' August 2024 report , and the letter the Media Rating Council (MRC) now links for the property-level row of Integral Ad Science (IAS) does not cover ad verification .

Brief v3, dated 9 November 2026, records each disposition and changed the method as well as the facts. The hypotheses gained numeric decision rules: H1 was split into accredited coverage and claims against scope, H2 into four components with first-party validation recorded apart, and H3 was coded by the estimand each study supports. All twelve scoring dimensions were rewritten so that each attribute has one home, with anchors for every value from 0 to 5, and the legal and incident record moved outside the quotient. Three of the reviewer's scoring proposals were adopted in a different form: a credible failed test still scores below marketing claims, though a 0 now needs a labelled test with a denominator; the authority dimension keeps its ladder toward settlement, as the paper's question; and non-disclosure scores 1, while ownership earns or loses nothing on its own.

The hypothesis register and the scoring rules were fixed on 9 November 2026, before any research agent ran or any cell was opened. The register holds 12 entries: hypotheses H1a, H1b and H2 to H7, each with its population, comparator, observable outcome and rules for supported, contradicted and unresolved, and questions Q1 to Q4, each with a coding rule. Absence of accessible evidence yields unresolved, never contradicted. The product register, seeded from the brief and completed by the sweep, holds 45 records, 19 of them listed and 9 historical.

How the evidence was gathered

Claude research agents gathered the original evidence under the author's direction on 9 and 10 November 2026. The OpenAI run-1 work on 10–11 November added sweep checks, scoring and chapter fact-checks; the author checked the combined work. Working from the shared rules, the brief and the registers, each wrote one structured file: sources with dates, funding and access limits; findings with the supporting passage, evidence class, confidence, limitations and counterevidence; and data tables whose rows name their sources. The rules require every cited source to be opened, treat a search snippet as a lead, and settle each claim from the source suited to its type, such as filings for ownership and the accreditor's record for accreditation. Where a page refused automated fetches, agents retried with a browser user agent or in a browser and recorded the route; X's help pages were read in a browser .

10 workstreams, A to J, each took one part of the brief, and 12 vendor-universe sweep, U, named every company and product it could find.

StreamWhat it readBears on
AMRC's lists, letters and standards; vendors' accreditation claims; certification and framework bodiesH1a, H1b, H6, Q4
BThe Association of National Advertisers' (ANA's) 2023 study, Adalytics' reports and the vendors' answers, academic work, platform transparency reportsH3
CVendor and commissioned claims of detection, accuracy and savings, with sponsor, design, sample and periodH3, kept apart from B
DFilings, measured volume, ownership, funding and consolidation since 2021H5, Q3
EBilling, credit and remedy terms on H2's fixed list, and public contractsH2
FWhat third-party verification can see and do inside the nine large platformsH4
GWhat is measured and accredited on connected TV (CTV) and in-appQ2, Q4, H1a; research question 10
HLetters, court filings, regulator actions and independent reports since 2023Q1
IWhat buyers, agencies and their bodies document about verification use, fees, contracts and disputesH2
JVerification verdicts in AdCP, the Agentic Advertising Management Protocol (AAMP) and vendor agent interfacesH7
UEvery named company and product, for the product registerThe cohorts

7 deep profiles covered the companies the brief chose for market weight and depth of public evidence: DoubleVerify, IAS, HUMAN, Pixalate, Zefr, Channel Factory and Adloox, which Peer39 bought from Scope3 in June 2026 . Claude agents produced those profiles with separate refute-mode verifiers. Each records the entity, its products, buyers, revenue model, financials and timeline, and a capability record per scoring dimension; 7 SWOTs summarise them, each point tied to its sources and labelled as a record, vendor documentation, a vendor claim or inference. A second-wave sweep opened the remaining candidate products' documentation. An OpenAI model completed the later sweep verifications and scored and self-checked the added rows; its chapter fact-checks and the combined work were checked by the author.

The research registered 1458 source records, including those the verifiers added: 798 vendor pages, documents and releases, 98 filings and 36 studies among them. 59 could be read only in part or as an abstract, and claims resting on them are limited to what could be read. Each finding became one claim in the ledger, with one evidence class; a vendor's documentation is a vendor assertion however technical, a study of unknown funding is not treated as independent, and the author's essays never count as corroboration.

Evidence classWhat it coversClaims
Direct recordA filing, registry entry, court record or other official document272
Independent measurementMeasurement by a party with no financial interest26
Commissioned measurementMeasurement paid for by an interested party10
Vendor assertionThe vendor's own pages, documentation and releases255
SynthesisA reading that combines sources209
InferenceA reasoned conclusion, marked as such28
ForecastA party's statement about the future4

How it was checked

In the verifier pass, the original Claude workstream files, universe sweep and deep profiles went to separate Claude verifiers told to refute them. An OpenAI model completed the later capability-sweep verifications. These file checks differ from the added score rows, whose scorers performed their own refute-mode second pass. The verifier re-opened the cited sources, checked each number's denominator, unit, date, entity and version, and returned one verdict per item, with the correction, the source's wording or the reason wherever the item fell short. Verifiers made 1125 checks of findings, profile capability records, profile fields and table rows: 696 supported as stated, 420 with a qualification, 9 contradicted and 0 unresolved. The figure traces each stage.

THE PIPELINE, AND WHAT EACH STAGE PRODUCED OR REMOVEDResearch10 workstreams12 universe sweep7 vendor profilesshared brief and schemaRegistered1458 sources opened804 claims in the ledger59 sources partly accessible2 author works, labelledVerified792 research claims checked542 as stated247 with qualification3 contradicted4 not checked · 8 synthesesScored17 products12 dimensions each171 cells scored (84%)16 n/e · 17 n/ascorer + cell verifierReviewed20 corrections logged0 claims withdrawn1 scoring correctionsWhat the process is not: two agents reading one source are not two measurements; a model checking a model is not an audit; a cellverifier re-reading the scorer’s sources is not a human inter-rater test.Every count on this figure is computed from the ledgers at build time (data/stats.json); the prose quotes the same values.Research cutoff 2026-11-11. Claims outside the verifier sample are marked unresolved with a note; they are not counted as supported.
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 cutoffCounts describe process volume, not evidence quality; a claim that survived verification is supported by its source, not proven true.Data: data/stats.json

Each claim is printed in its final wording, the verifier's correction where there is one. Of the 804 claims, 792 were checked: 550 stand as stated, 247 with a qualification and 3 were contradicted, which are cited only to say so. The 4 unresolved claims rest partly on HUMAN press releases that refused the verifier's fetches; the parts it re-opened held, and the rest is cited only as unresolved.

The contradicted claims show what the pass is for. The verifier contradicted stream C's top of the vendor-reported invalid-traffic range, 39%, with a 46.1% row in the stream's own table . It contradicted stream I's reading that MRC's 2018 content-level supplement left error disclosure optional, since the supplement makes several such disclosures mandatory . It also contradicted the sweep's proposal to list Pixability as bundled with its media, having found a self-service model sold apart . The verifiers had limits too: one had used up its web-search budget and re-bounded absences only through pages it opened directly, and another could not repeat a profile's browser reads of X and Snap . A model checking a model is not an audit, and two agents reading one source are not two measurements.

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

Claim ledger

The 426 claims marked in the text, each with its evidence class, review outcome and confidence. Click an ID to see the supporting passage, scope, derivation and limitations. Headline claims are listed first. The full research ledger of 804 claims, including those not used in the text, is in data/claim-ledger.csv.

IDClaim (final wording)Ch.Evidence classReviewConfidenceSources
Integral Ad Science's filed revenue by channel shows Publisher revenue of $62.8 million (FY2022), $63.8 million (FY2023) and $76.5 million (FY2024) against total revenue of $408.3 million, $474.4 million and $530.1 million, that is 15.4%, 13.5% and 14.4% of worldwide revenue, with Optimization at $190.6 million, $224.5 million and $242.6 million.1, 7direct recordsupported as statedhigh
Property-level ad verification is accredited for DoubleVerify on desktop, mobile web, in-app and CTV (letter of 13 April 2026), with accredited languages varying by level (about 176 for binary classification, 56 at domain level, 41 at page level and 10 on CTV after French was added in April 2026), and for Protected Media on desktop and mobile web only (initial grant 25 September 2025: display reporting and blocking, video reporting only, English, Spanish and Portuguese, excluding all other languages, gambling and fake news/misinformation categories, IVT ad verification, custom bundle lists and rules, sentiment and non-matching language).2direct recordsupported with qualificationhigh
Eight services on MRC's list hold SIVT accreditation that includes CTV, five of them scored candidates (DoubleVerify, IAS, HUMAN, Pixalate, Protected Media), each with its own CTV limit (IAS certified applications only; Protected Media video impressions by non-SDK measurement; HUMAN's row "applies to backend detection and IVT predictions"), while Adloox's SIVT accreditation covers desktop and mobile web only, its in-app metrics being net of GIVT.2direct recordsupported as statedhigh
On MRC's accredited and under-review digital lists read on 10 November 2026, four of the five cohort C candidates (Channel Factory's ViewIQ, Peer39's brand-safety product, Proximic by Comscore and illumaSentinel) have no entry for any metric or environment; Peer39's only accredited service is the separately scored Adloox measurement service, Comscore's own page claims MRC accreditation only for Comscore TV, and Zefr's accreditations are confined to YouTube.2direct recordsupported as statedhigh
Inference: on MRC's records at 10 November 2026 and the unit grid used in this file, cohort C's accredited share of sold units is about 44% (7 accredited of at least 16), below H1a's 50% cohort floor, because only Zefr holds accreditation while the four other products' own pages document at least 9 sold units MRC does not list; the figure depends on the grid: counting Zefr's ADH reporting as one YouTube unit, as the file does for DoubleVerify and IAS, gives 5/14 (36%), counting GIVT units for Zefr's ADH reporting while holding the others at one family gives 10/19 (53%), and counting every family the other products' pages document keeps the share below 50%.2inferencesupported with qualificationmedium
Protected Media's home page (read 10 November 2026) says it is accredited for "Rendered Display and Video Ad Impressions across Desktop, Mobile Web, Mobile In-App, and Connected TV", while MRC's 22 September 2026 letter accredits display impressions on desktop, mobile web and in-app only and, on CTV, video impressions by non-SDK measurement only; read literally the claim extends display impressions to CTV.2synthesissupported as statedmedium
DoubleVerify's MRC one-sheet (PDF created 8 April 2026) lists its pre-bid property-level brand suitability avoidance segments and its Authentic Brand Suitability segment against desktop, mobile web, in-app and CTV, while MRC's letter of 13 April 2026 accredits the pre-bid "Brand Safety and Suitability" segments on desktop, mobile web and in-app only and limits pre-bid CTV accreditation to Fraud/IVT, Fully On-Screen, app-store categories and site and app lists, the last two being the brand-suitability components the one-sheet's own footnote 2 lists by environment.2synthesissupported with qualificationmedium
DoubleVerify's one-sheet discloses limits that neither the 13 April 2026 letter nor DV's product pages state (Authentic Ads are not accredited when a campaign's settings include geographic targeting; segments and categories released after May 2024, made-for-advertising and Low Quality GenAI are not accredited) and disclaims content-level brand safety, while three DV product pages still call DV Authentic Ad "a proprietary, MRC-accredited metric" defined to include the intended geography.2vendor assertionsupported as statedhigh
The other accreditation claims read on scored products' own pages and releases on 10 November 2026 stay within MRC's listed scope (DoubleVerify's IVT, viewability and property-level claims; IAS's SIVT claim; HUMAN's two 2026 releases; Pixalate; Protected Media's company page; Zefr; Comscore), while Channel Factory, illuma, Fraudlogix and Escalated make none; two further claims are ambiguous and belong with F12 and F13: IAS's 'MRC-accredited' Quality Impression, defined to require a suitable environment and the correct geo, although its July 2026 platform letter excludes ad verification and lists no brand-safety or geo metrics, and HUMAN's MRC solution brief, which calls its pre-bid accreditation 'mitigation' and 'protection' where MRC accredits detection, not prevention.2synthesissupported with qualificationhigh
MRC's Policy for Property-Level Ad Verification Representations (18 October 2025) bars accredited property-level services from positioning themselves as "Brand Safety" or content-level without content-level capability in the environment, after a six-month grace period (to about 18 April 2026), while allowing suitability language with disclosures; DV's 13 April 2026 letter still names property-level metrics "Brand Safety Floor Blocks" and "Brand Safe Ads", whereas DV's one-sheet of 8 April 2026 uses "Brand Risk Floor Avoidance".2direct recordsupported as statedhigh
The MRC standards governing the four metric families are, as read on 10 November 2026: the Viewable Ad Impression Measurement Guidelines version 2.0 (18 August 2015) and mobile guidelines (28 June 2016); the IVT 2.0 Addendum (June 2020) with the 2024 Interim Updates (final 24 April 2024); the Conduct of Ad Verification Guidelines (14 February 2012); the Enhanced Content Level Context and Brand Safety supplement (September 2018); and the SSAI and OTT Guidance (August 2021); MRC's AI guidance (8 July 2026) maps AI-related requirements already in seven MRC documents (including the ad verification guidelines, the 2018 supplement and the IVT addendum, but not the viewability guidelines or the SSAI and OTT guidance), with new AI standards due in phases through 2026 and early 2027.2direct recordsupported with qualificationhigh
The standards MRC cites differ by service: DoubleVerify's property-level accreditation cites the IAB ad-verification guidelines (2012) but not the 2018 supplement, Protected Media's property-level letter cites both, Zefr's and Google's content-level letters cite the supplement, and IAS's 2026 letter cites no ad-verification standard.2direct recordsupported as statedhigh
MRC revoked Method MI's accreditation for rendered display ad impressions in November 2023, after a hiatus, on its withdrawal from the audit process.2direct recordsupported as statedhigh
MRC's public record of a verifier's accreditation history is incomplete: its site drops superseded letters (all seven tried on 10 November 2026 returned HTTP 404), and its quarterly reports for January 2023 to September 2024 list status changes only, so IAS's continuation of 9 October 2024, the last letter found granting its property-level ad verification, appears in no quarterly report.2direct recordsupported as statedhigh
Protected by Mediaocean's home page says it is MRC-accredited for rendered display and video ad impressions across desktop, mobile web, in-app and connected TV, whereas MRC's letter of 22 September 2026 accredits video impressions only on CTV and MRC's list shows display impressions on desktop, mobile web and in-app only.2synthesissupported as statedmedium
Adloox's company page (read 10 November 2026) says Adloox 'is accredited by the Media Rating Council (MRC) since 2014 for comprehensive ad measurement and traffic validation' without naming metrics or environments and without the footnote or link to specifics that MRC's 2024 policy asks general representations to carry; against MRC's record (SIVT on desktop and mobile web only, no brand-safety, CTV or pre-bid metric) the claim is unscoped rather than demonstrably broader than scope, and MRC's archived list, which shows Adloox accredited on 6 February 2015 but not on 8 November 2014, fits a 2014 start only if accreditation came in the last weeks of that year.2synthesissupported with qualificationmedium
Peer39's verification page (read 10 November 2026) says it offers 'MRC-accredited viewability measurement built into the same pre-bid signal'; MRC lists no pre-bid metric for the Peer39 Adloox service, Peer39's only MRC-accredited service, and MRC's only pre-bid viewability row is IAS's, so the sentence implies accreditation for a pre-bid viewability metric MRC does not list, which contradicts H1b for the owner's page under H1b's 'states or implies' rule.2synthesissupported with qualificationhigh
Adloox's own releases of 2023-2024 describe Adloox as 'MRC-accredited' when announcing services MRC does not accredit, without a disclaimer: brand safety and suitability measurement on Meta (24 October 2024), new pre-bid DV360 segments (dateline June 2024) and publisher measurement through Equativ that covers pre-bid traffic (25 January 2023); MRC's 2024 policy asks accredited services to attach clear disclaimers when non-accredited aspects are mentioned in accreditation communications.2inferencesupported as statedmedium
No independent test with a labelled test set and a denominator of any Adloox category was found by 10 November 2026: none of the fourteen Adalytics reports whose sitemap entries are dated 2023 or later mentions Adloox or Peer39, and MRC accreditation certifies conformance with standards, not an error rate; the public numbers are vendor claims without denominators or holdouts, such as Peer39's 'invalid traffic fall from around 12% to 3% within a month', Adloox's MFA centre figure of 20% to 40% MFA for new clients falling below 1% after optimisation (average since January 2021), and a customer story citing 'IVT at 3.1%, much lower than the 7.9% benchmark'.2vendor assertionsupported as statedmedium
DoubleVerify's MRC one-sheet (PDF created 8 April 2026) discloses limits that its product pages omit: Authentic Ads and Authentic Viewable Impressions are not accredited when a campaign's settings include geographic targeting; segments released after May 2024, including One-click Brand Suitability segments, and the Made for Advertising and Low Quality GenAI categories are not accredited; and its property-level brand-suitability accreditation is not content-level brand safety; meanwhile the fraud, brand-suitability and viewability pages call DV Authentic Ad 'a proprietary, MRC-accredited metric' defined to include the intended geography.2vendor assertionsupported as statedhigh
A DoubleVerify release of 24 April 2024, still live on its newsroom on 10 November 2026, says its accreditation was extended to CTV pre-bid property-level brand suitability, contextual and Fully On-screen segments, whereas MRC's 2024 report records the pre-bid grant only as 'several product offerings' and the 13 April 2026 letter lists contextual and brand-safety-and-suitability pre-bid segments for desktop, mobile web and in-app only; DV's 10-K also says, without qualification, that its solutions are MRC-accredited and its data can serve as 'a single source standard'.2synthesissupported as statedmedium
On the large platforms MRC accredits only DoubleVerify's third-party viewability reporting, and narrowly: Meta display and video impressions and viewability reported net of GIVT for Facebook and Instagram feed placements only; YouTube video impressions and viewability from Google's Ads Data Hub, net of GIVT, for in-stream and bumper placements with Shorts, in-feed and masthead out of scope; and, since 13 April 2026, TikTok video impressions and viewability in-app filtered for GIVT and SIVT; no DoubleVerify brand-safety or suitability measurement on any platform is accredited or under review.2direct recordsupported as statedhigh
HUMAN's MRC accreditation brief (PDF created 16 August 2023, still served on 10 November 2026) speaks of 'the first-ever accreditation for pre-bid mitigation of SIVT', while MRC lists the accredited unit as 'Pre-bid IVT detection' and said in April 2025 that HUMAN is accredited for pre-bid IVT detection, not prevention of ad serving: a candidate H1b claim broader than MRC's listed scope, for workstream A to code.2inferencesupported as statedmedium
HUMAN sells brand safety and suitability (in the June 2026 suite; on its solution page by February 2026) with no MRC accreditation and no MRC audit listed for it, and without a TAG Brand Safety Certified seal, as read on 10 November 2026.2synthesissupported as statedhigh
As stated, adding: the 9 October 2024 letter's text (English-only, display and video property-level scope) was not re-opened in the verifier pass because the Internet Archive returned HTTP 429 on every attempt; MRC's July-September 2026 report records no IAS action, so no later decision explains the row either.2synthesissupported with qualificationhigh
In January 2026 MRC removed IAS Attention Measurement and IAS In-Game Measurement from in-process status because of IAS's lack of intent to engage in third-party audit.2direct recordsupported as statedhigh
No IAS page fetched on 10 November 2026 claims MRC accreditation for brand safety or suitability; IAS calls its Quality Impression metric 'MRC-accredited', which MRC's 2026 letter supports for 'Quality Ads' on desktop, mobile web and in-app, and its FY2024 10-K lists MRC accreditation for Quality Impression, Campaign Monitor and Firewall Verification Services and SIVT.2synthesissupported as statedmedium
Of the 16 MRC accreditation decision reports covering October 2022 to September 2026 (opened 10 November 2026), only the report for April to June 2025 (dated 11 July 2025) mentions Pixalate, recording the April 2025 continuation; none of the five reports from July 2025 to September 2026 records a Pixalate decision, so the accredited scope still rests on the letter of 8 April 2025.2direct recordsupported as statedhigh
Two Pixalate pages live on 10 November 2026 describe outputs as MRC-accredited beyond the Analytics platform's listed scope: a 27 January 2022 post calls its pre-bid Blocking 'MRC-accredited', whereas MRC lists no accredited Pixalate pre-bid service and has had its pre-bid SIVT in audit since September 2025, and the MRT FAQ calls the global-pool IVT rates shown for apps and domains 'MRC-accredited IVT measurement', whereas MRC's letter accredits the Analytics platform's reported statistics.2synthesissupported as statedmedium
Pixalate's Analytics page calls its 12 SSAI metrics 'unique to Pixalate', while MRC's list also shows Integral Ad Science accredited for SSAI video metrics (CTV for certified traffic only), and MRC's February 2024 policy says it does not generally approve 'Only' accreditation claims in communications.2synthesissupported as statedmedium
Pixalate's brand-safety output is property-level: apps are scored from screenshots and from natural-language processing of titles, descriptions and advisories across seven categories, websites are assessed at the homepage (root-domain) level, the High Risk Domains brand-safety list contains known adult websites, and the Analytics brand-safety report applies those app ratings to measured mobile-app impressions; none of it is MRC-accredited or under MRC review.2vendor assertionsupported as statedhigh
MRC's letter of 6 November 2025 granted initial accreditation to Zefr's independent third-party processing and reporting of Google Ads Data Hub YouTube traffic from Google Ads and DV360 for video impressions and viewability metrics inclusive of GIVT filtration only, in desktop, mobile web and in-app (no CTV), for skippable and non-skippable in-stream and bumper ads, excluding in-feed video, masthead, Shorts and Demand Gen non-YouTube inventory; the standards listed include only the GIVT provisions of IVT 2.0.2direct recordsupported as statedhigh
Inference: of the platforms on which Zefr documents selling content-level brand safety and suitability (YouTube, Meta, TikTok, Snap and, from 30 September 2026, Reddit), only YouTube carries MRC accreditation, and even there Zefr sells inventory and languages the letter excludes (Shorts since September 2024, Google Video Partners since 29 January 2026, and coverage it describes as more than 50 languages against an English-only grant); the accredited share of sold units is therefore 1 of 5 counted by platform, or at most 4 of 8 counting YouTube's four MRC environments separately (data table zefr_unit_share_conventions).2inferencesupported as statedmedium
Zefr's company page claims a 2025 'MRC accreditation for viewability & IVT', while MRC's record shows only YouTube Ads Data Hub video impressions and viewability inclusive of GIVT filtration and lists Zefr under no invalid-traffic metric; the same page dates 'key industry accreditations for brand safety and suitability' to 2019, a claim that names no body and that MRC's records opened cannot confirm (its first Zefr decision found is November 2025).2synthesissupported as statedmedium
Zefr's 9 March 2026 release states its accreditation as content-level labelling and reporting 'of YouTube' and sends readers to MRC's list for scope, but omits the English-only and in-stream limits and glosses the accreditation as confirming 'validity, reliability, and governance'; its brand-safety solution page shows the MRC logo beside 'Content-Level Brand Safety and Suitability' and 'Viewability' on a page presenting YouTube, Meta, TikTok, Snap and Reddit, without saying the accreditation is YouTube-only.2synthesissupported as statedmedium
MRC's policy (18 October 2025, addressed to MRC-audited digital services, a group Zefr joined in November 2025) permits only vendors accredited for content-level brand safety for the specific environment to invoke 'Brand Safety', and after a six-month grace period withdraws accreditation from any vendor that positions text- or keyword-based property-level verification as Brand Safety; Zefr labels unaccredited Meta, TikTok, Snap and Reddit products and a domain-level SPN product 'Brand Safety', and whether the policy reaches an accredited vendor's unaudited domain-level product (whose classification basis Zefr does not disclose) is not stated by MRC (unresolved).2inferencesupported with qualificationmedium
On 11 April 2025 MRC said IAS was then neither audited nor accredited for pre-bid IVT detection or for the publisher tools Adalytics examined, and that IAS would submit pre-bid IVT detection for audit in 2025; MRC granted initial accreditation of IAS's pre-bid IVT-avoidance segments (desktop, mobile web, in-app) in June 2026.2, 3direct recordsupported as statedhigh
On MRC's accredited list read on 10 November 2026, third-party measurement of a large platform's own placements is accredited on 4 of the 9 platforms (Meta: DoubleVerify and IAS viewability net of GIVT; YouTube: DoubleVerify, IAS and Zefr ADH viewability and Zefr content-level; TikTok: DoubleVerify in-app video viewability with SIVT; Amazon: IAS server-to-server with SIVT), third-party content-level accreditation on 1 of 9 (YouTube, Zefr), and none on X, Snap, Pinterest, Reddit or LinkedIn, while first-party accreditation is listed for 6 of 9 (Google, Meta, Amazon, Snap, Pinterest, LinkedIn).2, 5synthesissupported as statedhigh
On MRC's accredited digital list read on 10 November 2026, two services hold content-level brand safety and suitability accreditation, both for YouTube: Google's first-party controls (desktop, mobile web, in-app; in-stream and, since the April 2026 decision, Shorts) and Zefr's third-party labelling and reporting (desktop, mobile web, in-app, CTV; English in-stream video and music video only); no open-web, non-YouTube CTV or other-platform service holds it after GumGum's and Meta's first-party accreditations were revoked in September 2025.2, 5, 6direct recordsupported as statedhigh
CTV viewability is accredited for only two services, both for video and both limited to certified apps (DoubleVerify and IAS); no service holds display viewability on CTV, Protected Media and Pixalate hold none on CTV, and Pixalate's video viewability on desktop, mobile web and in-app has been in audit since September 2025; IAS's 2024 letter had further confined its CTV accreditation to five operating systems or devices.2, 6direct recordsupported as statedhigh
Pre-bid IVT detection is accredited for three services: DoubleVerify and HUMAN on desktop, mobile web, in-app and CTV, and IAS on desktop, mobile web and in-app only by an initial grant in June 2026 (IP and user-agent segments and its Fraud IP list), although MRC said on 11 April 2025 that IAS was then neither audited nor accredited for it; Pixalate's pre-bid SIVT and Escalated's pre-bid scoring remain in audit.2, 6direct recordsupported as statedhigh
MRC's statement of 11 April 2025 says it requires post-serve GIVT filtration on every impression for all accredited digital vendors, does not require pre-bid filtration or blocking (its standards caution against it), limits pre-bid audits to the production and accuracy of the files and databases vendors supply, and has no insight into billing.2, 6direct recordsupported as statedhigh
IAS's ad-fraud page says IAS was the first verification provider accredited for SIVT in CTV, a priority claim MRC's own records contradict: MRC's digital list showed SIVT accreditation including CTV for DoubleVerify, HUMAN, Oracle Moat, Pixalate and Protected Media on 24 January 2023 and 23 February 2024 (Internet Archive captures), while IAS's SIVT row covered desktop, mobile web and in-app only until MRC granted its CTV SIVT filtration in April 2024; MRC's 2024 representation policy approves 'first' claims only when they are factually unquestionably correct and concern a matter MRC has an interest in emphasizing.2, 6synthesissupported with qualificationlow
TAG's CAF version 11.0 (July 2026) requires anti-fraud and measurement services acting as verifiers under its device attestation guidance to implement OM SDK 1.6 or higher, while the OM SDK JavaScript clients reached 1.6.12 on 2 November 2026 (1.6.0 of 29 October 2025 added device attestation for Fire TV and Apple devices) and IAB Tech Lab's ads.cert page, last updated 14 February 2024, still marks Authenticated Delivery and Authenticated Devices as future.2, 6direct recordsupported as statedhigh
MRC accredits three more scored products for CTV invalid traffic but none of them for CTV viewability: HUMAN (letter of 14 April 2026) for post-bid tracked ads, impressions and GIVT and SIVT detection and for pre-bid GIVT and SIVT requests, with viewability on desktop, mobile web and in-app only; Pixalate (8 April 2025) for ad counts, impressions, SSAI video metrics and GIVT and SIVT filtration, with display viewability on desktop, mobile web and in-app only and video viewability under review for those three environments; and Protected Media (22 September 2026) for video impressions by non-SDK measurement with GIVT and SIVT metrics.2, 6direct recordsupported as statedhigh
On the CTV pages opened on 10 November 2026, Pixalate's documentation matches MRC's CTV scope and says it reports no CTV viewability, IAS's CTV page makes no MRC claim, and DoubleVerify's release of 24 April 2024 said its MRC accreditation extended to CTV pre-bid property-level brand suitability and contextual segments, whereas the 13 April 2026 letter lists brand safety, suitability and IAB contextual pre-bid segments for desktop web, mobile web and in-app only.2, 6synthesissupported with qualificationmedium
IAB Tech Lab's ads.cert 2.0 page (last updated 14 February 2024) marks Authenticated Connections ready for adoption and Authenticated Devices as future; a January 2022 Tech Lab blog by Index Exchange described a CTV proof of concept with Publica, Google DV360 and Oracle Moat, the open-source repository's last commit is dated 8 March 2023, and no adoption figure was found in the sources opened.2, 6direct recordsupported as statedmedium
On MRC's accredited digital list read on 10 November 2026, the 'Adloox' entry covers tracked ads, rendered ad impressions, and viewable impressions and viewability for display and video on desktop, mobile web and in-app, and sophisticated invalid traffic (SIVT) detection and filtration on desktop and mobile web only; it has no CTV, pre-bid, property-level ad verification or content-level brand-safety row (MRC's only pre-bid viewability row is IAS's), and Adloox, Peer39 and Scope3 are absent from the services-under-review list.2, 6direct recordsupported as statedhigh
MRC's letter of 29 June 2026 continues accreditation of the 'Peer39 Adloox' Viewability and Fraud prevention service for display and video tracked ads (gross), impressions, decision rate, viewable impressions, viewability and exposure metrics on desktop, mobile app and mobile web, with SIVT filtration applied on desktop and mobile web only; in-app metrics are accredited net of GIVT, the decision rate included on the letter's p.1 and in its TAG addendum, although Enclosure 1's bracket can be read as extending the decision rate's 'Total Net of SIVT' to mobile app; the standards are MRC's Minimum Standards, the Viewable Impression Measurement Guidelines and the GIVT and SIVT provisions of IVT 2.0 with interim guidance, plus the IAB and MMA impression guidelines, and the TAG addendum names domain-list and data-centre IP-list filtering.2, 6direct recordsupported with qualificationhigh
MRC's letter of 13 April 2026 continued accreditation of DoubleVerify's Pinnacle platform for display and video impressions on desktop, mobile web, in-app and CTV inclusive of SIVT filtration; display and video viewability on desktop, mobile web and in-app (including OM SDK traffic); property-level ad-verification metrics on all four environments; and CTV video viewability, where only certified apps count toward Video Viewable Impressions and Fully On-Screen metrics (other CTV viewability metrics may include non-certified but tested apps); MRC's list (read 10 November 2026) shows the same Quality Analytics Platform rows plus pre-bid IVT detection on all four environments, and no DoubleVerify service under review.2, 6direct recordsupported with qualificationhigh
DoubleVerify's property-level ad-verification accreditation is limited by language: the 13 April 2026 letter's table marks 176 languages for binary classification, 56 at domain level, 41 at page level and 10 for CTV (Chinese, English, French, German, Hindi, Japanese, Korean, Malay, Spanish and Vietnamese), French for CTV and five domain-level languages having been added that month; DV's own brand-suitability page says 'over 40 languages' on desktop and mobile and nine for CTV.2, 6direct recordsupported as statedhigh
MRC's pre-bid accreditation for DoubleVerify is narrower than its post-bid scope: the April 2026 letter's pre-bid table lists brand safety and suitability segments (including one-click segments), viewability and IAB contextual categories on desktop, mobile web and in-app only, while on CTV it lists only Fraud/IVT, Fully On-Screen, app-store categories and site and app lists; for the Authentic Brand Suitability product the pre-bid Fraud/IVT offering covers site and app fraud only.2, 6direct recordsupported as statedhigh
On 14 April 2026 MRC continued accreditation of HUMAN's Ad Measurement Platform for Ad Fraud Sensor post-bid display and video tracked ads, impressions, decisions and incomplete loads with GIVT and SIVT detection, and for Ad Fraud Defense pre-bid GIVT and SIVT requests, each in desktop, mobile web, mobile in-app and CTV, and granted initial accreditation to HUMAN Ad Viewability Measurement for display and video in desktop, mobile web and mobile in-app only.2, 6direct recordsupported as statedhigh
On MRC's accredited digital list read on 10 November 2026 (no update stamp), HUMAN's entry carries tracked ads and rendered impressions (display and video), pre-bid IVT detection and SIVT detection/filtration in desktop, mobile web, in-app and CTV, the SIVT row annotated as applying to backend detection and IVT predictions, plus display and video viewability in desktop, mobile web and in-app; HUMAN has no entry of its own on MRC's services-under-review list and appears there only as the SIVT provider in Walmart Connect Onsite Display's audit (in process as of 11/2025).2, 6direct recordsupported as statedhigh
MRC's letter of 9 July 2026 continues accreditation of IAS's Campaign Monitor, Firewall and Viewability Systems platform and CTV measurement platform for display and video impressions, viewability and Quality Ads inclusive of SIVT on desktop, mobile web and in-app, CTV video metrics and SIVT for certified applications only, and display blocks on desktop and mobile web, and grants initial accreditation of pre-bid viewability segments and IVT-avoidance segments on desktop, mobile web and in-app.2, 6direct recordsupported as statedhigh
No IAS content-level brand safety or suitability service is MRC-accredited: MRC lists 'IAS YouTube Brand Safety' (English and Spanish YouTube user-generated video on desktop, mobile web, in-app and CTV) as an audit in process since April 2023, and MRC's platform chart of 22 October 2026 shows it as in the accreditation process, not accredited.2, 6direct recordsupported as statedhigh
On 8 April 2025 MRC continued accreditation of Pixalate's Analytics platform for display and video ad counts and impressions (gross, net of GIVT and total net of SIVT) in desktop, mobile web, mobile in-app and CTV, for display viewability (net of GIVT and total net of SIVT) in desktop, mobile web and in-app only, and for SSAI video metrics in all four environments.2, 6direct recordsupported as statedhigh
MRC's services-under-review list (read 10 November 2026) shows a Pixalate audit in process as of September 2025 for video viewability in desktop, mobile web and in-app and for pre-bid SIVT delivered through data feeds and the Ad Fraud API (no environment stated); CTV viewability is neither accredited nor under review, and MRC's accredited list shows pre-bid IVT detection only for DoubleVerify, HUMAN and Integral Ad Science.2, 6direct recordsupported as statedhigh
TAG's certification guidelines were all reissued in July 2026: Brand Safety Certified version 5.0 (change log back to version 1.0 of September 2020, which merged JICWEBS' DTSG with TAG's IQG programme), Certified Against Fraud version 11.0 (intermediaries must use PaymentID; the CTV SSAI billing-notification requirement removed), Certified Against Malvertising version 7.0 (renamed from Certified Against Malware) and Certified for Transparency Guidelines version 2; TAG certifies the entity, not a product.2, 6, 8direct recordsupported with qualificationhigh
On MRC's accredited digital list read on 10 November 2026, 18 of the 41 listed services hold at least one CTV metric, but CTV SIVT filtration is held by 8 (DoubleVerify, IAS, HUMAN, Pixalate, Protected Media, the New Innovid report of Protected Media-measured traffic, Google Ads and Google DV360), CTV video viewability by 2 (DoubleVerify for certified applications, IAS for certified traffic), CTV property-level ad verification by 1 (DoubleVerify) and CTV content-level brand safety and suitability by 1 (Zefr's YouTube reporting).2, 6, 12direct recordsupported as statedhigh
MRC's letter of 20 January 2026 granted initial accreditation to Zefr's content-level brand safety and suitability labelling and reporting of English-language YouTube in-stream video and in-stream music video content in desktop, mobile web, in-app and CTV, excluding Shorts, YouTube TV, Google TV, Masthead, Google Video Partner videos and 'Volumetric Data', against MRC's Minimum Standards, the 2018 Enhanced Content Level Context and Brand Safety supplement and the IAB Ad Verification Guidelines.2, 6, 12direct recordsupported as statedhigh
MRC's list on 10 November 2026 shows Integral Ad Science with property-level ad verification on desktop and mobile web, but the last MRC letter whose body grants it is the continuation of 9 October 2024 (display and video, English only), which the list linked in every capture read from 28 April 2025 (the first capture whose entries carry letter links) to 16 June 2026; the letter of 9 July 2026 that the list now links omits ad verification and MRC's Q2 2026 report says that June 2026 continuation excluded it, while a letter of 7 November 2025 for IAS's Amazon DSP integration lists brand-safety statistics for desktop and mobile web in its enclosure only.2, 6, 12, 13direct recordsupported with qualificationhigh
Escalated Solutions, LLC (Carmel, Indiana) describes itself as a programmatic ad-fraud vendor for 'the sell-side to the buy-side' offering pre-bid and post-bid GIVT and SIVT detection and blocking for display, video, app and CTV, and MRC lists its audit (display only) as in process as of June 2026, so it stays a cohort B candidate that is not accredited.2, 7synthesissupported as statedhigh
In TAG's registry on 10 November 2026, DoubleVerify and IAS hold Certified Against Fraud and Brand Safety Certified (both independently validated, global) plus Certified for Transparency (TAG-validated, North America and Europe), and TAG's Platinum page lists both as current Platinum members; HUMAN and Pixalate hold Certified Against Fraud; Protected Media holds Certified Against Fraud and Certified for Transparency; Adloox holds Certified Against Fraud and Brand Safety Certified; Channel Factory and illuma hold Brand Safety Certified; Zefr and Escalated are enrolled with no seal; and searches for Peer39, Comscore and Fraudlogix return no company.2, 8direct recordsupported with qualificationhigh
MRC's 2018 content-level supplement sets no suitability tiers of its own: it requires discrete reporting so buyers can set their own risk tolerance, points to the 4A's APB for the floor and the suitability framework, and makes either direct third-party content-level measurement or validated first-party data from the platform a condition of content-level accreditation for third-party verifiers.2, 8, 12direct recordsupported as statedhigh
Peer39's verification page (read 10 November 2026) claims "MRC-accredited viewability measurement built into the same pre-bid signal", but MRC lists no pre-bid metric for the Peer39 Adloox service, whose 29 June 2026 letter covers post-serve measurement on desktop, mobile web and in-app; the claim states accreditation for a pre-bid metric MRC does not list, which contradicts H1b.2, 12synthesissupported as statedmedium
MRC's Policy for Accreditation Representations (15 February 2024) asks accredited services to state what is and is not accredited, bars claims for anything a service is not accredited for (including services under review), restricts "first" and "only" claims, and names the accreditation letter as the authoritative reference.2, 12direct recordsupported as statedhigh
On MRC's accredited digital list as read on 10 November 2026 (41 services, no update stamp), seven third-party verification vendors hold 14 entries (DoubleVerify 3, Integral Ad Science 4, Protected Media 2, Zefr 2, HUMAN 1, Pixalate 1, Adloox 1), one ad-server entry reports Protected Media-measured traffic, and none of the other 26 entries is a third-party verification vendor.2, 13direct recordsupported as statedhigh
The ANA study's invalid-traffic, measurability and viewability figures are the three participating verification companies' own classifications matched to DSP log data (all three MRC-accredited, p. 89), not a test of those companies: IVT was 0.78% of impressions and 0.69% of spend (September 2022 to January 2023); measurability had a median of 86% across participants; 85% of measurable impressions were viewable; and the ANA's '73% of total impressions both measurable and viewable' is its own product of those two figures (86% x 85%), which combines a median with an aggregate share.3independent measurementsupported with qualificationhigh
In the ANA study, made-for-advertising sites (a subset of domains matched to DeepSee.io's list of 4,500 known MFA sites) took 21% of impressions and 15% of spend, and the ANA reported that they showed high measurability, good viewability and low IVT, with no meaningful IVT difference from other media; neither DoubleVerify nor IAS sold MFA detection in that window, since DoubleVerify announced its MFA beta on 14 September 2023 (for all advertisers from 20 September, after undated on-request handling) and IAS on 16 October 2023 (select customers).3synthesissupported with qualificationhigh
Brand safety was outside the 2023 ANA study's measurement, and the ANA's Q3 2025 benchmark report says the 2023 study included no brand-safety or suitability data because methods were inconsistent across vendors, so the report's statement that MFA sites "usually have brand-safe environments" is not backed by any brand-safety measurement in the pages read.3synthesissupported as statedmedium
The ANA's Complete Report (December 2023) tested no verification product's detection: it noted that many supply-chain companies began offering MFA-blocking after its June 2023 First Look, advised advertisers to verify that such offerings "function as sold", and suggested validating partners' IVT reporting with a different vendor.3inferencesupported as statedhigh
The ANA Programmatic Transparency Benchmark, run with TAG TrustNet (a joint TAG and Fiducia initiative) and Fiducia, grew from 37 participating and 21 active marketers (38.5 billion impressions, $235 million, April to September 2024) to 86 participating and 66 active marketers (20.9 billion impressions, $160 million, January to March 2026), and builds its IVT, measurability and viewability figures from log data matched between DSPs and the advertisers' verification providers, with MFA data from DeepSee.io and brand-risk data from Mobian, so it reports prevalence measured by those suppliers and tests none of them.3independent measurementsupported with qualificationhigh
In the benchmark's cost waterfall, IVT was 0.4% in Q3 2025, fell to 0.2% in Q1 2026 and was 0.1% in Q2 2026, while MFA exposure rose from 0.4% to 1.1% in Q1 2026; the median MFA share fell from 10% in the 2023 study to 1.1% in April to September 2024, and the benchmark warns that waterfall values and medians are not comparable.3independent measurementsupported with qualificationmedium
Between June 2023 and March 2025 Adalytics published ten reports that name a verification vendor (dated from first Internet Archive captures, its sitemap and the parties' dated replies), all observational (case reports, plus one post relaying an MRC statement), and the three that bear most directly on DoubleVerify, IAS and HUMAN (August 2024, February 2025, March 2025) each state that they make no claim about quantitative magnitude or relative abundance; its two later posts in the sitemap (July 2025, June 2026) name no verification vendor on their pages as read on 10 November 2026, though the July 2025 report itself is a PDF hosted off-site that has not been opened.3synthesissupported with qualificationhigh
DoubleVerify and IAS each answered on 7 August 2024 that the tags in the report's screenshots were from their publisher products (DV: all of those shown to it; IAS: 'almost all') and that Adalytics could not see advertisers' settings (monitoring rather than blocking, exception lists); DV added that its preliminary analysis found all content shared with it classified accurately, but neither statement published the classifications or counts behind those assertions, and IAS's statement does not address the low-risk values its publisher endpoint returned for explicit pages, saying only that the examples represent one part of its solution and that tags on a page need not match the final decision for the creative shown.3vendor assertionsupported with qualificationhigh
DoubleVerify's 7 April 2025 response also asserted, without method or sample, that its recent testing confirmed 100% post-bid detection of the known-bot and data-centre GIVT cited in the report, and that with its pre-bid avoidance GIVT from declared bots falls from below 1% to below 0.03% of programmatic impressions, a 75% to 98% reduction depending on client settings, figures produced by the classifier being defended.3vendor assertionsupported as statedhigh
Adalytics' MFA report (first archived 18 March 2024) observed, in January 2024, ads from hundreds of brands on 22 sites it judged MFA, many carrying tags from IAS, DoubleVerify, Oracle Moat, Pixalate, HUMAN or Fou Analytics, when DV's MFA product was in beta for all advertisers and IAS's had been announced in October 2023 as a beta for select customers with general availability expected in early 2024; DV answered that it had already classified virtually all the sites as MFA, which, if correct, places the gap in buyers' settings rather than in classification.3synthesissupported with qualificationmedium
Category-level agreement in Smith et al. was lower still for some categories: of the 31 articles that Oracle or IAS flagged for hate speech, the two disagreed on 27 (87.10%; alpha 0.19 across co-rated articles); the authors also report manually found misclassifications by each provider, without a rate, say no ground-truth dataset of brand-safety labels exists, and call for a public human-labelled benchmark to evaluate accuracy across tools.3independent measurementsupported with qualificationhigh
Adalytics' December 2020 analysis (first archived 16 December 2020, updated 2 February 2021) of brand-safety labels on 25,730 archived wsj.com articles found Comscore and Oracle Moat agreeing on safe or unsafe 58.7% of the time, an inter-vendor agreement estimate with a denominator inside H3's 2019-to-freeze window; a companion post was edited after IAS said the data it described were not advertiser keyword blocklists.3independent measurementsupported as statedmedium
arXiv searches run on 10 November 2026 found no study beyond Smith et al. that tests a named verification product's outputs, and none reporting a verification product's miss or false-positive rate on a labelled set; the labelled designs found test other products, such as FP-Inconsistent's evasion rates of 52.93% against DataDome and 44.56% against BotD over half a million requests from 20 bot services, while Vekaria et al. document ad-inventory pooling that circumvents brand-safety protections.3synthesissupported as statedmedium
Platform transparency reports screened estimate violating-content prevalence from labelled samples, YouTube's Violative View Rate from a sample of videos sent to its content reviewers (0.16% to 0.18% of views when introduced in April 2021) and Meta's prevalence from labelled samples of content views, but these are first-party measures of each platform's own policy enforcement, not error rates of a third-party verification product.3synthesissupported as statedmedium
DoubleVerify's release of 7 May 2026 for its 2026 Global Insights CTV report states that fraud rates were less than 1% in DV-protected CTV campaigns against nearly 9% in 'unprotected campaigns'. The release says its insights as a whole rest on 'billions of impressions' from DV-protected campaigns and on controlled tests where protection controls were not applied; that the 9% arm comes from those tests is an inference. No count of impressions, campaigns or tests, no period and no assignment method is published.3vendor assertionsupported with qualificationhigh
DoubleVerify's 7 May 2026 release states that in unprotected campaigns fraud can cost advertisers about $1.8 million per billion CTV impressions 'even at conservative estimates', without stating the CPM, the fraud rate or the method behind the figure.3vendor assertionsupported as statedhigh
DoubleVerify's FY2025 10-K (filed 26 February 2026) reports about 9.5 trillion Media Transactions Measured in 2025 (8.3 trillion in 2024, 7.0 trillion in 2023), the unit on which it bills advertisers, and acknowledges as a risk factor that data from its integration partners have in the past been inaccurate and caused errors in its technology, without quantifying any error rate.3direct recordsupported as statedhigh
IAS's 20th Media Quality Report (release of 20 May 2025) reports a fraud rate of 10.9% for non-optimized campaigns in H2 2024 against 0.7% for optimized ones (15 times higher) and presents this as showing that anti-fraud technologies work, although non-optimized campaigns, those without any fraud mitigation strategy, were only 4.0% of IAS-monitored campaigns (2024 in IAS's text, H2 2024 on its chart; the text gives 2.2% for 2023, while the chart labels 2.2% H1 2024 and shows 0.9% for H2 2023).3vendor assertionsupported with qualificationhigh
IAS's 21st Media Quality Report (released 9 July 2026) puts open-web invalid traffic at 1.1% across 2025, with optimized IVT about 1% on average, North America at 1.36% and the US at 1.40%, but its benchmark scope statement names display and video on desktop and mobile web, where the 20th edition named desktop, mobile web, mobile app and CTV, and it defines IVT by intent rather than in MRC's GIVT and SIVT terms.3vendor assertionsupported as statedhigh
IAS's 21st Media Quality Report gives brand safety and suitability fail rates only as multiples of baselines it does not publish (the global brand safety fail rate fell from 3.7X in January to 1.0X in December 2025) and infers from its own falling fail rates that the most severe adjacency risks are 'caught and blocked with growing reliability'.3vendor assertionsupported as statedhigh
Pixalate's Q1 2026 benchmarks (released 29 April 2026) put global IVT at 20% of web traffic, 39% of mobile app traffic and 25% of CTV traffic across over 82 billion programmatic impressions drawn predominantly from buy-side open-auction traffic, figures Pixalate labels its opinions; its Q2 2026 global rates, published in releases of 12 August 2026, were 20% web, 41% mobile app and 26% CTV across over 80 billion impressions (the full report is behind an email form).3vendor assertionsupported with qualificationhigh
TAG's 2024 US Fraud Benchmark (release of 29 October 2024), conducted by The 614 Group, found IVT under 1% in TAG Certified Channels for a fourth year, using impressions shared by agency holding companies and measured by their MRC-accredited vendors from 1 January to 30 June 2024 without sampling (745 billion TCC impressions).3commissioned measurementsupported as statedhigh
TAG's 2024 US Ad Fraud Savings Report (9 October 2024), developed by Cunningham.Tech Consulting for TAG, the 4A's, the ANA and the IAB, attributes $10.8 billion of 2023 savings to anti-fraud programmes by subtracting $979 million (a 0.92% weighted IVT rate on $106.4 billion of revenue through TAG-certified and non-certified channels) from a counterfactual $11.78 billion (a 9.96% 'unfiltered' IVT rate blended from DoubleVerify, IAS, Oracle Moat and Pixalate, applied to all $118.2 billion of US display and video revenue). Losses on the remaining $11.8 billion of revenue are not subtracted, so about $1.18 billion that the report's own unfiltered rate says was still lost is counted as saved (derived: 0.0996 x 11.8); on the method of TAG's 2025 European report the US figure would be about $9.6 billion (derived: 0.0996 x 106.4 - 0.0092 x 106.4 = 9.62; inference).3commissioned measurementsupported with qualificationhigh
Juniper Research's September 2023 ad-fraud whitepaper, in a copy hosted by the click-fraud vendor Fraud Blocker that contains a Fraud Blocker vendor chapter, forecasts global ad fraud of $84.2 billion in 2023 rising to $172.3 billion by 2028 (and 'almost $100 billion' by the end of 2024); it gives two definitions, a broad one (any attempt to deceive digital advertisers or ad networks for financial gain) and a quoted click-based one (intentionally repeated clicking on pay-per-click ads), and does not say which the forecast uses.3forecastsupported with qualificationhigh
IAS's 20th Media Quality Report states that ad fraud siphoned an estimated $100 billion of ad spend in 2024, citing Juniper Research's September 2023 'Quantifying the Cost of Ad Fraud: 2023-2028', a forecast whose whitepaper gives both a broad and a click-based definition of ad fraud; the 21st edition cites the same Juniper report without repeating the dollar figure.3synthesissupported with qualificationhigh
CHEQ, a bot- and click-protection vendor, produced with economist Roberto Cavazos of the University of Baltimore an estimate that online ad fraud would cost advertisers $35 billion in 2020, obtained by applying a 10.5% 'composite rate' drawn from 'industry insiders' insights' to eMarketer's $332.84 billion of 2020 global digital ad spend, under a definition spanning fraudulent impressions, clicks, conversions and data events.3forecastsupported with qualificationhigh
The ANA's release of 1 May 2019 on the fourth Bot Baseline, a joint study with White Ops (now HUMAN), projected global bot-fraud losses of $5.8 billion for 2019 (against $6.5 billion in the 2017 study) and a counterfactual $14 billion without anti-fraud measures, from data on 50 ANA members' campaigns analysed by White Ops between 1 August and 30 September 2018.3forecastsupported as statedhigh
The industry-wide ad-fraud totals in circulation, from $5.8 billion (bot fraud, 2019 projection, ANA and White Ops) through $35 billion (2020, CHEQ) to $84.2 billion (2023) and $172.3 billion (2028 forecast, Juniper), were each produced with, or distributed by, a fraud-protection vendor (White Ops, CHEQ, Fraud Blocker), use incompatible definitions and are projections or models, so none can serve as a measured market total or a hero number.3synthesissupported as statedhigh
Zefr's own researchers published (arXiv, 7 August 2025; accepted to the CVAM workshop at ICCV 2025) a zero-shot test of six multimodal LLMs and professional human reviewers on 1,500 videos (500 in each of three risk categories, sampled 1 July 2024 to 1 March 2025), reporting precision, recall and F1 (best LLM overall F1 0.91, humans 0.98); it scores general-purpose models against professional reviewers' labels, and the reviewers against a senior reviewer's check of a sample, not Zefr's sold product.3commissioned measurementsupported as statedhigh
The only Forrester Total Economic Impact work found for a verification product is Oracle's: a spotlight of two Oracle-commissioned studies (Oracle Moat Analytics, November 2021, ROI 287%; Oracle Contextual Intelligence, February 2022, ROI 105%) built from 11 customer interviews and a survey of 107 US customers into a composite organisation, for a product line since retired; searches on 9 November 2026 found no TEI commissioned by DoubleVerify, IAS, HUMAN (for advertising), Pixalate, Zefr, Channel Factory or Peer39.3commissioned measurementsupported as statedmedium
The commissioned studies found for cohort C vendors measure opinions or adjacency effects rather than classification accuracy: Zefr commissioned Forrester Consulting in late 2021 to survey over 325 US and UK marketing decision-makers ('The Maturation of Brand Suitability'), and MAGNA Media Trials with Channel Factory reported in September 2021 an 8% loss of purchase-intent impact for ads next to misaligned content.3synthesissupported as statedmedium
The joint made-for-advertising definition issued on 26 September 2023 by the ANA, the 4A's, the WFA and ISBA relays the ANA report's finding that MFA sites show high measurability, good viewability, low invalid traffic and usually brand-safe environments, and the report's recommendation that advertisers decide independently whether MFA fits their brand suitability standards; ISBA's 2025 guide reproduces a short form of the definition for UK advertisers to avoid such inventory.3direct recordsupported with qualificationhigh
Adloox's DV360 guide, archived on 8 December 2022 at its original address (its blog index dates it 15 November 2022; the current site index says 23 November 2022), offered pre-bid DV360 segments 'preventing your ads from ending up on made for advertising websites' at $0.03 CPM and described the 'Fraud' segment as covering impressions on fake and 'made for advertising' websites, and a post the same archive dates to 18 November 2021 already pitched pre-bid avoidance of 'Arbitrage sites' in DV360 and Xandr; Adloox therefore offered an MFA-labelled pre-bid exclusion inside the ANA study's window (September 2022 to January 2023), which the brief's statement that no verification product then sold MFA detection does not account for; whether Adloox's 2022 label matches the joint MFA definition of 26 September 2023 is not documented.3vendor assertionsupported with qualificationlow
Adalytics' report on YouTube ads and children's data (sitemap date 17 August 2023; observations as of June and July 2023) names Channel Factory once, saying Apple appeared to use it for some iPhone 14 campaigns whose ads were observed on 'made for kids' YouTube channels; the list is a non-exhaustive sample with no denominator, and the report allows that some brands intended such placements.3independent measurementsupported as statedmedium
A study by five NJIT authors and one DeepSee.io author (arXiv, 3 January 2026; supported by NSF grants, the Martin Tuchman '62 Chair Endowment and the Leir Foundation) read DoubleVerify, IAS and Oracle brand-safety key-values exposed on news pages and found 1,416 of 3,219 classified articles (43.99%) unsafe for at least one provider under the authors' own rules, with agreement between IAS and DV on the 94 articles both rated (all on forbes.com; Oracle also rated 89 of them) at Krippendorff's alpha 0.19; it measures consistency between providers, not accuracy, because no ground-truth labels existed.3independent measurementsupported with qualificationhigh
Adalytics' report of 7 February 2025 relayed advertisers' statements that their brand-safety vendors marked 100% of measured impressions on imgbb.com and ibb.co safe or suitable; DoubleVerify answered that day that its customers' impressions there were 0.000047% of the media transactions it measured in 30 days and that pages are classified only once they reach an unpublished level of ad traffic, and its archived statement added that 83% of those impressions served to the site's home page, a sentence the live page (modified 15 October 2025) no longer carries.3synthesissupported as statedhigh
DoubleVerify sold made-for-advertising detection from September 2023, after the ANA study's window: it announced MFA measurement and avoidance on 14 September 2023, in beta to all advertisers from 20 September 2023 after handling MFA sites on request, classifies MFA sites into high, medium and low tiers by human and AI review, re-evaluates active MFA sites about every 90 days, and says its MFA category is not MRC-accredited.3vendor assertionsupported as statedhigh
Smith et al. (arXiv, 3 January 2026) measured agreement, not accuracy: IAS gives no overall article rating, so the authors built one, and across 94 Forbes articles rated by both IAS and DoubleVerify the disagreement was 37.23% with Krippendorff's alpha of 0.19; the authors say no ground-truth dataset of brand-safety labels exists.3independent measurementsupported as statedmedium
IAS's own accuracy claims carry no published method: its FY2024 10-K says its frame-by-frame analysis delivers 'up to 74% more accurate' social brand-suitability reporting than limited-sample frame analysis, its brand-safety page says IAS is '42% more accurate than the next best provider at classifying online content', its why-IAS page claims it is '3X more effective at identifying high-risk content', and its ad-fraud page claims to identify more bot traffic than 'a leading competitor'.3vendor assertionsupported with qualificationhigh
IAS announced its made-for-advertising detection on 16 October 2023 as a beta for select customers, with general availability expected in early 2024, trained on Jounce Media's MFA domain list with Sincera signals; it later announced general availability globally; on IAS's own dating (inference), IAS sold no MFA detection during the ANA study's window (September 2022 to January 2023).3vendor assertionsupported as statedhigh
Zefr's researchers published (arXiv, 7 August 2025) a test of six multimodal LLMs against professional reviewers' labels on 1,500 videos (500 in each of three risk categories, 823 risky and 677 safe, sampled 1 July 2024 to 1 March 2025): the best models reached an overall F1 of 0.91, and the human reviewers scored precision 0.98, recall 0.97 and F1 0.98 against a senior team member's check of a sample; it does not test Zefr's sold classifier or settings.3commissioned measurementsupported as statedmedium
From Q3 2025 the benchmark scores brand risk with Mobian's AI classification of web and mobile pages (on average 99.1% of spend in low-risk environments in Q3 2025) and relays Mobian's own analysis, done outside the benchmark, that roughly one-third of professional news pages were misclassified as unsafe by "legacy systems", a vendor's false-positive claim with no published sample, labels or denominator.3, 4vendor assertionsupported as statedmedium
The ANA's study of 21 advertisers' log data (September 2022 to January 2023) found a median measurability of 86% across participants (range 0.02% to 99.61%, four below 20%) and 85% of measurable impressions viewable, from which the ANA derives that 73% of total impressions were measurable and viewable, treating the 86% median as a share of total impressions; it found an average IVT rate of 0.78% of impressions (0.69% of spend) using three MRC-accredited measurement companies' categorizations, and left agency fees and brand safety outside its cost waterfall.3, 4commissioned measurementsupported with qualificationhigh
Adalytics' Google Video Partners report (public by 27 June 2023) argued that YouTube Measurement Program vendors such as IAS and DV rely on aggregated server-to-server 'YouTube data feed' transfers, and a follow-up post (July 2023) relayed MRC's statement that no third party was audited or accredited for Google brand safety; Google's rebuttal cited DV's and IAS's 93% GVP viewability as independent contradiction, and IAS's own data for January to early July 2023 put GVP at 3.4% of its YouTube and GVP impressions with 92.74% viewability and 0.40% IVT.3, 5synthesissupported with qualificationmedium
Oracle retired Moat rather than selling it: its Advertising EOL FAQ (Version 3.0, 12 August 2024) says Oracle would end-of-life all Oracle Advertising products and support, including Moat Analytics, Moat Reach and Oracle Contextual Intelligence, effective 30 September 2024, after notices in the week of 17 June 2024, and would enter into no new agreements.3, 6, 7vendor assertionsupported as statedhigh
HUMAN's brand-safety page (captured 4 February 2026) says pages, apps or impressions are classified by AI with explainable reasons and benchmarked against 400K+ expert-labelled web pages with 'high accuracy', without publishing an accuracy figure, framework or edition, category definitions or languages; its GlancelotBot crawler retrieves page content for brand safety, suitability and contextual signals.3, 8vendor assertionsupported as statedhigh
Adalytics' report of 7 August 2024, begun when a multinational brand told that its campaigns were '100% brand safe' asked for a review, showed IAS's publisher-side endpoint returning 'veryLow' or 'low' on all seven brand-safety keys (meanings per IAS's own implementation guide) for explicit pages, mostly Fandom wikis, and ads carrying DoubleVerify or IAS code, including DSP charge strings for DV and IAS brand safety, on such pages; it also showed the same endpoint labelling mainstream news pages 'medium' or 'high'. It used a convenience sample whose explicit-page examples are undated and whose dated examples run from April 2023 to July 2024, and it states that it makes no claim about any classifier's sensitivity, specificity or accuracy.3, 9independent measurementsupported with qualificationhigh
Adalytics' report of 7 February 2025 reviewed thousands of URLScan.io recordings of imgbb.com and ibb.co from 2017 to 2024, relayed several advertisers' statements that their brand-safety vendors had marked 100% of measured impressions there brand safe or suitable, showed ads carrying DoubleVerify or IAS tags on explicit adult pages, and says advertisers could not count ads beside child sexual abuse material for lack of page-level reporting.3, 9independent measurementsupported as statedhigh
Both vendors answered on 7 February 2025 with denominators the report lacked: DoubleVerify put its customers' impressions on imgBB.com at 0.000047% of the media transactions it measured over 30 days, said in its original statement (dropped from the live text on 15 October 2025) that 83% of those measured impressions were served to the site's home page, and said pages are classified only once they reach a threshold of ad traffic; IAS said the two domains accounted for 88,000 of nearly 1.9 billion webpages it observed across eligible ad opportunities in 60 days, and labelled them ineligible for monetization.3, 9vendor assertionsupported with qualificationmedium
On 13 February 2025, six days after the report, DoubleVerify launched a default-on "Highly Illicit: Do Not Monetize" category built from three years of public NCMEC data on over 300 domains and services, of which it said roughly 100 were already covered by its classifications, so a category aimed at CSAM-notified domains was not sold during the report's 2017 to 2024 observation period, while its "Adult & Sexual" category was.3, 9vendor assertionsupported as statedmedium
Adalytics, which sells analytics to advertisers and self-declares that no publisher or ad-tech vendor funds its research, examined in its report of 28 March 2025 ads served to three crawler bots (URLScan.io, HTTP Archive and one undisclosed) in datasets spanning over a petabyte, over two million websites and over seven years, while its statement that advertisers were billed for such impressions rests on non-public log files and invoices from work for Fortune 500 advertisers, and its caveats disclaim any finding of fault, magnitude or prevalence.3, 9independent measurementsupported as statedhigh
IAS answered on 28 March and 3 April 2025 (after a pre-emptive methodology post of 21 March) that the report evaluated its publisher tag, not its advertiser tag, and that this tag by design does not block ads during post-bid processing, while saying publishers see the same measurement data it reports to buy-side customers; that it flagged HTTP Archive traffic as GIVT and classified URLScan.io impressions as GIVT, SIVT or reduced-value inventory (a non-IVT category in IAS's own definition), relying on impression-level thresholds to confirm fraud; and that outside analysis cannot read its back-end adjudication, without publishing counts for the sessions in question.3, 9vendor assertionsupported with qualificationhigh
MRC's statement of 11 April 2025 said that, at that date, IAS's publisher tools and pre-bid IVT detection, and DoubleVerify's publisher tools and Scibids product, were not audited or accredited (MRC's accredited list read on 10 November 2026 now shows IAS accredited for pre-bid IVT detection on desktop, mobile web and in-app, while the publisher tools and Scibids remain unlisted); that MRC requires post-serve filtration of general invalid traffic but not pre-bid filtration; that HTTP Archive is on the IAB bots list while URLScan.io's bot does not declare its user agent (MRC adds that its referrer domain is declared); that it had not tested each of the report's observations; and that it has no insight into billing.3, 9direct recordsupported with qualificationhigh
DoubleVerify answered the bot report in advance on 10 January 2025 and again on 28 March and 7 April 2025: general invalid traffic that reaches an ad is removed from billable counts post-bid, URLScan.io is not a declared bot, the cited 'DV source code' is DSP code, and of the report's 115 examples 62 had no DV measurement, 10 (37 impressions) fell outside DV's data retention, and all 43 it could match were identified and invalidated post-bid; DV published these example counts but no per-impression data.3, 9vendor assertionsupported with qualificationhigh
No public HUMAN response to the 28 March 2025 report was found in the searches run on 10 November 2026, and the report's HUMAN observations are ads served by platforms that announced HUMAN pre-bid partnerships, which MRC says receive HUMAN's IVT suggestions through an API and make the bid decisions themselves.3, 9synthesissupported as statedmedium
Adalytics' Forbes report (public by 4 April 2024) observed page URLs truncated and a hidden www3 subdomain mis-declared as www.forbes.com in bid requests sent through Media.net's Prebid server to four SSPs until 22 March 2024, with verification vendors' publisher code absent from www3 while ads there loaded IAS, DV, Oracle Moat or HUMAN tags; DoubleVerify said it had classified that inventory as low-tier MFA before publication.3, 9synthesissupported as statedmedium
In DoubleVerify v. Adalytics (D. Md. 8:25-cv-01535), filed on 12 May 2025 over Adalytics' report of 28 March 2025, the court on 27 April 2026 dismissed only the tortious-interference count, without prejudice, on allegations accepted as true, holding that DoubleVerify had not shown the report's statement that some ads reached bots pre-bid was false but had plausibly alleged defamation by implication; Adalytics answered on 11 May 2026 pleading truth and Maryland's anti-SLAPP statute, and the next status report is due on 9 February 2027.3, 9direct recordsupported as statedhigh
On 7 February 2025, the day Adalytics' image-host report appeared, Senators Blackburn and Blumenthal sent six letters (Amazon, Google, DoubleVerify, IAS, MRC and TAG), having received the report before publication; the letter to DoubleVerify asks five questions, including how verified ads appeared on a site known to host CSAM and how much revenue DoubleVerify derived from measuring ads on such sites, with answers requested by 14 February 2025.3, 9direct recordsupported as statedhigh
In letters dated 24 March 2025 and released on 28 March 2025, Senator Warner asked the FTC to investigate whether IAS, DoubleVerify and HUMAN made false bot-filtering claims (answers requested by 28 April 2025) and asked the Justice Department to investigate knowing misrepresentation to federal clients and False Claims Act liability; the FTC letter itself says the failures discussed are GIVT rather than advertising fraud under industry definitions, and no FTC or DOJ action concerning these vendors was found in the sources screened by 10 November 2026.3, 9direct recordsupported as statedmedium
Of the fourteen blog URLs in Adalytics' sitemap with lastmod in 2023 or later (the image-host report has both a summary and a full-report URL), six reports name scored vendors (made-for-advertising sites in January 2024; the Forbes subdomain; Colossus user-ID rotation, which names HUMAN only as the SSP's declared partner; 'AI' brand safety of 7 August 2024; the image-host report of 7 February 2025; the bot report of 28 March 2025), all naming DoubleVerify, IAS or HUMAN and one also Pixalate; none names Zefr, Peer39, Adloox or Protected Media, the bot report mentions comScore only in passing, and the reports read carry disclaimers against causal or magnitude claims.3, 9synthesissupported with qualificationmedium
Adalytics' made-for-advertising report (public by 18 March 2024) listed Oracle Moat, IAS, DoubleVerify, HUMAN, Pixalate and Fou Analytics among the verification platforms observed monitoring ads on MFA sites in January 2024, after DoubleVerify and IAS had announced MFA products in late 2023, and noted that some verification code is injected by SSPs and DSPs rather than chosen by the advertiser; DoubleVerify answered on 7 August 2024 that it had already classified virtually all of the sites as MFA.3, 9independent measurementsupported as statedmedium
Coded by estimand, the independent record screened to 10 November 2026 yields prevalence in samples (the ANA study, the benchmark and its July 2026 AI-slop analysis, using the vendors' and data suppliers' classifications), failure-mode cases (ten Adalytics reports, June 2023 to March 2025, and the manually found cases in Smith et al.) and agreement between vendors (Smith et al. 2026; Adalytics 2020); its one candidate product error rate, the March 2025 report's shares for IAS's publisher-side 'fr' flag (an IVT label by IAS's own documentation), publishes no session counts and concerns a tool MRC does not accredit and the seed register does not score, so H3 is supported on this record only while that tool stays off the scored register; if workstream U scores it, the registered rule for an undisclosed denominator makes H3 unresolved.3, 12inferencesupported with qualificationmedium
The ANA's Programmatic Media Supply Chain Transparency Study analysed log-level data from 21 ANA member advertisers (of 67 that wanted to take part), covering $123 million of open-web programmatic spend and 35.5 billion impressions from September 2022 to January 2023, with three DSPs, six SSPs and three verification companies (DoubleVerify, IAS and Oracle Moat) supplying data but taking no part in analysis.3, 13independent measurementsupported as statedhigh
In the vendor and commissioned evidence screened on 9 November 2026 (publications by or with DoubleVerify, IAS, HUMAN, Pixalate, Zefr, Peer39 and Channel Factory, plus TAG, Juniper, CHEQ, ANA-White Ops and Forrester studies), no candidate product's own or commissioned publication reports a miss rate, false-positive rate or precision for a category it sells, measured on a labelled holdout with a denominator; the only labelled test with denominators (Zefr's, August 2025) tests general-purpose LLMs, not a Zefr product.3, 13synthesissupported with qualificationmedium
The Trade Desk's Inventory Supplier Terms and Conditions (last updated 24 September 2020) carry the same conclusive-report clause for supply partners (clause 4.5) and let the parties agree in writing that a third party's records, such as a fraud or viewability vendor's, are the controlling measurement for invoicing guaranteed placements (clause 5.1).4direct recordsupported as statedhigh
The Trade Desk's Advertising Terms (last updated 21 May 2026) give The Trade Desk sole responsibility for calculating the metrics that are used to calculate advertisers' payments (clause 5.1), and the page text read on 10 November 2026 contains no invalid-traffic, SIVT or viewability provision.4direct recordsupported as statedhigh
The Trade Desk's SIVT Credit Program, described in a blog post dated 1 May 2018 (Internet Archive capture of 6 June 2023), required client-initiated claims to exceed US$1,000 of media spend, arrive within 30 days of the incident and include a report verified by an MRC-accredited SIVT vendor such as IAS or DoubleVerify, but issued the credit only after the relevant supplier approved and credited back the amount; by the capture of 1 December 2023 the page had been replaced by a redirect (content entry updated 20 October 2023), so the public description was withdrawn in late 2023 and whether the programme still runs is unknown.4vendor assertionsupported with qualificationmedium
Magnite's Buyer Master Terms and Conditions (last updated 17 June 2026), section 2.3, let a buyer withhold or dispute invoiced amounts for invalid activity only after written notice with reasonable evidence within 10 days of the invoice, and if the parties have not resolved it within 30 days Magnite's own determination is dispositive.4direct recordsupported as statedhigh
Yahoo's Advertising Terms and Conditions (last updated 1 January 2026) make Yahoo's measurements definitive for charges and exclude from charges the clicks or impressions Yahoo determines are fraudulent; self-service Yahoo DSP buys add the DSP Program Terms, which contain no invalid-traffic, refund or dispute provision, while managed Yahoo DSP buys placed by insertion order fall under the same document's Display Advertising Program Terms (item 14).4direct recordsupported with qualificationhigh
In the Display Advertising Program Terms of the same Yahoo document, vCPM line items carry a viewability guarantee under which the billable count is the viewable share, determined by Yahoo or by an MRC-certified Approved Third-Party Vendor, multiplied by delivered impressions, with Yahoo reserving a right to dispute vendor measurements and the parties agreeing to reconcile within 10 days (clause 10.1).4direct recordsupported with qualificationhigh
Google Ads says advertisers are not charged for invalid traffic detected before billing and are credited 'where appropriate and possible' for invalid traffic found later, limits advertiser-requested investigations to the past 60 days, and states that charges are based solely on Google's metrics, noting that third-party tools may flag as invalid traffic that Google has already filtered from the bill.4vendor assertionsupported as statedhigh
The US Google LLC Advertising Program Terms give a third party's numbers one billing role, in the text dated 29 July 2025 and unchanged in the text dated 1 July 2026 that Google's payments info finder serves for US Google Ads accounts: when Google's impression count for CPM or vCPM display ads exceeds the advertiser's third-party ad server count by more than 10% and the gap is unresolved, a claim within 60 days that Google determines valid earns credits of 90% of Google's count minus the ad server's count, times the average CPM or vCPM (section 3(c)).4direct recordsupported with qualificationmedium
In Display & Video 360, every exchange partner has agreed to refund traffic that Google considers invalid, with HUMAN integrated as an extra check, while third-party verification is sold as pre-bid brand-safety and fraud targeting at prices the verification providers set.4vendor assertionsupported as statedhigh
Google Ad Manager's publisher help says Google deducts publisher earnings for invalid clicks and impressions it detects and reimburses the advertisers who paid, credits advertisers for invalid traffic accrued within 60 days before the invoice date, allows publishers no appeal, and does not bill Ad Manager 360 fees on impressions identified as invalid.4vendor assertionsupported as statedhigh
The one advertiser settlement over invalid-traffic credits found in this stream, AdTrader v. Google (N.D. Cal. No. 17-cv-07082-BLF), received final approval on 1 November 2022 for a non-reversionary US$7 million fund for US AdWords advertisers charged for clicks or impressions on AdX publisher sites that were not refunded although Google withheld the publisher's payment for invalid activity or other breach, including policy violations, so the money moved on Google's own determinations rather than a third-party verdict.4direct recordsupported with qualificationhigh
Amazon's Advertising Agreement (posted 23 July 2026) bases fees solely on Amazon's measurements and has the customer accept the risk of improper third-party impressions or clicks (section 4.A), while its Traffic validation help page (revised 2 November 2026) says invalid traffic detected within 30 days is removed from invoices, invoices may be restated after an investigation, and a gap of up to 5% from third-party invalid-traffic rates is expected.4direct recordsupported as statedhigh
Meta's Self-Serve Ad Terms (effective 22 August 2025, with a preview of updated terms on hold) calculate the amount billed on Meta's own tracking mechanisms and disclaim responsibility for click fraud and other potentially invalid click activity, and Meta's help centre says advertisers do not pay for impressions that Meta determines to be invalid.4direct recordsupported as statedhigh
TikTok's Ad Services Payment Terms (last revised September 2025) base all metric-based billing on TikTok's measurements or data (clause 2.3), its Advertising Terms (last revised January 2026) make no commitments about fraudulent, invalid or non-human activity or about content adjacency (clause 3.8), and its Ad Serving Policy (October 2024) lets a third-party analytics company's impression or click counts on reservation (insertion-order) display ads trigger a joint data review only when they differ from TikTok's by more than 20%, with TikTok's data final if the review finds no error.4direct recordsupported with qualificationhigh
Snap (Payment Terms effective 1 April 2024), Pinterest (US Advertising Services Agreement, April 2025), X (US Master Services Agreement, 14 August 2026) and LinkedIn (Ads Agreement, 3 November 2025) each bill on their own measurements, which Snap, Pinterest and LinkedIn call the sole basis and X calls definitive, with charge disputes waived after 60 days at Pinterest and X and 90 days at LinkedIn, and LinkedIn's sole remedy for invalid activity is an Ad Services credit at its discretion.4direct recordsupported as statedhigh
Reddit's Advertising Services Agreement (effective 1 January 2026) makes Reddit's measurement the controlling measurement for billing and rules out make-goods (section 6), and its Business Tool Terms (effective 7 August 2025) say charges are never based on third-party reporting or metrics even when the advertiser uses a Reddit-approved measurement partner (section 4).4direct recordsupported as statedhigh
The ANA's Master Media Buying Services Agreement Template, Version 3.0 (June 2023, downloadable without sign-in), asks the agency to make media owners and resellers accept the advertiser's ad-verification tool as the basis for measuring the billable metric and to seek terms under which the advertiser does not pay for invalid traffic, brand-unsafe or non-viewable placements that tool identifies, with refund, credit or makegood at the advertiser's discretion (Schedule 10, sections 3.4, 4.1, 5.6 and 6).4direct recordsupported as statedhigh
IAB's Measurement Services Addendum v1.0, posted as final on 28 October 2026 for contracts with measurement and verification providers, provides for the case where a provider's outputs are designated the controlling measurement under a related agreement (requiring reports under both old and new methodology for the affected invoice period) but sets no default that makes a verification report decide billing.4direct recordsupported as statedmedium
Disney Advertising's addendum to the 4A's/IAB Standard Terms v3.0 (dated 28 March 2023, US media buys) says third-party ad verification vendor reports will be used to determine a campaign's compliance with editorial adjacency guidelines, while Disney reserves the right to review the incident list before issuing a remedy and keeps its own ad server as the controlling measurement unless the IO states otherwise.4direct recordsupported as statedhigh
Older and newer seller-side standard terms keep billing on delivery counts rather than verification verdicts: the 4A's/IAB Standard Terms v3.0 make removal with makegood, credit or non-billing the sole remedy for adjacency violations without saying who determines a violation and take the controlling measurement from a compliant third-party ad server, and Spotify's Master Advertising Terms (last updated 1 July 2026) let Spotify's own metrics control over an approved third party's statistics unless an IO says otherwise.4direct recordsupported as statedhigh
In the UK government's RM6123 Media Services framework (agreement dated 14 December 2021, expired 13 June 2026), the Lot 1 and Lot 2 specifications say that where ad fraud tools are employed the public-sector client pays only for impressions, clicks or conversions the tool verifies as human traffic, while the agency must use reasonable endeavours to recover the value of fraudulent, non-viewable or unsafe ads it has paid for, in cash, credits or other compensation, and pass a proportionate share to clients.4direct recordsupported as statedhigh
The 4A's/IAB Standard Terms and Conditions for Internet Advertising, Version 3.0 (footers dated 2010), make the count of an ad server certified compliant with the IAB/AAAA measurement guidelines the controlling measurement for invoices when both parties track delivery (the agency's third-party ad server controls when both comply and it offers an automated daily reporting interface); where the controlling count is more than 10% lower over the invoice period and reconciliation fails, the agency may treat the gap as under-delivery with a makegood or pay on the controlling count plus a 10% upward delivery adjustment; and for violations of an IO's editorial adjacency guidelines the advertiser's sole remedy, on written request, is removal plus makegoods, a credit or non-billing. The template contains no viewability, invalid-traffic, fraud or verification term.4direct recordsupported with qualificationhigh
The ANA's Complete Report of December 2023 tells advertisers to pre-agree how invalid traffic will be remediated (refunds, make-goods), not to pay for impressions with IVT or for impressions that are not measurable or not viewable, and to settle with DSPs in advance how refunds, make-goods and credits for fraud are processed, while naming the study's three verification participants, DoubleVerify, IAS and Oracle Moat, as vendors that 'should at least be considered'.4direct recordsupported as statedhigh
The WFA's Global Media Charter of 21 June 2018, developed with more than ten large advertisers including P&G, Unilever, Mastercard and Diageo, demands a streamlined process to refund all media investment, fees and commissions included, associated with invalid traffic, says advertisers seek accredited third-party verification to assess fraud exposure, and calls self-reported data unacceptable.4direct recordsupported as statedhigh
TAG's Brand Safety Certified Guidelines v5.0 (July 2026) require every certified Direct Buyer, which includes a brand-appointed agency, to define procedures covering make goods and reversals or clawbacks and to screen 100% of the monetizable transactions it handles with independently validated content verification services or with inclusion and exclusion lists, so verification is one of two permitted means, not a requirement.4direct recordsupported as statedhigh
On 10 November 2026 the TAG Registry showed Brand Safety Certified seals in the buyer category for media entities of each of the six holding companies in this workstream's scope (IPG now within Omnicom), independently validated in each case but with scope ranging from global (Publicis Groupe/MMS USA Holdings, IPG Mediabrands) to the UK alone (Omnicom Media Group), the UK and Ireland (Havas Media, Dentsu Aegis Network) and Asia Pacific, the UK, the US and Ireland (WPP Media), and Certified Against Fraud seals that were self-attested for WPP Media and Omnicom Media Group.4direct recordsupported as statedhigh
IAS's FY2024 10-K, its last, lists Dentsu, IPG, Omnicom, Publicis Groupe, WPP and Havas as global agencies through which it serves advertisers, alongside DSPs (Amazon, Display & Video 360, The Trade Desk, Yahoo), and says its long-term revenue-share agreements with DSPs including Google and The Trade Desk pay those platforms a percentage of the revenue IAS receives from buyers using its products through them.4direct recordsupported as statedhigh
In Adalytics' report of 7 February 2025, an unnamed advertising executive said that finding ads on a site hosting child sexual abuse material would most likely mean pausing spend across all clients and requesting refunds, and the report says advertiser data confirmed invoicing for ads on imgbb.com or ibb.co, but no refund, credit or settlement is documented; IAS answered that Adalytics could not tell whether its tags were used for blocking or measurement.4independent measurementsupported as statedmedium
The FTC's final Decision and Order of 26 September 2025 (Docket C-4823; ten-year term, a compliance monitor for five years, annual compliance reports) in Omnicom's acquisition of IPG bars Omnicom, in the United States, from directing advertisers' spend on 'Covered Bases' (political or ideological viewpoints, including labelling news as misinformation or disinformation; adherence to journalistic standards set by a third party; or diversity commitments, but not fraudulent content) through agreements with third parties, and requires Omnicom Media Group and Mediabrands not to rely on exclusion lists, inclusion lists or other ratings of publishers on those bases, while expressly permitting lists developed at a particular client's express direction, which Omnicom may not share with another client or a third party, and requiring its compliance reports to count how often publishers appear on client-directed exclusion lists based on political ideology; the complaint behind it alleged that agencies had coordinated, including through industry associations, on decisions not to advertise on certain websites; Omnicom completed the merger on 26 November 2025.4direct recordsupported with qualificationhigh
Omnicom's FY2025 10-K and WPP's FY2025 20-F contain no reference to brand safety, viewability, invalid traffic or ad fraud, use the word 'verification' only for audit matters (WPP) or not at all (Omnicom), and Omnicom's makes no mention of the FTC order; holding companies' verification practice is documented mainly by third parties (TAG's registry, vendors' filings and case studies, researchers' reports).4inferencesupported as statedmedium
In the ANA's Programmatic Benchmark Survey of client-side members (fielded 6 September to 5 October 2023), 78% said they own direct contracts with ad verification providers and 73% plan new ones, behind DSPs (80% and 77%) and ahead of SSPs (27% and 32%), and brand safety was the top-rated metric for programmatic campaigns (4.61), above viewability (4.41) and invalid traffic (4.19).4commissioned measurementsupported with qualificationmedium
The ANA/White Ops Bot Baseline release of 1 May 2019, a report from a fraud-detection vendor (White Ops, now HUMAN) and the ANA, said that for the first time most fraudulent impressions were being invalidated by DSPs or SSPs, filtered as SIVT before payment, or recovered after campaigns through claw backs, but gave no record of any claw back.4commissioned measurementsupported with qualificationlow
Inference: of the buyer-side terms screened, only RM6123 gives a verification tool's verdict (an unnamed ad fraud tool agreed with CCS and GCS, which the contract does not require to be independent or accredited) determining effect on what a buyer pays, and it does so between a public client and its agency; the 4A's/IAB template, TAG's certification rules (including Certified Against Fraud's 100% IVT-filtration duty) and the ANA, WFA, 4A's and ISBA guidance set procedures or demands, and no record of a settlement, refund or credit paid on a third-party verdict was found in the sources screened.4inferencesupported with qualificationmedium
DoubleVerify asserts that its post-serve verdicts change what advertisers are charged: its complaint against Adalytics says GIVT impressions are deducted from the billable count before a customer is billed and SIVT impressions are flagged so the advertiser can seek compensation from the seller, and its March 2025 statements say detected bot impressions are removed from billable counts reported to advertisers; MRC says it has no insight into billing arrangements, and DV's own Main Service Agreement makes DV usage statistics potentially determinative only for billing its own fees.4vendor assertionsupported as statedmedium
IAS's own Standard Terms make IAS's reporting and usage statistics 'determinative for billing purposes' for IAS's fees, make fees nonrefundable, give clients 30 days to dispute an invoice, and disclaim any warranty that the services are error-free; nothing in the terms read bars clients from using IAS measurements to settle media billing.4vendor assertionsupported as statedhigh
Since 4 November 2025 Pixalate sells a Clawback Report, a timestamped, digitally signed PDF of GIVT and SIVT ad counts by app or domain, seller, publisher, campaign and advertiser that it markets for refund negotiations, and since 6 October 2026 it labels apps and websites named in clawbacks filed through its Clawback Tool; neither source names a counterparty, an amount refunded or a paid settlement.4vendor assertionsupported as statedhigh
Pixalate's Master Subscription Terms (effective 12 February 2026) bar clients from using Analytics output in connection with litigation or advertising claims without Pixalate's prior written consent, call its IVT and viewability designations opinions rather than facts or guarantees, and bar clients from disputing their position in Pixalate's indexes.4vendor assertionsupported as statedhigh
DoubleVerify's FY2025 10-K says advertisers pay it a fee per thousand impressions measured, that media agencies pay it on behalf of multiple customers, that two programmatic partner platforms collected about 24% and 12% of its 2025 revenue on behalf of advertisers, and that agencies have significant control over which verification solutions their clients' campaigns use, some of them marketing competing products.4, 7direct recordsupported as statedhigh
ISBA's Responsible Media Guide (brand safety chapter V1, March 2025) tells UK advertisers that an ad verification partner such as Adloox, DoubleVerify, IAS, Mantis or Zefr is the principal way to manage brand safety, that the advertiser controls what the partner does and how it is paid even when engaged through the agency, and that buyers should ask whether they pay for blocked impressions; ISBA says its guidance feeds its Media Services Framework, the member-only contract template members use with agencies.4, 8direct recordsupported as statedhigh
The ANA's 2023 Complete Report defines a 'verified impression' as one on which two or more log sources (DSP, SSP, verification provider) agree, and its quarterly benchmark, built with TAG TrustNet and Fiducia on log-level data, added brand safety and suitability metrics from an unnamed 'AI-powered standard' in Q3 2025 reporting 99.1% of spend in low-risk environments, and counted 86 participating marketers (66 active contributors) in Q1 2026.4, 8commissioned measurementsupported as statedmedium
Rumble's antitrust suit against the WFA, Diageo, WPP plc and GroupM Worldwide (N.D. Tex. 7:24-cv-00115, filed 6 August 2024) was dismissed without prejudice on 13 August 2025 for lack of personal jurisdiction and improper venue, so no court reached the merits of the claims against the holding company.4, 8, 9direct recordsupported as statedhigh
In its May 2025 complaint against Adalytics, DoubleVerify states that its post-serve verdicts move money: GIVT impressions are deducted from the billable count before a customer is billed and SIVT impressions are flagged so the advertiser can seek compensation from the seller; it cites no contract, platform term or settlement, and the MRC says it has no insight into billing arrangements.4, 9vendor assertionsupported as statedmedium
Both public verifiers' filings acknowledge billing disputes with clients without naming any (DoubleVerify FY2025: clients have disputed how it calculates billing; IAS FY2024: it may be involved in disputes over billings), and CourtListener's RECAP index on 10 November 2026 showed no suit by an advertiser or agency against either by case name: the only party identifiable as buy-side was Sizmek DSP, whose bankruptcy adversary proceedings against DoubleVerify (20-01119) and IAS (20-01134), filed 3 June 2020, closed in 2021 with no document public.4, 9inferencesupported with qualificationmedium
The Trade Desk's OpenPath Publisher Terms (v.7.5.23 clause 4.5, unchanged as clause 4.4 of v.10.31.24, the newest version found) deem reports from MRC-accredited third-party services used by The Trade Desk or its advertisers conclusive in calculating the fees it owes a publisher, exclude 'Invalid Inventory' (including zero viewability and SIVT) from payment, and oblige the publisher to refund amounts already paid within 30 days of The Trade Desk's notice.4, 12direct recordsupported as statedhigh
Microsoft's documentation for its Invest DSP (last updated 4 April 2025) describes an invalid-traffic refund process in place since March 2018 in which a buyer's monthly claim, due within 10 days of month-end, must include reporting from an approved verification partner; Microsoft refunds the claimed impressions at an average CPM calculated from its own purchase and invoice data and adjusts the seller's account, and a seller may dispute the claimed amount through 'the representative or verification partner responsible for the relationship'. Microsoft announced on 14 May 2025 that it would stop supporting media buying through Invest from 28 February 2026; the pages read on 10 November 2026 do not confirm or date the retirement.4, 12vendor assertionsupported with qualificationmedium
PubMatic's Demand Partner Master Services Agreement (page modified 18 August 2026), section 8.4, credits a demand partner for impressions deemed fraudulent on an MRC-accredited vendor's scoring only if the partner uses such a vendor, reports within 30 days of month-end, claims more than US$500 for the month and PubMatic verifies the impressions in good faith, and it rules out makegoods for viewability or brand safety; PubMatic announced the programme as a money-back guarantee on 7 November 2017.4, 12direct recordsupported as statedhigh
The IAB's modular contract terms released on 18 May 2026 (General Terms v1.1, Direct Buy Addendum v1.0 and Appendix 1 v1.1) leave the controlling measurement to each order and, in the Direct Buy Addendum's section 18, which applies only where a third-party ad server is used, default to the ad server or measurement system that renders the ad; they allow reconciliation on viewable impressions if the parties agree, define an Ad Verification Company as a mutually agreed third party, and settle invalid-traffic and other compliance issues through notice, a ten-business-day correction period and remedies the parties mutually choose.4, 12direct recordsupported with qualificationhigh
On this stream's coding applied to H2's registered population (the 17 listed platforms' public terms plus every public agency or advertiser contract found), up to 26 entries are readable: the stream's 16 platform entries and 7 contracts, plus RM6123 (the UK public-sector media framework, expired 13 June 2026), the OpenX addendum and Index Exchange's Marketplaces Terms of Use. Counting all terms found, a third-party report has determining effect (component b) in 4 entries strictly (The Trade Desk's supply-side terms, Microsoft Invest's refund process, the ANA template and RM6123: 15.4% of 26 to 17.4% of 23), in 6 counting clauses qualified by a review or dispute right (Disney, Yahoo vCPM: 23.1% to 26.1%), and in 7 also counting PubMatic's verification-gated credit (26.9% to 30.4%), the range depending on whether OpenX and Index are counted and whether Yahoo's display clause is folded into Yahoo DSP. Counting only terms in force at reading (dropping RM6123 and Microsoft Invest, whose end of buying on 28 February 2026 Microsoft announced in May 2025), the shares are 8.3% to 9.5%, 16.7% to 19.0% and 20.8% to 23.8%.4, 12synthesissupported with qualificationmedium
Google's own viewability measurement sets billable quantity in two Google buying products read on 10 November 2026: Google Ads viewable-CPM advertisers on Display Network campaigns pay for impressions Active View measures as viewable (a small portion of served impressions might also be charged), and Display & Video 360 bills outstream video only on viewable impressions, after DV360 archived its pay-per-viewable-impression insertion orders on 1 August 2023.4, 13vendor assertionsupported as statedhigh
Adalytics' report of 7 August 2024 relays that a multinational brand was told by its media agency holding company, its DSP and its verification vendor that its campaigns were 100% brand safe, and reports ads from campaigns transacted by agencies of IPG, GroupM/WPP, Omnicom and Publicis on pages with explicit titles, some carrying codes the report reads as charges for brand safety services; IAS answered the same day that a script on a page does not show the advertiser's settings.4, 13independent measurementsupported as statedmedium
Meta's update of February 2025, as posted on Zefr's press page, states that in the previous 30 days more than 99% of content next to ads in Facebook and Instagram Feed and Reels was brand safe as measured by its brand safety and suitability partners, without defining 'brand safe' or giving a denominator or partner breakdown.5vendor assertionsupported as statedhigh
MRC's letter of 13 April 2026 continues accreditation of DoubleVerify's independent third-party reporting of Meta display and video impressions and viewability on desktop, mobile web and in-app, reported net of GIVT, only for Meta-measured and accredited placements in the Facebook and Instagram feeds, and MRC's list (read 10 November 2026) annotates the service as applying to Newsfeed placement types only.5direct recordsupported as statedhigh
Meta's Marketing API documents two partner-facing data products: a Content Delivery Reports API, gated by the brand_safety_third_party_partners capability, that returns for Facebook in-stream video and Facebook Reels overlay positions each content item's ID, creator ID and estimated summed ad impressions, hourly, for public content only, with a 15-day lookback; and a Business Third Party Partner Viewability Requests edge that returns a business's viewability reports by day or by hour.5vendor assertionsupported as statedhigh
Meta's own announcement page dates third-party brand-suitability verification on its feeds to Zefr (Facebook Feed 30 March 2023, Instagram Feed 31 July 2023, Reels 10 October 2023), DoubleVerify (17 January 2024), IAS (5 February 2024) and Adloox (24 October 2024), and describes Zefr's AI as labelling Feed content from video, image, text and audio, without stating what data Meta supplies; DoubleVerify and IAS announced content-level pre-screen controls on Facebook and Instagram Feed and Reels on 10 October 2024, expected for all advertisers in early 2025.5vendor assertionsupported as statedhigh
Ads Data Hub, which Google names for third-party Shorts measurement and which MRC's letters name as the data source for DoubleVerify's, IAS's and Zefr's accredited YouTube viewability, returns privacy-checked aggregates: Google says rows need roughly 20 unique users under noise injection, 50 under difference checks or 10 for click and conversion queries, and that its checks are designed to help prevent the transmission of data about individual users, so verifiers query Google-held event data without receiving event-level records (inference, though Google's page lists measurement partners among ADH users); MRC also lists a Google 'YouTube Relay Log' of aggregated views and watch time for third-party partners as in audit since June 2026.5inferencesupported with qualificationmedium
On YouTube the only MRC-accredited content-level services are Google's first-party Inventory Modes controls (in-stream video and, from the 28 May 2026 letter, Shorts) and Zefr's third-party labelling and reporting of English-language in-stream video and music video on desktop, mobile web, in-app and CTV (20 January 2026; Shorts, YouTube TV, Google TV, Masthead, Google Video Partner videos and volumetric data excluded), which MRC's chart calls 'Independent URL classification'; IAS's YouTube classification has been in audit since April 2023, and Zefr says clients reach its YouTube measurement through Ads Data Hub.5direct recordsupported as statedhigh
Channel Factory says it uses YouTube's API along with its own ViewIQ engine to scan video, audio and metadata and builds custom suitability inclusion and exclusion lists; read as a pre-bid list product (the page does not use the word), in which the vendor classifies content through the platform's API and its own scanning (how video and audio are obtained is not stated) rather than measuring delivered impressions; the page also claims unspecified 'real-time monitoring' and makes no MRC claim (read 10 November 2026).5vendor assertionsupported with qualificationmedium
TikTok's help pages (last updated September and November 2026) offer DoubleVerify and IAS invalid-traffic and viewability measurement on the For You Feed and TikTok Ad Network (IAS also on CapCut and Lemon8), requiring campaigns set up in DV Pinnacle or IAS tags generated on the IAS Signal Platform, while TikTok's list of supported impression and click trackers (January 2026) names neither verifier; MRC's letter of 13 April 2026 granted DoubleVerify initial accreditation for TikTok video impressions and viewability in mobile in-app filtered for GIVT and SIVT, and MRC removed TikTok itself from in-process status in December 2025.5synthesissupported as statedhigh
TikTok badged DoubleVerify, IAS and Zefr as brand-safety and suitability measurement partners on 19 May 2023, saying they had measured more than 26 billion impressions in over 20 languages in the prior year, and makes partner-managed video-ID exclusion lists its pre-bid control (up to 200,000 videos per addition, February 2026; Standard and Limited inventory only per its April 2025 playbook), with a partner-managed Profile Feed exclusion list 'coming soon' in April 2025; none of these pages states what content data TikTok supplies to the partners.5vendor assertionsupported as statedhigh
Amazon says its own systems, enhanced with signals from HUMAN Security, automatically exclude identified invalid traffic from reports and invoices, and MRC's letter of 29 May 2026 continues Amazon DSP's first-party measurement of display impressions and click-throughs net of GIVT for Amazon Retail, IMDb, APS and third-party exchanges and grants first viewability accreditation, net of GIVT, across those sources and Alexa devices; MRC's chart of 22 October 2026 shows no accreditation and no audit for Twitch or Prime Video video ads.5synthesissupported as statedhigh
X's own controls are account-level Sensitivity Settings (Standard by default, or Limited), up to 4,000 keyword and 2,000 account exclusions and exclusion of suspected automated accounts from monetizable audiences; X says it renewed TAG Brand Safety Certification in March 2026 and completed an MRC pre-assessment in December 2022 without a full audit, MRC's chart of 22 October 2026 marks X 'Platform Not in MRC Audit', and X's 2025 averages of 99.99% brand safety and 97% suitability are X's statements about partner reports.5vendor assertionsupported as statedhigh
Snap's help pages say third-party verification on Snapchat runs on approved vendors' tracking URLs, an impression tracker being a 1x1 tag that fires when the ad renders; they list IAS and DoubleVerify for viewability and attention on all five Snap ad products and IAS, DoubleVerify and Zefr for brand safety on Snap Ads Video, Commercials and Story Ads only, and say IAS brand-safety reporting requires IAS tags (no tagless integration), covers Creator Stories, in-between content and Spotlight, is charged per CPM and feeds floor classifications back to Snap.5vendor assertionsupported as statedhigh
Snap's own measurement is MRC-accredited only in mobile in-app (letter of 29 July 2026: impressions, viewability, swipe-ups, reach and video duration at the GIVT standard) after MRC denied desktop-web accreditation in September 2025 on Snap's withdrawal; MRC's chart of 22 October 2026 marks Story Ads and Sponsored Snaps first-party viewability and Snap's SIVT as in process and no third-party Snap measurement as accredited or in process, and DoubleVerify, which says it has measured viewability and IVT on Snap since September 2019, said on 5 November 2024 that its brand-safety integration with Snap was still being implemented.5synthesissupported as statedhigh
Pinterest, DoubleVerify and IAS document third-party viewability and IVT measurement on Pinterest since 2020 (IAS reports refreshed daily) and brand-safety measurement since 2024, but none of these pages states the data route (n/e); MRC accredits only Pinterest's first-party display impressions and pin clicks at the GIVT standard (20 January 2026), and its chart of 22 October 2026 marks Pinterest SIVT as in process for six display formats and no third-party Pinterest measurement as accredited or in process.5synthesissupported as statedhigh
Reddit offers a 'Limited inventory with Integral Ad Science' tier for managed accounts, and DoubleVerify (11 June 2024: content directly above and below in-feed ads and above Conversation ads) and IAS (3 April 2025: viewability and IVT on Feed, Conversation and Takeover) document measurement on Reddit, but Reddit's tracker page names ten approved tracker partners and neither verifier, its Business Tool Terms (effective 7 August 2025) treat measurement partners as Reddit-approved service providers to the advertiser whose metrics Reddit may request, and no page opened states how DV or IAS obtain Reddit data.5vendor assertionsupported as statedhigh
LinkedIn's help (updated about two weeks before 10 November 2026) says DoubleVerify measures post-bid IVT, viewability and intended geography on the LinkedIn feed and Audience Network but brand suitability only on the Audience Network, that IAS measures IVT and viewability on video ads only, and that LinkedIn accepts impression, click, video and image tags but not iFrame or JavaScript (DoubleVerify uses a site-served tag for video and a 1x1 image URL for display); MRC accredits only LinkedIn's first-party impressions and clicks, excluding the Audience Network (28 May 2026).5synthesissupported as statedhigh
DoubleVerify's FY2025 10-K (filed 26 February 2026) lists its social-platform integrations as Facebook, Instagram, YouTube, TikTok, X, Snapchat and Pinterest, says integration partners exercise significant control over its solutions on their platforms, that it relies on social and other partners to report usage of its solutions, and that data or signals from integration partners have in the past been inaccurate; DV's explainer of 30 May 2024 says its approach goes beyond simply accessing platform data: it embeds its technology, uses its own tags when possible and establishes server-to-server connections to confirm that the data meets its quality standards, and it needs access to all underlying event signals.5synthesissupported with qualificationhigh
Among the nine platforms, Google's YouTube Inventory Modes controls are the only first-party brand-safety and suitability controls holding MRC content-level accreditation on 10 November 2026 (in-stream video and, from the 28 May 2026 letter, Shorts), Meta's having been revoked in September 2025; the other platforms' tiers (Meta's inventory filter, TikTok's Inventory Filter with Category Exclusion and Vertical Sensitivity, Snap's Full, Standard and Limited suite, Reddit's inventory types, X's Sensitivity Settings, Pinterest's inventory filters) are unaccredited platform controls, and TikTok describes its first-party invalid-traffic handling as TAG Certified Against Fraud and 'powered by Protected by Mediaocean'.5synthesissupported as statedhigh
In the sources screened on 9 and 10 November 2026, every documented third-party pre-bid control inside the nine platforms is a verifier-supplied list, classification or tier that the platform executes (TikTok exclusion lists, Meta pre-screen and block lists, Reddit's IAS tier, IAS pre-bid on X's Vertical Video, IAS floor classifications sent to Snap, Channel Factory's YouTube lists), no page documents a verifier blocking an ad after the bid on a platform's own placement, and no page states a sampling rate; freshness is documented only for Meta (hourly or daily reports, 15-day lookback), Snap (trackers fire in real time) and IAS on Snap and Pinterest (daily).5synthesissupported with qualificationmedium
Protected Media's owner on 10 November 2026 is Mediaocean: Flashtalking bought Protected Media in 2021, Mediaocean agreed to buy Flashtalking on 13 July 2021, Mediaocean completed its acquisition of Innovid on 13 February 2025 and names 'Protected by Mediaocean' among its units, and Innovid's merger proxy describes Mediaocean as a portfolio company of funds managed by CVC Capital Partners and of TA Associates and Charlesbank Capital Partners.5direct recordsupported as statedmedium
Meta's own announcement page dates third-party brand safety and suitability verification for Facebook and Instagram Feeds and Reels through Adloox to its update of 24 October 2024, and a second Meta page (update of 27 March 2025) says advertisers can implement third-party content blocklists through Scope3, 'which acquired Adloox’s measurement operations'; Adloox's own pages add that it was badged as a Meta viewability partner on 20 February 2024, with access to 'shared and permissioned' viewability data for listed Facebook and Instagram placements, and list IVT among its Meta metrics; Peer39 adds Threads, which no Meta page opened confirms.5vendor assertionsupported with qualificationmedium
The French competition authority's decision of 4 May 2023 records how Meta's viewability and brand-safety partner programmes work: Meta collects, processes and supplies the data to partners through an API, removes general and sophisticated invalid traffic as Meta identifies it (applying partner-specific criteria) before passing raw impression-level data to viewability partners, lets brand-safety partners download delivery reports, and does not let partners collect raw data with their own technology. Adloox joined these programmes in 2024, so its Meta measurement runs on Meta-supplied data (inference for the post-2023 period, consistent with Adloox's 'shared and permissioned' wording).5synthesissupported as statedmedium
Amazon's approved-providers policy (last updated 1 June 2026) states that verification by DoubleVerify, Integral Ad Science and Adloox is available on Amazon DSP through server-to-server (S2S) integrations, and lists Adloox for viewability, brand safety and fraud or invalid traffic; it does not describe what data Amazon passes, and its tag table lists monitoring and 'Monitoring & Blocking' tags for all approved verification vendors without naming Adloox.5vendor assertionsupported as statedhigh
Inference: YouTube's derived-metrics policy (last updated 14 September 2026), for API clients that accept its amendment, permits API metadata to assess channels and videos for brand suitability and brand safety at the planning stage 'for the purpose of influencer marketing, vetting, and matchmaking', permits suitability scoring of Google-supplied post-campaign reports, and bars brand safety claims about specific impressions, which need impression-level data restricted to Brand Safety Reporting Partners (unnamed in the policy). Since Channel Factory says it uses YouTube's API and appears in none of Google's brand-safety measurement lists, its YouTube verdicts are documented only as pre-bid lists (refreshed every 4 hours, per the vendor, including during campaigns) and suitability reporting; whether building placement lists falls within the policy's planning purpose is not stated by either party.5inferencesupported with qualificationmedium
MRC's own records disagree on Amazon: the 22 October 2026 platform chart footnotes Amazon DSP server-to-server third-party viewability as 'Accredited for DV, In-process for IAS', but MRC's accredited list carries an 'Integral Ad Science Server to Server (S2S) Integrated Third-party Amazon DSP Reporting' row and no DoubleVerify Amazon row, and MRC's Q4 2025 report records the November 2025 grant to IAS.5direct recordsupported as statedhigh
Pixalate documents pre-bid filtering of GIVT and SIVT inside Amazon DSP, selectable per line item and configured and run by Amazon, and dated its availability to January 2022; Amazon's own pages opened do not name Pixalate.5vendor assertionsupported as statedmedium
On Meta, Meta's own pages date third-party brand-suitability verification through Zefr to Facebook Feed (30 March 2023) and Facebook and Instagram Reels (10 October 2023), describe Zefr's AI as labelling Feed content from video, image, text and audio against GARM suitability standards, and made third-party content block lists available through DoubleVerify and Zefr on 13 February 2025 (34 languages); Meta's Content Delivery Reports API gives partners approximate content-level impressions for public content only, with a 15-day lookback.5direct recordsupported with qualificationhigh
On Reddit, the only documentation found is Zefr's own: a release of 30 September 2026 announcing a partnership that gives advertisers 'more precise suitability controls' on Reddit, and a Reddit-specific contact page; no Reddit page opened names Zefr, and the joint microsite is script-rendered and marked noindex.5vendor assertionsupported as statedlow
The brand-safety rates published about campaigns Zefr measured are prevalence statements by interested parties, not error rates: Meta said early testing with Zefr found under 1% of Facebook Feed content in the high-risk GARM suitability category (March 2023); Zefr said it had measured a 99%+ brand safety rate on TikTok since 2022 on GARM floor definitions (September 2023); TikTok's playbook put misinformation adjacency below 0.1% on Zefr data for May to August 2024; and TikTok's case tile credits Rexona and Zefr with 99.9% brand safety and suitability rates.5vendor assertionsupported as statedmedium
MRC's 2018 supplement makes platform cooperation a condition of content-level accreditation for third-party verifiers: platforms either allow direct third-party content-level measurement or provide validated first-party data; on 10 November 2026 MRC lists a Google 'YouTube Relay Log' of viewership data provided to third-party partners as in audit, and Zefr describes its integrations as built on platform access rather than scraped data.5synthesissupported as statedmedium
MRC's chart of large digital platforms, updated 22 October 2026, marks third-party viewability as accredited only on Meta (Facebook Feed on desktop, mobile web and in-app; Facebook in-stream Reels video, Instagram Feed and Instagram Stories in-app, each footnoted 'DV, and IAS both accredited'), Google Ads Data Hub (YouTube and Google Video Partner in-stream, footnoted as integrated with Zefr, DV and IAS), Amazon DSP (server to server: Retail, IMDb, APS, third-party exchanges) and TikTok (DoubleVerify, video, in-app); as in process on Facebook in-stream Reels display, Facebook Reels and Instagram Reels; and as neither accredited nor in process on Pinterest, Snap, LinkedIn, YouTube Shorts, Fire TV, Twitch and Prime Video, with X shown as not in MRC audit and Reddit absent. Where the chart is broader than the letters (DoubleVerify on Instagram Stories and Facebook in-stream Reels; the vendors on GVP; DoubleVerify versus IAS on Amazon), the letters and list are the operative record.5, 6direct recordsupported with qualificationmedium
MRC accredits third-party processing of Google's Ads Data Hub YouTube data for DoubleVerify (13 April 2026), IAS (9 July 2026) and Zefr (6 November 2025) only for video impressions and viewability on desktop, mobile web and in-app at GIVT level (IAS's 'GIVT+' level is expressly not SIVT accreditation) and only for in-stream and bumper placements, excluding Shorts, in-feed video and masthead, while Google's own ADH accreditation (20 January 2026) covers 'third-party passthrough reporting' generated through vendor queries of ADH.5, 6, 12direct recordsupported as statedhigh
DoubleVerify (28 September 2021, known here from a syndicated Business Wire copy) and IAS (9 May 2022 for viewability and IVT; 17 October 2022 for post-bid brand safety and suitability in unnamed 'select markets') said they measure TikTok's own placements through TikTok's integration of IAB Tech Lab's Open Measurement SDK, a route in which the measurement provider's tag collects signals inside the app: verifier-executed observation under H4's comparator for viewability and IVT. No 2026 page names the route: TikTok's November 2026 page still requires IAS tags and DoubleVerify's April 2026 release calls its accredited TikTok reporting 'direct measurement', but MRC's April 2026 letter says DoubleVerify's TikTok traffic is segregated from its 'tag-based measurement'. For brand safety the OM SDK statement documents the impression-side tag, not how IAS obtains the adjacent TikTok content it classifies, so that leg is weaker.5, 6, 12synthesissupported with qualificationmedium
Amazon says it built a server-to-server approach that lets DoubleVerify and IAS apply their detection methods within Amazon's secure environment, and its approved-providers policy (last updated 1 June 2026) says verification by DoubleVerify, IAS and Adloox is available on Amazon DSP through server-to-server integrations and that JavaScript tags are not accepted for third-party tracking in the Amazon mobile app; MRC's letter of 7 November 2025 accredits IAS's server-to-server integration, including GIVT and SIVT filtration per the letter body (the enclosure labels every metric 'Net of GIVT'), for Amazon Retail and IMDb and the third-party supply sources it calls 'APD and 3PX', excluding Fire TV and Twitch.5, 6, 12synthesissupported with qualificationhigh
HUMAN (White Ops until it renamed itself on 30 March 2021) was acquired in December 2020 by Goldman Sachs Merchant Banking Division with ClearSky Security and NightDragon (terms not disclosed), raised a $100 million round led by WestCap on 4 January 2022, announced on 27 July 2022 a merger with PerimeterX under the HUMAN name, approved by both boards and with regulatory approval received (terms not disclosed; no completion date stated; with a $100 million Blackstone Credit debt facility), and raised $50 million-plus led by WestCap on 9 October 2024; no release opened states who controls it now.5, 7vendor assertionsupported with qualificationmedium
Peer39 announced on 2 June 2026 that it had acquired Adloox from Scope3, with terms not disclosed, and a filing records the completion: Adloox Limited's accounts, approved on 30 September 2026, state that in June 2026 Peer39 Tech LLC acquired Adloox S.A.S from Scope3 S.A.S and became the ultimate controlling party; the BODACC notice published on 5 July 2026 records P39 Technologies UK Limited replacing Scope3 PBC as president of the French company, and MRC continued the service's accreditation under the Peer39 name on 29 June 2026.5, 7, 8, 9direct recordsupported as statedhigh
IAS describes its social-platform measurement as impression-level, says it scores content directly adjacent to ads 'via our integrations with social platforms' and only content aligned with each platform's privacy guidelines, that methodology may vary by platform, and defines a brand-safety rating both as the percentage of ads passed for the GARM floor 'out of all ads served' and as the percentage of 'measured ad impressions' next to safe content, so platform-relayed '99%' figures are pass rates over the impressions and content each platform makes available for measurement (inference).5, 8inferencesupported with qualificationmedium
The WFA announced on 9 August 2024 that GARM would discontinue its activities, citing allegations that had drained its resources and finances, and on 29 July 2026, announcing the end of its litigation with X, stated that it will not form or restart GARM or a similar initiative, publishing no settlement terms; Rumble's parallel suit (N.D. Tex. 7:24-cv-0115) was dismissed without prejudice on 13 August 2025 for lack of personal jurisdiction and improper venue, the court not reaching the merits.5, 8, 9direct recordsupported as statedhigh
On 9 October 2022 Adloox complained to the French competition authority that Meta had refused it access to its viewability and brand-safety partner programmes since 2016; on 4 May 2023 the authority, on the state of its investigation and without deciding the merits, ordered interim measures requiring Meta to publish objective, transparent, non-discriminatory and proportionate access criteria within two months, with a re-examination route, and to answer an Adloox application within two weeks; the decision was not appealed, and no decision on the merits appears in the authority's decision list searched on 10 November 2026.5, 9direct recordsupported as statedhigh
Apostra (formerly Scope3; scope3.com now redirects to apostra.com) documents that its Interchange buyer platform began judging ivt and brand_safety performance standards against a vendor's reported metric for a campaign seller scorecard on 22 October 2026, a day after noting that 'Invalid traffic and brand safety are not scored yet'; the scorecard is a buyer-side quality score, not billing, and Apostra's 9.4 MB documentation corpus read on 10 November 2026 never mentions Brand Standards, the LLM-based suitability product Scope3 documented in 2025.5, 10vendor assertionsupported as statedhigh
MRC's letter of 9 July 2026 continues accreditation of IAS's third-party processing and reporting of Meta display and video impressions and viewability on desktop, mobile web and in-app net of GIVT only, limited to placements audited for Meta's own measurement (Facebook Feed; Instagram Feed and Stories and Facebook in-stream video in-app; Reels, Audience Network and others excluded), and records that, because of limits Meta imposes, IAS cannot independently identify and filter invalid traffic and has no user- or device-level data.5, 12direct recordsupported as statedhigh
On 10 November 2026 MRC listed a 'Meta Third-Party Brand Suitability Pipeline' as under review (audit reviewed 8/2026), described as defined data fields Meta delivers to third-party partners for text, image, audio and video in Facebook in-stream video and the Facebook and Instagram feeds, and MRC's platform chart of 22 October 2026 marks third-party content-level brand safety on Facebook Feed, Facebook in-stream Reels and Instagram Feed as in the accreditation process. On those placements third-party suitability measurement is documented as resting on data Meta supplies (inference from MRC's description; Meta's Content Delivery Report API also hands partners content IDs for retrieving video source files on in-stream and Reels-overlay positions); how partners obtain content on Feed Reels, Stories and Threads is not documented; and no third-party content-level accreditation exists on any Meta placement.5, 12direct recordsupported with qualificationhigh
Meta's own measurement holds MRC accreditation under the letter of 2 October 2026 for Facebook and Instagram ad impressions, video plays and first-party viewability report metrics at total net (including SIVT filtration) by placement, with initial grants for Facebook and Instagram Reels, after MRC revoked Meta's content-level brand safety and suitability controls and Brand Safety Error Rate in September 2025 on Meta's withdrawal and, in September 2026, revoked Facebook Video Feed metrics and Instagram Feed video viewability on desktop and mobile web.5, 12direct recordsupported as statedhigh
Google's help pages say Google does not accept third-party tracking pixels for YouTube measurement, that enabling a measurement vendor means sharing campaign data with that vendor, and that third-party partners measure YouTube Shorts reach and viewability through Ads Data Hub; Display & Video 360 lists DoubleVerify, IAS and Zefr as its YouTube viewability and brand-safety vendors, while its brand-suitability help lists third-party pre-bid verification without saying it applies to YouTube.5, 12vendor assertionsupported as statedhigh
Coded per platform at 10 November 2026 under the registered rule, H4 is contradicted on one platform: DoubleVerify (2021) and IAS (2022) documented measuring TikTok's own placements through TikTok's Open Measurement SDK integration (viewability and IVT; IAS also post-bid brand safety, though how IAS receives the adjacent content is not stated), and TikTok's November 2026 page still requires IAS tags, while MRC's April 2026 letter separates DoubleVerify's TikTok reporting from its 'tag-based measurement'. Platform-supplied routes are documented for Meta's feeds and in-stream, YouTube and Amazon (server to server, alongside JavaScript monitoring tags on Amazon DSP-hosted creatives whose supply Amazon does not name); Snap and LinkedIn document vendor tracking pixels or URLs (Snap auto-applies DoubleVerify's; LinkedIn refuses JavaScript and iFrame tags), so no verifier code is documented running in their placements, but neither says whether the trackers fire from the device; X, Pinterest and Reddit name partners without a route (n/e); and no platform or verifier documents an independent crawl of the content next to ads.5, 12inferencesupported with qualificationmedium
X's help pages (read in the browser pane on 10 November 2026) name IAS and DoubleVerify for third-party viewability and brand-safety measurement and for post-bid adjacency measurement in the Home Timeline in English- and Japanese-language markets (IAS also on Profile from 2026), with adjacency defined as one organic post above and one below the ad; DoubleVerify says it processes and classifies the Tweets above and below ads (25 January 2023), but no page opened states how those posts reach the verifiers (n/e on H4), and IAS's pre-bid optimisation runs only on X's Vertical Video surface.5, 13vendor assertionsupported as statedhigh
Peer39 states in an 11 August 2026 blog that roughly 60% of CTV bid requests lack usable program-level signal (no source given) and, citing its own benchmark data on authenticated CTV inventory, that where such signal is present 97% of impressions fall within standard TV ratings; neither figure comes with a sample, period or method.6vendor assertionsupported with qualificationmedium
Of 17 CTV units (product x metric family) that this stream could document as sold or accredited, 14 hold MRC accreditation on 10 November 2026 (all 10 GIVT and SIVT units of DoubleVerify, IAS, HUMAN, Pixalate and Protected Media; DoubleVerify's and IAS's video viewability, both limited to certified apps; DoubleVerify's property-level ad verification; Zefr's YouTube content-level reporting), while 3 brand-safety and suitability units are unaccredited (DoubleVerify's program-level Do Not Air and Verified Streaming TV, IAS's YouTube brand safety, listed as under audit 'as of 04/2023', and IAS's other CTV safety and suitability claim).6synthesissupported with qualificationlow
The same guidance requires measurement vendors to verify that providers declaring SSAI are SSAI providers and to treat their traffic as unknown unless it is discretely measurable for IVT, allows proxy and data-centre traffic to be filtered only when known to be invalid through direct signals, warns that SSAI artifacts may be spoofed and that CTV may be more vulnerable to ad fraud, partly because measurers' CTV telemetry may be more limited, and encourages ads.cert and an SSAI-provider certification process that it proposes rather than establishes.6direct recordsupported with qualificationhigh
MRC's IVT 2.0 addendum (June 2020) requires SIVT vendors to consider mobile in-app and OTT traffic discretely in risk assessments, business-partner qualification and parameter setting where it is a material portion of measured traffic, because SIVT risks and detection assets differ there; its 2024 interim update (published 24 April 2024) says CTV bundle IDs should be included in app-misrepresentation requirements and that non-CTV apps misrepresented as CTV should be considered SIVT, and encourages use of ads.txt and app-ads.txt telemetry.6direct recordsupported with qualificationhigh
IAB Tech Lab's OM SDK first supported CTV in version 1.4 (2022: Android TV and tvOS, with a display-connection signal for TV-off), added Samsung Tizen and LG webOS through the Web Video SDK in version 1.5 (announced 30 May 2024) and device attestation for Fire TV and Apple devices in 1.6.0 (29 October 2025); the JavaScript client reached 1.6.12 on 2 November 2026 with no change from 1.6.11, and no Tech Lab page opened names Roku or Vizio support, which an August 2022 Tech Lab deck listed as not supported.6direct recordsupported with qualificationhigh
On 10 November 2026 IAB Tech Lab's OM SDK compliance API listed 341 active certified integrations from 134 companies (and 125 expired records), of which 26 active records from 12 companies carried a CTV platform label (11 CTV-Android, 10 CTV-iOS, 4 CTV-Web, 1 CTV); these count certified SDK and player integrations, not apps or impressions, and the only CTV coverage share found is Tech Lab's unsourced statement that Samsung and LG support brought OM SDK to 40% of the CTV market (30 May 2024).6direct recordsupported with qualificationhigh
MRC's verifier letters mention OM SDK only for non-CTV environments: DoubleVerify's desktop, mobile web and in-app viewability accreditation is inclusive of traffic measured through OM SDK (13 April 2026) and IAS's in-app measurement includes OM SDK and MRAID with MRAID impression and viewability metrics out of scope (9 July 2026), while DoubleVerify said on 4 December 2025 that it now supports OM SDK for CTV to scale its CTV viewability coverage.6synthesissupported as statedhigh
IAB Tech Lab's app-ads.txt extends authorised-seller files to connected-TV app stores and works by crawling store listings and developer websites, with CTV inventory sharing handled by a partner-domain variable, while Tech Lab's CTV Programmatic Guide (published 11 December 2025, updated 30 January 2026) treats device spoofing as a separate problem because device information in bid requests and beacons is a text string that can be easily manipulated.6direct recordsupported as statedhigh
IAB Tech Lab's VAST 4.3 (December 2022) tells ad-stitching providers to send the device's IP address and user agent in X-Device-IP and X-Device-User-Agent headers so that server-side tracking is not mistaken for fraud, and White Ops' ICEBUCKET report of April 2020 described these headers as the limited information available in SSAI and as simple to falsify.6synthesissupported as statedhigh
White Ops (now HUMAN) disclosed ICEBUCKET on 16 April 2020 as an SSAI-spoofing operation that near its peak in January 2020 made up nearly 28% of the programmatic CTV traffic White Ops could see, about 1.9 billion ad requests a day, and 66% of the programmatic CTV-related SSAI traffic it protected that month, and that impersonated more than 2 million people in over 30 countries; the TAG and White Ops briefing of 30 April 2020 captioned the 28% 'Observed amount of global CTV traffic affected', dropping 'programmatic' and the visibility basis.6vendor assertionsupported with qualificationmedium
Oracle, then owner of the Moat verification service, announced on 17 December 2020 that StreamScam had spoofed more than 28.8 million U.S. household IP addresses, about 3,600 apps and 3,400 CTV device models by having its own servers send ad-impression events to Moat and advertisers without delivering ads, an operation Oracle called larger than ICEBUCKET.6vendor assertionsupported as statedmedium
For CycloneBot write 'capable of generating up to 250 million falsified ad requests and spoofing about 1.5 million devices a day'. Replace the limitation 'no other release states its dollar method' with 'DV's November 2025 release with Roku says only that the $7.5 million a month rests on industry CPM estimates; no release states impressions billed or a period total'.6vendor assertionsupported with qualificationmedium
HUMAN disclosed four CTV-related operations with its own scale figures: PARETO (21 April 2021; nearly one million infected Android phones spoofing more than 6,000 CTV apps at an average 650 million ad requests a day), BADBOX (4 October 2023; more than 74,000 infected phones, tablets and CTV boxes), BADBOX 2.0 (5 March 2025; more than one million infected devices, which HUMAN called the largest CTV botnet ever disclosed) and NewsJunkie (7 July 2026; up to nearly two billion invalid CTV bid requests a day per seller at peak).6vendor assertionsupported with qualificationmedium
In mobile in-app inventory White Ops and HUMAN disclosed CopyCatz (January 2021; 164 Google Play apps with more than 10 million downloads), VASTFLUX (19 January 2023; more than 1,700 spoofed apps on nearly 11 million devices at a peak of 12 billion ad requests a day, stacking invisible video ads), PEACHPIT (4 October 2023; 39 apps averaging four billion ad requests a day) and Konfety (16 July 2024; more than 250 decoy apps and a peak of 10 billion bid requests a day).6vendor assertionsupported as statedmedium
Pixalate's 'Monarch' report of 25 March 2020 alleged app spoofing and falsified measurement events across more than a dozen passive Roku apps that, on its conservative estimate, generated hundreds of thousands of invalid-traffic impressions a day for over six months, while stating that it did not assert or assign culpability.6vendor assertionsupported as statedmedium
BADBOX 2.0, with the original BADBOX that the same records describe, is the only scheme in this register with records outside its finder: the FBI's public service announcement of 5 June 2025 says the BADBOX 2.0 botnet consists of millions of infected devices, without mentioning advertising, notes that the original BADBOX was identified in 2023 and disrupted in 2024, and credits Google, HUMAN, Trend Micro and Shadowserver, while Google's complaint in Google LLC v. Does 1-25 (S.D.N.Y. 1:25-cv-04503; Document 22, filed 11 July 2025) alleges more than ten million infected devices as of April 2025 and says Google refunds advertisers for fraudulent traffic, allegations on which the court had already granted preliminary relief (order signed 27 June 2025, after a temporary restraining order of 30 May 2025).6direct recordsupported with qualificationhigh
None of the 14 CTV and in-app scheme disclosures opened (2020-2026) reports a detection miss rate, false-positive rate or precision for any product: each states scale in the finder's own units (ad or bid requests, devices, IP addresses, apps, downloads or modelled dollars) from the finder's own traffic, every finder (White Ops and HUMAN, DoubleVerify, Oracle Moat, Pixalate) sold the detection it publicised at the time, and 12 of the 14 say the finder's customers were protected or its product detected the scheme.6synthesissupported with qualificationmedium
Keep; re-open the NewsJunkie release before publication: its two quoted passages are unverified by this pass.6synthesissupported with qualificationmedium
MRC's 2021 guidance does not require TV-off detection for accredited CTV impressions, OM SDK added a display-connection signal only with version 1.4 in 2022, DoubleVerify's accredited CTV viewable impressions count certified apps only, and DoubleVerify stated on 24 April 2024, without a published method, that over one in three CTV impressions serve into environments that fire ads when the TV is off, an estimated $1 billion of annual waste.6synthesissupported as statedmedium
DoubleVerify's release of 7 May 2026 states that it detected 140% more CTV fraud schemes and variants in Q1 2026 than in Q1 2025 and more than 50 distinct CTV bot attacks and variants in 2025, and that bots took 34% and 25% of impressions in two direct-deal CTV campaigns, figures from its own measurement data without published counts, periods or campaign selection; its 5 November 2025 release with Roku cites 4 million infected CTV bot devices.6vendor assertionsupported as statedlow
Responses to CTV device spoofing remain proprietary or unmeasured: Roku's Advertising Watermark is designed to let only genuine Roku devices display authenticated ads and DoubleVerify used its signals against CycloneBot, OM SDK 1.6.0 added Privacy Pass device attestation for Fire TV in October 2025, ads.cert Authenticated Devices is still marked future, and no source opened reports how much CTV inventory carries any of these signals.6synthesissupported as statedmedium
Channel Factory dated two classification changes in autumn 2026: a Premium CTV layer (28 September 2026) adding video-level metadata tagging and genre and mood classification across a premium publisher network, and AI Slop Detection (2 October 2026), a five-tier pre-bid classification at channel, video and creative level in 50+ languages.6vendor assertionsupported as statedmedium
MRC's statement of 11 April 2025 says HUMAN's audited pre-bid product (Ad Fraud Defense) supplies IVT suggestions derived from back-end detection through an API to DSPs and SSPs, not advertisers or agencies (elsewhere: its clients are 'generally' DSPs and SSPs); that HUMAN does not directly filter requests, prevent ad serving or enable blocking, post-serve blocking included; that platforms' bid decisions on those suggestions fall outside MRC's audits; and that pre-bid databases update near real time and flat files every several minutes (a sentence covering DV's and HUMAN's assets together).6direct recordsupported with qualificationhigh
HUMAN sells viewability for display and video in desktop, mobile web and in-app only (solution page captured 21 February 2026), and its release of 27 April 2026 claims MRC viewability accreditation for display and video in desktop, mobile web and mobile app, which matches MRC's listed scope; no CTV viewability product or claim was found.6synthesissupported as statedhigh
HUMAN says a substantial share of CTV inventory is delivered without JavaScript, leaving fewer detection signals, and that its Ad Fraud Defense and Ad Fraud Sensor customers are defended against the NewsJunkie spoofing scheme (peak up to nearly two billion invalid CTV bid requests a day per seller), without stating a catch rate.6vendor assertionsupported as statedhigh
IAS's ad-fraud page (modified 16 October 2026) claims MRC SIVT accreditation in desktop, mobile web, mobile app and CTV and says IAS was the first verification provider to receive it in CTV; the environments match MRC's list, but the page omits MRC's 'certified applications only' limit, and the priority claim cannot be confirmed from the MRC records opened, while MRC's 2024 policy approves 'MRC First' claims only when unquestionably correct.6synthesissupported with qualificationmedium
MRC's accredited digital list, read on 10 November 2026 with no update stamp, shows Pixalate accredited for tracked display and video ads, rendered video impressions, SSAI video metrics and SIVT detection and filtration in desktop, mobile web, in-app and CTV, and for rendered display impressions and display viewability in desktop, mobile web and in-app; it links only the 8 April 2025 letter, whose enclosure also grants rendered display impressions in CTV.6direct recordsupported as statedhigh
Pixalate classifies traffic as CTV by the device on which its tag fires, decides per transaction whether server-side ad insertion traffic is invalid using more than 30 composite signals about the SSAI server rather than treating all SSAI as IVT, and reports 12 MRC-accredited SSAI metrics.6vendor assertionsupported as statedmedium
Pixalate's video viewability and engagement measurement is in beta: 54 metrics at three IVT tiers, a video impression counted viewable when 50% of the player's pixels are on screen for 2 continuous seconds during playback, signals from the Open Measurement SDK through a VAST verification script, and no historical backfill.6vendor assertionsupported as statedhigh
DoubleVerify measured about 9.5 trillion Media Transactions Measured in 2025 (8.3 trillion in 2024, 7.0 trillion in 2023, 5.5 trillion in 2022, 4.5 trillion in 2021), the count of media transactions, or ads, measured for advertising customers on which it bills; it estimates 44% display and 56% video, and, within post-campaign measurement in 2025, 74% mobile, 14% desktop and 12% CTV.6, 7direct recordsupported as statedhigh
MRC's letter of 13 April 2026 accredits DoubleVerify's Pinnacle platform on CTV for display and video impressions inclusive of SIVT, for video viewable impressions and Fully On-Screen metrics from certified apps only (other viewability metrics may include tested, non-certified apps), for property-level ad verification in ten CTV languages (Chinese, English, French, German, Hindi, Japanese, Korean, Malay, Spanish, Vietnamese; French added in April 2026) and for pre-bid Fraud/IVT, Fully On-Screen, app-store-category and app-list segments, while its pre-bid brand safety, suitability and IAB contextual segments are accredited on desktop web, mobile web and in-app only.6, 12direct recordsupported as statedhigh
MRC's letter of 9 July 2026 accredits IAS on CTV for video tracked ads, impressions, viewable impressions, viewability metrics and SIVT filtration for certified applications only (the enclosure itemises CTV impression and viewability metrics and SSAI Noncompliant Ads, but no CTV tracked-ads or Total Net Impressions row), while its GIVT and SIVT count and rate metrics are listed for desktop web, mobile web and in-app only; MRC granted IAS's CTV SIVT filtration and reporting in April 2024.6, 12direct recordsupported with qualificationhigh
The only CTV content-level brand safety and suitability accreditation on MRC's list on 10 November 2026 is Zefr's labelling and reporting of English-language YouTube in-stream video and music video (granted 20 January 2026, excluding Shorts, YouTube TV, Google TV, Masthead, Google Video Partner videos and volumetric data); Google's own YouTube content-level row lists no CTV, IAS's YouTube brand-safety audit covering CTV is listed as 'in process as of 04/2023', and GumGum's open-web content-level accreditation covering CTV, continued in May 2025, was revoked in September 2025 on its withdrawal.6, 12direct recordsupported as statedhigh
Vendors sell or describe CTV content signals at program, genre or show level without MRC content-level accreditation: on 6 November 2025 DoubleVerify launched Verified Streaming TV (verification of premium streaming-player environments) and 'Do Not Air' automation inside Authentic Brand Suitability, which curates streaming inventory by DV-normalised genre and maturity-rating data (early access through The Trade Desk), and said licensed IMDb data would power capabilities planned for early 2026; Pixalate markets OpenEPG, which resolves the show in a CTV bid request and returns title, genre, rating and channel; IAS sells CTV Brand Safety for YouTube while its YouTube audit is listed as in process; and Peer39 states that roughly 60% of CTV bid requests lack usable program-level signal.6, 12synthesissupported with qualificationmedium
MRC's Server-Side Ad Insertion and OTT Guidance (August 2021, listed by MRC as the OTT/CTV and SSAI Digital Video Measurement Guidelines) keeps CTV, defined by delivery of digital video to televisions, as a discretely accredited environment while OTT is not discretely accredited; it does not currently require TV-off detection for CTV video impressions, says events confirmed as collected while the TV was off should be removed and not counted, asks vendors to disclose TV-off as a general limitation, and requires auditable support from vendors that forgo available TV-off signals.6, 12direct recordsupported with qualificationhigh
DoubleVerify's filed revenue by customer type shows Supply-side revenue of $43.3 million (FY2022), $45.6 million (FY2023), $56.8 million (FY2024) and $71.3 million (FY2025) against total revenue of $452.4 million, $572.5 million, $656.8 million and $748.3 million, that is 9.6%, 8.0%, 8.6% and 9.5% of worldwide revenue in calendar years 2022 to 2025.7direct recordsupported as statedhigh
In the nine months to 30 September 2026 DoubleVerify booked Supply-side revenue of $56.8 million, 9.8% of total revenue of $580.3 million (Activation $325.8 million, Measurement $197.7 million), up from $51.7 million of $542.7 million in the same period of 2025.7direct recordsupported as statedhigh
IAS's last filed revenue is for the nine months to 30 September 2025: Publisher revenue of $68.7 million of $437.6 million total (15.7%), up from $53.1 million of $377.1 million a year earlier, with Optimization at $206.4 million and Measurement at $162.6 million.7direct recordsupported as statedhigh
DoubleVerify charges advertisers a Measured Transaction Fee per thousand impressions on the volume of Media Transactions Measured, with most prices fixed and caps on platforms that charge a percent of media spend; the average fee fell 4% in FY2024 and 3% in FY2025 (each excluding an introductory fixed-fee deal with one large customer) and 4% in the nine months to September 2026, while measured volume rose 19%, 15% and 11%.7direct recordsupported as statedhigh
DoubleVerify reports Media Transactions Measured, its count of media transactions or ads measured for advertiser customers, of about 4.5 trillion (2021), 5.5 trillion (2022), 7.0 trillion (2023), 8.3 trillion (2024) and 9.5 trillion (2025), and estimates that in 2025 44% were display and 56% video, and that within post-campaign measurement 74% were mobile, 14% desktop and 12% CTV.7direct recordsupported as statedhigh
IAS's FY2024 10-K gives no annual count of impressions measured: it reports processing over 280 billion daily digital interactions (as of 31 December 2024), a different unit, and growth rates instead (2024: measurement impressions up 23% with average CPMs down 7%; optimization impressions up 4% with CPMs up 2%), and it recognizes revenue as CPM times impressions measured.7direct recordsupported as statedhigh
No DoubleVerify customer exceeded 10% of revenue in 2023-2025 and no IAS customer exceeded 10% of accounts receivable at the end of 2023 or 2024 (IAS's 10-K states no single-customer revenue share), but both route much of their revenue through a few programmatic platforms: two programmatic partner platforms collected about 18% and 17% (2023), 22% and 14% (2024) and 24% and 12% (2025) of DoubleVerify's total revenue on advertisers' behalf; DSP partnerships facilitated 47%, 47% and 46% of IAS's revenue in 2022-2024, the largest one 22%, 21% and 18%.7direct recordsupported with qualificationhigh
The HSR waiting period expired at 11:59 p.m. ET on 21 September 2026; Germany (5 October), Cyprus (7 October), the ACCC and Morocco (13 October 2026) cleared the merger, which could not complete until 14 calendar days after the ACCC published its reasons; DoubleVerify's stockholders adopted the merger agreement on 29 October 2026 by 118,218,914 votes for, 7,592,263 against and 214,454 abstentions.7direct recordsupported as statedhigh
Counting 14 calendar days from the ACCC's publication of its determination on 13 October 2026, the window for applications to the Australian Competition Tribunal closed at the end of 27 October 2026, and the Tribunal's current-matters list on 10 November 2026 showed four 2026 matters, none concerning Nielsen, Neptune BidCo or DoubleVerify.7synthesissupported as statedmedium
DoubleVerify's expected closing moved from the end of the fourth quarter of 2026 (joint release, 6 August 2026) to the first quarter of 2027 (proxy, 29 September, and results release, 30 October 2026) and back to the fourth quarter of 2026 (8-K of 9 November 2026, 'in a position to close'); all four are forecasts.7forecastsupported as statedhigh
DoubleVerify's other documents screened contain no conflict policy for its seller-side revenue: its Code of Ethics defines conflicts as personal benefit, the FY2025 10-K's only conflicts policy concerns Providence's corporate opportunities, and its Impact Report 2023, whose reporting period is calendar 2022, says it has 'no inherent bias or conflict of interest' because it is not party to the media transaction and operates solely in advertisers' interest, while 9.6% of its 2022 revenue ($43.3 million of $452.4 million) came from supply-side customers (8.0% in 2023).7synthesissupported with qualificationmedium
On 24 September 2025 Integral Ad Science agreed to merge with a subsidiary of Igloo Group Parent, Inc., an affiliate of funds managed by Novacap, for $10.30 a share in cash; holders of about 52.6% of the voting power consented in writing the same day; Novacap funds committed up to $857.8 million of equity and lenders a $1.0 billion term loan and a $150 million revolver; termination fees were $52.5 million (IAS) and $100 million (Parent).7direct recordsupported as statedhigh
IAS's take-private closed on 23 December 2025; the closing 8-K puts the aggregate price paid for the outstanding shares (excluding the Owned Company Shares and Appraisal Shares) at about $1.6 billion, funded by Novacap equity, a credit agreement and IAS's cash, while the joint release of 24 September 2025 says the deal 'values IAS at approximately $1.9 billion' and states no basis.7direct recordsupported as statedhigh
After the closing, Nasdaq filed Form 25 on 23 December 2025 and IAS filed Form 15 on 2 January 2026 under Rules 12g-4(a)(1) and 12h-3(b)(1)(i), certifying one holder of record; EDGAR shows no IAS 10-K or 10-Q after its 10-Q for the quarter to 30 September 2025, so IAS's FY2025 revenue by channel is not public.7direct recordsupported as statedhigh
Since 2021 DoubleVerify completed four acquisitions, only one of which (Meetrics) its filings describe as an ad verification provider: Meetrics (31 August 2021, $24.3 million net cash), OpenSlate (22 November 2021, $148.2 million net of cash; a pre-campaign contextual targeting platform that, by DoubleVerify's own announcement, helped brands align ads 'with suitable or contextually relevant content across social video and CTV'), Scibids (14 August 2023, $121.4 million including $52.9 million of stock, AI campaign optimization) and Rockerbox (13 March 2025, $82.3 million net of cash, marketing attribution).7direct recordsupported with qualificationhigh
IAS listed through an IPO priced on 29 June 2021 at $18.00 a share and closed on 2 July 2021, three years after Vista Equity Partners acquired it (2018); in 2021 it acquired Publica (9 August; $171.4 million cash plus 2,888,889 shares valued at $49.6 million), Context (Nobora SAS, 31 December; $22.6 million cash plus shares valued at $10.4 million) and Amino Payments (date and price not stated in the pages read).7direct recordsupported as statedhigh
Protected Media's ownership runs from Flashtalking, which announced in February 2021 that it had acquired Protected Media (no terms stated), to Mediaocean, which agreed on 13 July 2021 to acquire Flashtalking (terms not disclosed; closing expected in the third quarter of 2021) and called Flashtalking part of Mediaocean by 12 May 2022; Innovid's proxy of 10 January 2025 describes Mediaocean as a portfolio company of funds managed by CVC Capital Partners and of TA Associates and Charlesbank Capital Partners, offering 'Protected by Mediaocean's ad verification solutions'.7synthesissupported as statedmedium
Mediaocean, Protected Media's owner, agreed on 21 November 2024 to acquire Innovid for $3.15 a share in cash and completed the merger on 13 February 2025, combining Innovid with Flashtalking; funds affiliated with CVC Capital Partners and Charlesbank Capital Partners committed the equity ($47 million per the proxy), and neither filing states an aggregate deal value.7direct recordsupported as statedhigh
Mediaocean owns both ad servers (Flashtalking and, since February 2025, Innovid) and Protected Media's verification, which it announced on 12 May 2022 as fully integrated into Flashtalking's ad serving; its statements frame independence as freedom from media ownership, and no policy separating verification from ad serving was found in the pages and filings opened.7vendor assertionsupported as statedmedium
The IVT specialists document seller-side customers without filing revenue: HUMAN announced on 5 October 2026 an Ad Integrity Suite for Platforms for DSPs, SSPs and exchanges, with OpenX the first platform to adopt it, and its 2022 site offered MediaGuard to DSPs, SSPs, media owners and brands; Pixalate names SSPs and publishers among its customers and says its Seller Trust Index rankings use only buy-side data, a method safeguard rather than a conflict policy; no conflict or separation policy was found for either.7vendor assertionsupported with qualificationmedium
Few funding or ownership events were found for the other specialists since 2021: Pixalate announced $18.1 million of growth capital on 2 September 2021 (Western Technology Investment and Javelin Venture Partners; $22.7 million in total), Zefr acquired Adverif.ai on 20 July 2022 (terms not disclosed), and Truelink Capital completed a 'significant investment' in Channel Factory on 24 April 2025 (terms not disclosed), through which Star Mountain exited while remaining a minority owner; none states who controls the company.7vendor assertionsupported as statedmedium
Fraudlogix (Hallandale Beach, Florida; founded 2010 and still founder-led) sells its programmatic IVT detection 'exclusively' to high-volume RTB platforms as free post-bid pixel analytics of served impressions (GIVT and SIVT, brand-safety flags, domain verification, viewability) plus a fee-based pre-bid IP blocklist, which fits cohort B's trading-platform route; it is on neither of MRC's digital lists.7vendor assertionsupported as statedmedium
Kubient, Inc. sold KAI, a pre-bid ad-fraud detection product deployable on third-party real-time bidding platforms (FY2022 10-K, filed 30 March 2023); its 8-K filed 25 September 2023 reported that its board had concluded that the financial statements for fiscal 2020 to 2022 and the quarterly reports through the quarter ended 31 March 2023 should no longer be relied upon; a Form 25 removing its securities from listing was filed on 21 November 2023, and an 8-K of 8 May 2024 reported a workforce reduction in connection with the potential wind-down of its business, so KAI is a historical cohort B row.7direct recordsupported with qualificationhigh
Since Peer39 bought Adloox (blog of 3 June 2026), Peer39 sells pre-bid brand safety, suitability, IVT and viewability controls with post-buy reporting for which it does not charge separate fees (heading 'No Verification Fees.'), while MRC's continuing accreditation of 29 June 2026, addressed to Peer39 P39 France, covers the Adloox service's tracked ads, impressions, decision rate, viewability and exposure metrics on desktop, mobile web and in-app, with SIVT filtration applied on desktop and mobile web (the enclosure also lists the decision rate net of SIVT for mobile app) and no pre-bid metric.7synthesissupported with qualificationhigh
Mantis, which Reach plc developed in 2019, is sold mainly to publishers (charged per URL scanned, with brand-safety key-values passed to their ad servers for PMP and PG deals, and Mantis Video added on 2 June 2026); on its advertiser page of 10 November 2026 the only post-serve monitoring product is the Mantis Sentinel tag, which captures every URL an ad serves on and reports brand-safety insight, while the same page also offers buyers contextual targeting and Mantis Signals (contextual, brand-safety and product signals for planning, activation and measurement), whose buying task and environments are not documented.7vendor assertionsupported with qualificationmedium
Two cohort C seeds are confirmed on 10 November 2026: illuma Sentinel, a pre-bid page-level brand-safety layer with reporting for advertisers, sold in The Trade Desk's brand-safety marketplace since 27 July 2026 by a London and New York company, and Proximic by Comscore's pre-bid brand safety and suitability filters across desktop, mobile, CTV and podcasts, for which Comscore claims no MRC accreditation (its page lists MRC accreditation only for Comscore TV).7vendor assertionsupported as statedmedium
Four acquired brands from verification, suitability and fraud detection now sit inside other companies: DoubleVerify acquired Meetrics on 31 August 2021 and OpenSlate on 22 November 2021 (on 10 November 2026 openslate.com redirected to doubleverify.com and meetrics.com did not resolve), Zefr acquired Adverif.ai on 20 July 2022, and Impact Radius acquired Forensiq on 14 June 2016.7direct recordsupported as statedhigh
Scope3 announced on 4 November 2024 that it had acquired Adloox (Adloox's post of 5 November 2024 and Scope3's release dated 5 November 2024 give no terms or separate closing date); the registers record the change of control: Adloox Limited's directors changed on 4 November 2024, the BODACC notice published on 31 May 2025 records Scope3 PBC as president of the French company, then a single-shareholder SAS, and Adloox Limited's confirmation statement of 23 September 2025 names SCOPE3 FRANCE SAS as its shareholder.7direct recordsupported as statedhigh
No filing found settles who owns Peer39: its terms and privacy policy name P39 Tech, LLC, a Delaware company located in New York; it describes itself since 2019 as 'Wholly independent, employee owned'; and the only ownership record found for a Peer39 entity, the UK register for P39 Technologies UK Limited (incorporated 4 September 2019, now president of P39 France), shows one individual (named in the register, a US national) holding 75% or more of its shares and voting rights since incorporation and serving as its sole director.7synthesissupported as statedmedium
Adloox sells to publishers as well as buyers: its publisher page offers ad-server-activated brand-safety and fraud exclusion segments and a native Google Ad Manager integration, its January 2023 Equativ integration lets publishers measure their own traffic, a publisher (Brut) chose Adloox for its programmatic operations in September 2023, and its data-use terms say the Solution serves 'online media sellers and buyers'; no revenue split is published, and no conflict or separation policy was found in the Adloox and Peer39 pages or the filings screened on 10 November 2026.7vendor assertionsupported as statedhigh
Peer39, the owner, sells pre-bid controls and says its post-buy reporting exists to confirm 'that the pre-bid controls worked', with no separate verification fee, and it also sells publishers categorisation of their own inventory; the owner thus reports on the effect of controls it sells and has publisher customers, and no separation policy covering these interests was found in the pages screened on 10 November 2026.7inferencesupported as statedmedium
Truelink Capital, a Los Angeles private equity firm, completed a 'significant investment' in Channel Factory on 24 April 2025 with terms undisclosed, and Star Mountain Capital exited through that transaction while remaining a minority owner; no source opened states Truelink's stake or who controls Channel Factory.7vendor assertionsupported as statedmedium
Channel Factory discloses no revenue by customer type and no conflict or separation policy, while its own documents describe relationships with the content side it classifies: influencer partnerships, creator branded content and sponsorships, and YouTube channel owners whose authenticated data is shared with 'Channel Factory and its Media Partners' (read 10 November 2026).7synthesissupported as statedmedium
Inference: Channel Factory is neither the seller nor the platform, so its classification is third-party under the paper's definition, but the same company builds the suitability lists and is paid to buy media against them, and it publishes no separation between list-building and media trading.7inferencesupported as statedmedium
On 6 August 2026 DoubleVerify agreed that Wallace Merger Sub Inc. would merge into it, making it a wholly owned subsidiary of Neptune BidCo US Inc., which the definitive proxy describes as the parent of the Nielsen Company (US), LLC and an affiliate of funds managed by Elliott and Brookfield, for $13.60 a share in cash (an enterprise value of about $2.15 billion by the parties' joint release), with termination fees of $60 million payable by DoubleVerify and $144 million by the parent and an initial termination date of 6 May 2027.7direct recordsupported as statedhigh
DoubleVerify completed its IPO on 23 April 2021 at $27.00 a share and has since completed four acquisitions, of which only Meetrics is described in its filings as an ad-verification provider: Meetrics (31 August 2021, $24.3 million), OpenSlate (22 November 2021, $147.4 million including $22.5 million of stock; described as a pre-campaign contextual targeting platform for social video and CTV), Scibids (14 August 2023, $121.4 million, AI campaign optimisation) and Rockerbox (13 March 2025, marketing attribution; $82.6 million of 2025 investing cash flows attributed to it).7direct recordsupported with qualificationhigh
DoubleVerify's filed revenue by customer type shows seller-side revenue: Supply-side revenue was $71.3 million of $748.3 million in FY2025 (9.5%) and $56.8 million of $656.8 million in FY2024 (8.6%), the rest coming from advertiser customers through Activation ($427.3 million in FY2025) and Measurement ($249.7 million); in the nine months to 30 September 2026 Supply-side was $56.8 million of $580.3 million (9.8%).7direct recordsupported as statedhigh
DoubleVerify bills advertisers a Measured Transaction Fee per thousand impressions measured, mostly at fixed prices with caps where platforms charge a percent of media; supply-side customers pay under contracts with minimum guarantees; and its Legal Center publishes partner-pay CPM list prices (effective 21 May 2026) of $0.135 for DV Authentic Ad, $0.150 (display) and $0.370 (video) for DV Authentic Ad Plus for Open Web, $0.400 for DV Social Pre-Bid Controls, $0.110 for DV Video Complete and $0.050 for DV Authentic Attention for Open Web.7direct recordsupported as statedhigh
DoubleVerify publishes no conflict or separation policy covering its seller-side revenue or its prospective owner: its June 2024 Transparency Center post describes structural separation of assessment teams from business development without calling it a policy; its 2023 Impact Report says it has no inherent bias or conflict of interest because it is not party to the media transaction; its 10-K's conflicts policy concerns Providence's corporate opportunities; and the merger documents offer only the joint release's statement that both operating structures will stay independent, the definitive proxy containing no 'conflict of interest' text.7synthesissupported as statedhigh
On 5 October 2026 HUMAN announced Ad Integrity Suite for Platforms for DSPs, SSPs, exchanges and other programmatic platforms, combining IVT, viewability and brand-safety signals with reason codes; OpenX is the first adopter, using the signals in supply curation and deal creation, and further capabilities are announced for H1 2027.7vendor assertionsupported as statedhigh
In a Rule 7.1 corporate disclosure statement docketed on 26 June 2026 in Chavez v. Human Security (S.D.N.Y. 1:26-cv-03727), HUMAN identified Singularity Buyer LLC and Singularity TopCo LLC as its corporate parents; Singularity TopCo LLC (Delaware LLC, incorporated 2020, at 841 Broadway, New York, HUMAN's own address in its January 2025 Data Use Agreement) filed Form Ds in 2024 naming HUMAN's CEO Stuart Solomon as an executive officer and director; EDGAR lists no registrant named HUMAN Security or Singularity Buyer.7direct recordsupported as statedhigh
HUMAN's releases record a $100 million growth round led by WestCap (4 January 2022); on 27 July 2022 HUMAN and PerimeterX announced they were uniting in a merger under the HUMAN name (board and regulatory approvals received; completion not separately stated; terms not disclosed), with a $100 million Blackstone Credit debt facility and a statement that the combined company will have more than $100 million in ARR; and $50 million-plus led by WestCap with Goldman Sachs, ClearSky, NightDragon and Vertex Ventures US (9 October 2024); HUMAN files no revenue, and its parent's Form Ds decline to disclose a revenue range.7vendor assertionsupported with qualificationmedium
HUMAN's Magnite case study (PDF created 1 December 2025) says Magnite has used Ad Fraud Defense since 2015 and is a founding member of the HUMAN Collective; with its platform suite and publisher products HUMAN discloses sell-side and platform-paid relationships, but no revenue by customer type and, in the pages and terms read, no conflict or separation policy.7synthesissupported as statedmedium
Integral Ad Science's take-private closed on 23 December 2025, when Merger Sub merged into the company and IAS became a wholly owned subsidiary of Igloo Group Parent, Inc., an affiliate of funds managed by Novacap Management Inc.; the closing 8-K puts the aggregate price paid for outstanding shares at about $1.6 billion.7direct recordsupported as statedhigh
In FY2024, IAS's last filed fiscal year, revenue of $530.1 million split into Optimization $242.6 million, Measurement $211.0 million and Publisher $76.5 million (14.4%), with 86% of revenue from advertiser customers; the Publisher channel includes Publica's CTV ad-serving products.7direct recordsupported as statedhigh
DSP partnerships facilitated 46%, 47% and 47% of IAS's revenue in 2024, 2023 and 2022, the largest DSP 18%, 21% and 22%, and IAS pays DSPs including Google and The Trade Desk a percentage of revenue received through its products under long-term revenue-share agreements.7direct recordsupported as statedhigh
IAS's owner also owns a digital media company and a TV advertising platform: Novacap's portfolio pages list IAS (invested 2025), Static Media, a digital media company operating nineteen brands with more than 100 million monthly visitors (invested 2023), and Cadent, a TV advertising platform (invested 2023); no conflict-of-interest or separation policy covering IAS's supply-side lines or its owner's holdings was found on integralads.com or in the filings read.7synthesissupported as statedmedium
The FY2024 10-K says IAS generates no material revenue directly from its integration partners (the platforms), and that IAS charges a CPM on the volume of ads it analyzes for advertiser and publisher customers.7direct recordsupported as statedhigh
Pixalate discloses sell-side customers and products (SSPs and publishers among its customers, with platform customers named in partner releases such as Screencore, Gravite and Pangle; publisher Analytics, a publisher Tag Server, clawback evidence for SSPs, seller and publisher indexes) but no revenue split by customer type, and its published safeguards are narrow: the MST bars use of a client's first-party data to rank that client on its indexes, the Seller Trust Index ranks on buy-side data only 'to keep them impartial', with client data contributing only in aggregate, and Clawback Warning labels cannot be added or removed manually; no general conflict-of-interest or separation policy was found.7synthesissupported with qualificationmedium
In its Rule 7.1 corporate disclosure statement docketed on 30 April 2025 in Pixalate, Inc. v. Criteo S.A., Pixalate stated that it has no corporate parent.7direct recordsupported as statedmedium
Pixalate announced on 2 September 2021 $18.1 million of additional growth capital from Western Technology Investment, with Javelin Venture Partners strengthening its investment, bringing total investment to $22.7 million; no later financing or ownership event was found in the sources opened.7vendor assertionsupported as statedmedium
Zefr is a private Delaware corporation (EDGAR registrant ZEFR, INC., formerly MOVIECLIPS, INC., which filed under that name from November 2009 to August 2012) whose last SEC filing is a Form D of 8 March 2016; its Form D notices record $18.5 million of equity sold in August 2012 (as Movieclips), $30.0 million in February 2014, $0.9 million in September 2014 and $5.0 million of debt with options or warrants in March 2016, each declining to disclose revenue and naming directors and officers but not investors.7direct recordsupported with qualificationhigh
Zefr's interests are disclosed only in its own releases and reposts: it describes itself as independent ('We are not a media owner'; 'No hidden incentives'), yet it runs Zefr Media Services buying YouTube media for advertisers, supplied video-level signals to TikTok's first-party Inventory Filter and flags unsuitable content to TikTok (September 2023), and had a revenue-sharing agreement with IRIS.TV, a sell-side contextual marketplace (Adweek, November 2020, reposted); no revenue by customer type and no conflict or separation policy were found.7synthesissupported as statedmedium
Zefr acquired Adverif.ai, an Israeli AI company focused on identifying misinformation, on 20 July 2022 (terms not disclosed); its team joined Zefr's technology organisation, and Zefr has since sold a misinformation category on TikTok (February 2023) and YouTube (September 2024).7vendor assertionsupported as statedmedium
mFilterIt, which lists a US entity (MFILTERIT TECHNOLOGIES INC, Northborough, Massachusetts) beside offices in India, the UAE and Indonesia, documents on 10 November 2026 real-time GIVT and SIVT detection with pre-bid avoidance and post-bid measurement, viewability measurement across display, video and CTV, and a brand safety and suitability solution with post-bid placement-level verification and GARM-based categories, which meets cohort A's entry test on vendor documentation alone; it is on neither of MRC's digital lists.7, 8vendor assertionsupported as statedmedium
As of 20 October 2026 DoubleVerify had received fourteen stockholder demand letters and two stockholder complaints over the Neptune BidCo merger, none served, alleging proxy disclosure deficiencies; its Q3 2026 10-Q of 30 October 2026 repeats 'two complaints' without naming a court or caption, and the only caption found is Michael Kent v. DoubleVerify Holdings, Inc. et al., filed in New York County Supreme Court on 8 October 2026 against DoubleVerify and eight individuals and visible only through an aggregator.7, 9direct recordsupported as statedmedium
The only regulator decision on a scored vendor found in the competition registers screened is the ACCC's Phase 1 approval of Neptune BidCo's acquisition of DoubleVerify (MN-55036, determination published 13 October 2026), one of the merger reviews DoubleVerify reported complete on 20 October 2026; on 10 November 2026 the CMA's case finder returned no case for DoubleVerify, IAS, HUMAN Security, Pixalate, Zefr, Peer39, Adloox, Protected Media, Channel Factory, Scope3 or 'brand safety', and the European Commission's case search returned none for DoubleVerify, Pixalate or Zefr.7, 9direct recordsupported as statedmedium
DoubleVerify's pending acquirer is the parent of a measurement company: the proxy describes Neptune BidCo US Inc. as the parent of the Nielsen Company (US), LLC and an affiliate of funds managed by Elliott and Brookfield, formed on 17 March 2022 to acquire Nielsen Holdings plc; the joint release says 'The combined company will maintain the independence of both operating structures', but neither it nor the proxy pages read sets out a conflict or separation policy.7, 12direct recordsupported as statedmedium
On 13 June 2024 DoubleVerify published a page stating that its media quality assessment teams operate independently from business development and client relations teams ('structural separation'), that it does not soften assessments for publisher clients, and that publishers can raise site-classification concerns directly; the page does not call itself a policy and names no scope, owner or enforcement for the separation, and its only audit statement is general ('Our processes and outcomes are regularly audited by respected external bodies like the Interactive Advertising Bureau'), not tied to the separation.7, 12vendor assertionsupported with qualificationmedium
IAS sells to the supply side: publisher tools, Publica CTV ad serving (with an exclusive global partnership with Samsung Ads extended on 30 September 2025), Quality Connect (June 2026) giving publishers visibility into advertisers' opted-in settings, resale of pre-bid segments by DSPs and other sell-side customers, and 'Media Partner Fees' under which a media partner can pay the verification fees for an advertiser's campaign.7, 12vendor assertionsupported as statedhigh
Adloox changed owners twice since 2024: Scope3 announced its acquisition on 4 November 2024 (no terms or separate closing date), and Peer39, which describes itself as 'Wholly independent, employee owned' since 2019 and files nothing, announced on 2 June 2026 that it had acquired Adloox from Scope3 (terms not disclosed); MRC's letter of 29 June 2026 continues accreditation of the 'Peer39 Adloox (formerly d/b/a Scope3 France Adloox)' service; Scope3's announcement of 23 September 2026 that it 'is becoming Apostra' renames the former owner and does not change Adloox's.7, 13synthesissupported as statedmedium
At 11:40 UTC on 10 November 2026 DoubleVerify Holdings, Inc. was still a NYSE-listed company not owned by Nielsen's parent: its newest EDGAR filing was the 8-K (Item 7.01) of 9 November 2026, with no 8-K Item 2.01, Form 25 or Form 15, and that 8-K said the company currently expects the parties to be in a position to close the merger with a subsidiary of Neptune BidCo US Inc. during the fourth quarter of 2026, a forecast.7, 13direct recordsupported as statedhigh
White Ops announced on 23 December 2020 that Goldman Sachs Merchant Banking Division, with ClearSky Security and NightDragon, had acquired it from Paladin Capital Group, Grotech Ventures and other shareholders (closed; terms not disclosed), and renamed itself HUMAN on 30 March 2021.7, 13vendor assertionsupported as statedhigh
TAG's Brand Safety Certified Guidelines v5.0 reproduce the twelve floor categories and IAB Tech Lab's eleven Sensitive Topics at the Floor risk-rating without tiers or GARM's name, BSI hosts the floor-and-tiers framework without a date, version or owner, and the 4A's APB framework dates from 21 September 2018 (13 categories), its link on the 4A's page now resolves to an unrelated members-only page, and the only dated APB-related output found since June 2022 is the Media Responsibility Principles updated by the 4A's Media, Tech and Data practice in collaboration with the APB (published 11 December 2023, page modified 5 January 2026), eleven principles with no floor categories or tiers; the APB's other listed outputs (Call to Action, April 2020; Brand Safety Playbook, 2020) predate June 2022.8direct recordsupported with qualificationhigh
IAB's last brand safety and suitability guide on iab.com is dated December 2020; IAB Tech Lab's Content Taxonomy 3.1 (posted for public comment on 9 December 2024, comment closed 24 January 2025, file last revised 22 July 2025) keeps the eleven GARM-derived Sensitive Topics categories but carries no tiers, the only tier labels (Floor, High, Medium and Low Risk) sitting without definitions in the June 2022 3.0 Descriptive Vectors; the ANA's benchmark release of 5 November 2025 introduced suitability metrics without publishing a framework; and no publication on EASA's five publication pages is a brand safety or suitability framework, though its December 2023 repository of responsible-advertising initiatives describes GARM and the IAB UK Gold Standard's brand-safety aim.8direct recordsupported with qualificationmedium
JICWEBS, a body controlled by the UK trade bodies AOP, IAB UK, IPA and ISBA, ran UK digital trading standards: its Digital Trading Standards Group (set up 2012) set Good Practice Principles under which buyers and sellers of digital display and audio advertising must trade under a Primary Agreement or Contract that should state where advertising should or should not appear, using independently certified content verification tools or schedules (lists); signatories were verified by JICWEBS-approved providers (ABC, BPA Worldwide and ePrivacy for DTSG brand safety), with ABC the approved provider for content verification, viewability and anti-fraud product certification, before TAG's Certified Against Fraud replaced JICWEBS's anti-fraud principles from January 2019.8direct recordsupported with qualificationhigh
JICWEBS passed to TAG in stages: a formal partnership in March 2017, a merger agreement announced on 29 July 2020 for completion in early September, TAG recorded at Companies House as JICWEBS Ltd's controller (75% or more of voting rights) from 1 September 2020, when all four founding trade-body controllers (AOP, ISBA, IPA and IAB UK) ceased, and TAG Brand Safety Certified launched on 10 September 2020, its v1.0 merging the DTSG programme with TAG's Inventory Quality Guidelines; JICWEBS Ltd was still 'Active' at Companies House on 10 November 2026, with its confirmation statement shown as overdue.8direct recordsupported with qualificationhigh
GroupM's Digital Risk Director stated on JICWEBS's site (archived December 2019) that GroupM was a JICWEBS signatory certified for DTSG Brand Safety and expected its trading partners to follow suit, and at TAG Brand Safety Certified's launch on 10 September 2020 the 112 certified companies included GroupM, Havas Media, IPG Mediabrands, Omnicom Media Group and Publicis Media.8vendor assertionsupported as statedmedium
The UK Content Verification Guide of 21 January 2020, issued by AOP, IAB UK, IPA, ISBA, JICWEBS and Newsworks, tells buyers to ask verification providers whether blocked impressions are paid for and whether their numbers are accredited, and its DoubleVerify case study says that in DSP buying the media cost is paid once the bid is won, so an ad blocked afterwards is wasted spend.8vendor assertionsupported as statedmedium
Buyer trade bodies sit inside the governance of the bodies that grade verifiers: the ANA says it, the 4A's and the IAB created TAG; ISBA says it sits on MRC's Board of Directors and TAG's Leadership Council; and the 2020 merger announcement said AOP, IAB UK, IPA and ISBA would join TAG's Leadership Council.8synthesissupported as statedhigh
AgenticAdvertising.org's policy registry serves Common Sense Brand Standards (CSBS) version 1.0.0, effective 1 January 2026, as a content-adjacency standard that originated at Scope3, with an exclusion floor and 'graduated assessment' categories but no defined suitability tiers; its revision history returned zero revisions on 10 November 2026, and AdCP's documentation says the IP donation had not yet been formalised.8direct recordsupported as statedhigh
Apostra (whose rename from Scope3 was announced on 23 September 2026; 'Scope3 by Apostra' remains its sustainability brand) does not sell Brand Standards on any current page or document opened on 10 November 2026: the product page reads 'Pending page design and copy', the home page does not mention it, and the developer corpus (llms-full.txt, about 9.4 MB of text) contains no occurrence of 'brand standards', while Scope3's legacy docs still describe it with an update stamp of 28 August 2025.8synthesissupported with qualificationmedium
Google's Display & Video 360 help (read 10 November 2026) still lists 'Scope3 (formerly Adloox)' among third-party pre-bid verification providers, with brand safety and suitability content filtration 'based on the GARM framework' and fraud and invalid-traffic filtration across URL, app, domain and subdomain levels on open-web display and video, plus viewability targeting of the top share of DV360 inventory by IAB viewed rate (no level stated); on CTV only the content filtration, fraud and IVT filtration and content-category settings are available; prices are set by the vendor, and the contact is a scope3.com address.8vendor assertionsupported with qualificationhigh
The suitability framework is named only in general terms: Google's DV360 help says the Adloox categories are 'based on the GARM framework', Adloox's DV360 page lists 12 brand-suitability exclusion segments and gives three example content types whose wording follows GARM floor categories, including misinformation, its Meta release cites 'the IAB brand safety guidelines and industry-standard BS&S framework', and Peer39 calls its own basis 'the industry standard brand safety and suitability framework, formerly known as GARM'; no edition, date or maintainer is named, so alignment with the archived June 2022 GARM edition (twelve categories) is an inference from the segment count and the examples.8synthesissupported with qualificationmedium
Google's Enterprise Marketing Portal lists Channel Factory as one of four YouTube Activation Partners (with MiQ Digital, Pixability and Zefr), a campaign-support programme Google keeps separate from its Advertising Technology Partners directory, whose Brand Safety category lists only DoubleVerify, Integral Ad Science and Zefr (read 10 November 2026).8direct recordsupported with qualificationhigh
Channel Factory names no framework or dated edition for ViewIQ's categories on its current pages: its releases cite 'IAB standard category lists', its About page invites buyers to events on 'GARM+ strategies' without defining the term, and the 13+ categories are not listed (read 10 November 2026).8synthesissupported as statedmedium
On 15 April 2026 the FTC and eight states sued Dentsu, WPP Media and Publicis in the Northern District of Texas, alleging the agencies used the 4A's Advertiser Protection Bureau and the WFA's GARM to set a common 'Brand Safety Floor' targeting 'misinformation', and settled the same day by consent orders; the Dentsu order, read in full, bars Dentsu for ten years from relying on exclusion or inclusion lists or ratings based on 'Covered Bases' (political or ideological viewpoints, third-party journalistic standards, DEI) unless an advertiser expressly directs it; no verification vendor is a defendant.8, 9direct recordsupported as statedhigh
The House Judiciary Committee's interim staff report of 10 July 2024 names DoubleVerify, IAS, Oracle Moat, Peer39, Channel Factory, Pixability and Zefr among eleven ad-tech providers invited to join GARM's community and says committee documents suggest GARM was incorporating its framework into their products, citing Zefr's GARM-mapped products; it is a staff characterization, not a finding by a court or agency.8, 9direct recordsupported as statedhigh
X Corp. v. WFA (N.D. Tex. 7:24-cv-00114, filed 6 August 2024) was dismissed on 26 March 2026, with prejudice as to the WFA and the domestic defendants for want of antitrust injury and without prejudice as to four foreign defendants; on appeal (5th Cir. 26-10394) X and the WFA stipulated on 5 August 2026 to a final settlement between them while X kept its appeal against the others, and on 12 August 2026 the United States (supporting X but expressing no view on ultimate liability) and West Virginia with 11 other states (supporting reversal) filed amicus briefs.8, 9direct recordsupported as statedhigh
The WFA discontinued GARM on 9 August 2024 (its framework page says 8 August) and stated on 29 July 2026 that it "will not form or restart GARM or a similar initiative"; the GARM framework page still reads "Last updated: June 2022".8, 9, 12direct recordsupported as statedhigh
ISBA's Responsible Media Guide brand safety chapter (V1, March 2025) publishes a brand safety floor of ten content categories with descriptions and a brand suitability framework of high, medium and low risk against the same ten, plus an adjacency framework, without naming GARM; its descriptions match the unattributed, undated floor-and-tiers text BSI hosts, which the paper identifies as GARM's June 2022 framework (the WFA's own copy is members-only and was not compared); it drops GARM's debated sensitive social issues and misinformation categories; and ISBA's hub page records the guide as 'Updated May 2026', says it is overseen by ISBA's Media Leaders group and pledges regular review and update.8, 12direct recordsupported with qualificationhigh
Inference: on the registered H6 rule, the screen to 10 November 2026 finds one candidate, ISBA's Responsible Media Guide (brand safety chapter V1, March 2025; guide "Updated May 2026"), that meets the dated-edition and floor-and-tiers conditions with a stated review-and-update pledge but no change log or published procedure, so H6 is contradicted if that pledge counts as a stated change process and unresolved if a documented procedure is required, while the other nine screened bodies publish no such edition.8, 12inferencesupported as statedmedium
DoubleVerify's suitability framework is its own taxonomy described as aligned with GARM: releases of January 2023 (aligned with 4A's APB and GARM standards), December 2023 ('GARM-aligned' settings) and January 2024 (DV's 'Brand Safety Floor and Suitability Tiers' mapped to the GARM framework) name no edition; its 10-K offers over 100 configurable avoidance categories classified by AI with human expertise in over 40 languages; Google's DV360 help exposes two DV risk tiers (high, medium) and 20+ categories; and none of DV's product pages read on 10 November 2026 names GARM or publishes category definitions.8, 12synthesissupported as statedmedium
IAS's brand-safety method as documented: GARM-aligned reporting (Floor plus Low, Medium and High across 'the supported 12 GARM content categories'), an open-web taxonomy of 8 brand-safety categories with 5 risk thresholds, page-level machine learning backed by IAS's own crawlers, frame-level multimedia classification for video, 40+ languages, and methodology that 'may vary' by platform; no dated framework edition, category definitions or thresholds are published on the pages read.8, 12vendor assertionsupported as statedhigh
Zefr's releases named GARM through 2023 (11 GARM categories at the TikTok launch of January 2022; every GARM category including misinformation from February 2023); its September 2024 YouTube and October 2024 Snap releases describe twelve 'industry-aligned' categories without naming GARM, so their link to the archived June 2022 edition is inferred from the count; after GARM's closure only the Media Services page says 'GARM-aligned', and the Standards page names no framework or edition.8, 12inferencesupported with qualificationmedium
Martin v. Zagorski (S.D.N.Y. 1:25-cv-02715), filed pro se on 31 March 2025 against DoubleVerify, Providence Equity and four individuals, is a whistleblower-retaliation and civil-rights conspiracy suit, not a claim about a DoubleVerify product: the amended complaint pleaded conspiracy under 42 U.S.C. § 1985(3), negligence and intentional infliction of emotional distress, and on 16 September 2026 the court dismissed it with prejudice for failure to state a claim (clerk's judgment 17 September 2026; no notice of appeal on the docket at 10 November 2026).9direct recordsupported as statedhigh
The derivative suit Kaszirer v. Zagorski (S.D.N.Y. 1:25-cv-10200), filed on 9 December 2025 against ten DoubleVerify officers and directors over statements from about 10 November 2023 to 27 February 2025, ended with a notice of voluntary dismissal filed on 4 February 2026 and entered as signed on 6 February 2026; whether the dismissal is without prejudice could not be read in any primary document on 10 November 2026.9direct recordsupported as statedmedium
The DoubleVerify securities class action (S.D.N.Y. 1:25-cv-04332), filed on 22 May 2025 for purchasers from 10 November 2023 to 27 February 2025, alleged among other things that DoubleVerify systematically overbilled customers for impressions served to declared bots; the lead plaintiffs voluntarily dismissed it without prejudice on 22 December 2025, before filing any amended complaint, and DoubleVerify's FY2025 10-K reports the dismissal.9direct recordsupported as statedhigh
DoubleVerify's complaint against Adalytics states that its pre-bid filtering keeps customers from bidding on or being charged for most invalid traffic, 'including 99% of unwanted GIVT impressions', concedes that not all invalid traffic can be detected or avoided pre-bid, and says that when pre-bid avoidance is enabled known-bot IVT averages less than 0.03% of the programmatic impressions served to its customers; it says it could identify 43 of the report's impressions in its records, which its blog of 7 April 2025 presents as the examples it could match after 62 of the report's 115 examples lacked DV tags and 10 fell outside its data retention. None of the figures comes with a published test set, denominator or period.9vendor assertionsupported with qualificationhigh
A securities class action against IAS that the brief's register omits, Oklahoma Firefighters Pension and Retirement System v. Integral Ad Science Holding Corp. (S.D.N.Y. 1:25-cv-00847), was filed on 29 January 2025 against IAS, two officers (Lisa Utzschneider and Tania Secor) and Vista Equity Partners over the nondisclosure of price competition (relevant period 2 March 2023 to 27 February 2024, after which the stock fell about 41%); on 2 March 2026 the court dismissed the amended complaint with prejudice for failure to plead a material misrepresentation or omission, judgment followed on 5 March and the plaintiff noticed an appeal to the Second Circuit on 26 March 2026.9direct recordsupported as statedhigh
IAS's FY2024 10-K, filed on 28 February 2025 three weeks after IAS was served with the Oklahoma Firefighters complaint, and its 10-Q of 12 May 2025 both stated that IAS was party to no proceeding expected to have a material adverse effect and did not name the action; IAS first disclosed it in its 10-Q of 7 August 2025 (Note 14, p. 20) and repeated the disclosure in its 10-Q of 4 November 2025, each saying it intended to defend vigorously and could not assess the likelihood of loss; IAS filed no periodic report afterwards, so no IAS filing records the March 2026 dismissal or the appeal.9direct recordsupported with qualificationhigh
Two IAS shareholder derivative suits on the same pricing allegations, Tauber (filed 11 March 2025) and Newman v. Utzschneider (filed 2 June 2025), were consolidated on 12 June 2025 as In re Integral Ad Science Holding Corp. Derivative Litigation (S.D.N.Y. 1:25-cv-02039), stayed pending the securities action, and ended with a stipulation and order of voluntary dismissal entered on 15 January 2026, after the Novacap take-private had closed.9direct recordsupported as statedmedium
Chavez v. Human Security, Inc. (S.D.N.Y. 1:26-cv-03727), a putative class action filed on 5 May 2026, alleged that HUMAN's bot-detection code covertly collects personal data from visitors to more than 1,200 client websites and that HUMAN monetizes the data through advertising companies including The Trade Desk and Magnite; the plaintiffs voluntarily dismissed all claims with prejudice, so ordered on 21 July 2026.9direct recordsupported as statedhigh
A CourtListener party-name search of federal dockets filed since 1 January 2023 (run 10 November 2026) found five naming DoubleVerify, three naming IAS and two naming HUMAN Security; for Pixalate it found two, its contract suit against Criteo (voluntarily dismissed in June 2025) and an unreviewed 'Schedule A' trademark case (Khara Inc., N.D. Ill. 1:23-cv-02880, filed 8 May 2023); the stream's queries found none naming Zefr, Peer39, Adloox, Scope3, Protected Media, Channel Factory, Fraudlogix, TrafficGuard, Illuma or Pixability. Comscore's federal docket count was not obtained, but its FY2025 10-K discloses a February 2026 privacy class action (Singer v. Comscore, C.D. Cal. 2:26-cv-01108), which the court dismissed with leave to amend on 30 June 2026 (Comscore 10-Q of 14 August 2026).9synthesissupported with qualificationmedium
On 20 May 2025 the FTC issued a civil investigative demand in its advertising-boycott investigation (File No. 251-0061) to the Global Disinformation Index asking for documents about entities that rate news sources for misinformation and similar categories, naming Integral Ad Science and DoubleVerify among thirteen such entities; GDI later alleged, on information and belief, that IAS also received a CID, and on 16 April 2026 the FTC withdrew GDI's demand, telling the court that several other CIDs in the same investigation were withdrawn for the same reason.9direct recordsupported with qualificationhigh
On 18 December 2025 IAS announced that it would discontinue the limited input from the Global Disinformation Index in its Context Control Avoidance brand-safety product and review whether its other avoidance tools remain suitable and unbiased; GDI's amended complaint against the FTC attributes the decision to the FTC investigation, which IAS's release does not say.9vendor assertionsupported as statedhigh
Marketing Brew reported on 11 October 2024, from anonymous sources, that Justice Department and NCIS officials had asked advertising executives how they work with IAS and DoubleVerify, including about the tools' effectiveness; both 2025 Senate letter sets cite that article, and no primary record of an investigation was found.9synthesissupported as statedlow
Adalytics' 'AI' brand safety report of 7 August 2024, built on a convenience sample, drew same-day rebuttals from both vendors whose code it showed: DoubleVerify called the results manufactured and said the report lacked campaign settings, and IAS said almost all tags shown came from its publisher products and that the report gave no timeframe or sample.9independent measurementsupported as statedmedium
Adalytics' image-host report of 7 February 2025 relayed advertisers' statements that their vendors marked all measured impressions on imgbb.com and ibb.co safe or suitable and said it was unclear whether IAS's or DoubleVerify's technology identified CSAM there; the same day DoubleVerify said its customers' impressions on the site were 0.000047% of the total media transactions it measured in the preceding 30 days and that page classification starts only once a page reaches a certain ad-traffic level (live text, revised 15 October 2025), and IAS said it labelled the domains ineligible for monetization.9independent measurementsupported with qualificationmedium
Adalytics' bot report of 28 March 2025 showed ads served to bots in crawl datasets and relayed that advertisers' log files and invoices suggested billing for impressions served to declared bots, while disclaiming magnitude and fault; IAS answered on 28 March and 3 April 2025 (URLScan impressions classified as GIVT, SIVT or reduced-value inventory), DoubleVerify on 7 April 2025 (GIVT removed post-bid from billable counts), and the MRC on 11 April 2025 (post-serve GIVT filtration is required, pre-bid filtration is not, and MRC has no insight into billing).9synthesissupported as statedhigh
The NJIT and DeepSee.io study first posted on 3 January 2026 found that 1,416 of 3,219 news articles with any brand-safety classification (43.99%) were unsafe for at least one of DoubleVerify, IAS and Oracle under the authors' rules, and that providers disagreed in 22.98% of cases where two rated the same article; it measures consistency between providers' publisher-side labels, not accuracy. The collection date (September 2024 per the brief) is not stated in arXiv v1 and should be read in the ICWSM 2026 version before it is printed.9independent measurementsupported with qualificationhigh
No SEC filing by any filer contains the word 'Adalytics' (EDGAR full-text search, 10 November 2026), so neither DoubleVerify's suit against Adalytics nor the reports, Senate letters and vendor responses in this register appear in either verifier's filings; DoubleVerify's FY2025 10-K names only the dismissed class action, under a materiality statement.9direct recordsupported with qualificationhigh
On 10 November 2026 DoubleVerify was still a public company with its sale pending (EDGAR shows its latest filing as an 8-K of 9 November 2026 under Item 7.01 naming Neptune BidCo US Inc. as Parent and expecting closing in the fourth quarter of 2026, and no Item 2.01, Form 25 or Form 15), while Integral Ad Science has been owned by Novacap since 23 December 2025.9direct recordsupported as statedhigh
The House Judiciary Committee's interim staff report of 10 July 2024 names Channel Factory among eleven ad-tech providers invited to join GARM's community, says CISA used Channel Factory to collaborate on a mis- and disinformation 'common lexicon', and calls it a YouTube Measurement Program member; these are staff characterizations, not findings.9direct recordsupported as statedhigh
The rest of DoubleVerify's dated legal record from 2023 to 10 November 2026 is allegations, letters and dismissals, not findings: a securities class action filed 22 May 2025 (alleging, among other things, systematic overbilling for impressions served to declared bots) was voluntarily dismissed without prejudice on 22 December 2025; a derivative suit (Kaszirer, filed 9 December 2025) ended with a notice of voluntary dismissal entered 6 February 2026; Martin v. Zagorski, a civil-rights suit, was dismissed with prejudice on 16 September 2026; two unserved stockholder complaints over the merger were pending; Senators Blackburn and Blumenthal (7 February 2025) and Senator Warner (24 March 2025) sent letters asking questions or requesting investigation; and DV booked $3.5 million of 2025 legal fees for litigation and regulatory matters outside the ordinary course.9direct recordsupported as statedhigh
Montgomery Venkersammy v. Human Security, Inc. (C.D. Cal. 2:23-cv-07914) was removed to federal court on 21 September 2023, reported 'completely settled' by mediation on 21 May 2024 and terminated on 22 May 2024; its subject is not established because no document is in RECAP.9direct recordsupported as statedmedium
On 7 February 2025 Senators Blackburn and Blumenthal wrote to IAS's CEO (and separately to DoubleVerify, Amazon, Google, MRC and TAG), asking how ads 'verified' by IAS appeared on a site known to host child sexual abuse material and requesting answers by 14 February 2025; on 28 March 2025 Senator Warner released letters asking the FTC to examine whether vendors including IAS claimed in their marketing to perform real-time bot filtering. These are questions, not findings.9direct recordsupported as statedhigh
A CourtListener party-name search for 'Pixalate' on 10 November 2026 returned five federal dockets, of which only Pixalate, Inc. v. Criteo S.A. (S.D.N.Y. 1:25-cv-03614) falls within Q1's window: a breach-of-contract complaint filed by Pixalate on 30 April 2025, a notice of voluntary dismissal on 18 June 2025 and termination on 20 June 2025; the complaint is not in RECAP, so its subject is not established.9direct recordsupported as statedhigh
The House Judiciary Committee's interim staff report of 10 July 2024 names Zefr among the eleven ad-tech providers GARM invited into its community, describes Zefr as a GARM ad-tech partner whose solutions YouTube incorporates to keep ads off content violating GARM standards, and quotes internal GARM documents; it is a staff characterization, not a finding, and no Zefr response to it was found in Zefr's press pages.9direct recordsupported as statedhigh
X Corp.'s antitrust complaint against the WFA (N.D. Tex., filed 6 August 2024) attached as Exhibit A an internal WFA email chain of 16 March 2022 titled 'Meta: Zefr partnership on post campaign reporting', in which GARM's lead called Meta's coming post-campaign transparency with Zefr 'a significant win we forced' and said he had given Meta and Zefr a quote; Zefr was not a party, and on 26 March 2026 the court dismissed X's claims (with prejudice as to the defendants not dismissed for lack of jurisdiction), a ruling that does not mention Zefr.9direct recordsupported as statedhigh
CourtListener's RECAP index returned no federal docket with Zefr as a party on 10 November 2026 (party-name search, all courts and years); a full-text search for 'Zefr' matched 39 cases and 56 docket entries, of which the X Corp. v. WFA exhibits are the only ones concerning verification on the first results page.9direct recordsupported as statedmedium
AdCP's Measurement Taxonomy, served on 10 November 2026 from the AdCP 3.2 documentation and unchanged on the point in the 3.3.0-beta.5 repository file of 7 November 2026, states that AdCP 'does not adjudicate between competing verification vendors', makes a third party verification's source of truth, and says the protocol runs no measurement models.10direct recordsupported as statedhigh
Since 3.0.0 AdCP has also represented measurement terms that let a confirmed package name the vendor whose count governs invoicing, with a variance tolerance and a maturation window such as post_sivt, and a makegood menu (additional delivery, credit, invoice adjustment) for breaches of a performance standard; the docs call the confirmed package 'the binding contract', but no public agreement using these terms with a verification vendor was found by 10 November 2026.10direct recordsupported as statedhigh
AdCP 3.2's own documentation stops short of settlement: a structured dispute task 'does not exist in 3.2 and is not scheduled for it', a verification vendor's own integration is 'out of AdCP scope', and measurement agent interfaces are 'a follow-up'. The page names RFC #7440 (open, Spec Backlog) as the tracker for a shared dispute lifecycle, and issue #7064, titled a 3.3 epic, which asks what dispute lifecycle should cover invalid traffic, credits and makegoods, was routed to the Spec Backlog on 21 September 2026 with no 3.3 slice selected; both were open on 10 November 2026.10direct recordsupported with qualificationhigh
In AdCP 3.2 an ivt or brand_safety performance standard names the vendor but not which of its metrics is judged: Apostra added a private extension on 22 October 2026 because such standards otherwise read 'Not reported' on its scorecards, and the 3.3.0-beta.5 prerelease of 7 November 2026 adds an optional metric_id, saying that without it no party may guess which vendor metric answers the standard.10synthesissupported as statedhigh
On 10 November 2026 (10:49 and 12:14 UTC) AgenticAdvertising.org's registry API listed 35 agents and returned none of type measurement or governance; the capability block through which a verifier would publish its metric catalogue and accreditations is still experimental (absent from 3.0.0, present from 3.1.0 of 18 June 2026), and the registry says accreditation claims in it are vendor-asserted and not verified by AAO.10direct recordsupported as statedhigh
Checks of twelve domains on 10 November 2026 found no AdCP brand.json declaring an agent for any verification vendor: doubleverify.com, integralads.com, humansecurity.com, zefr.com, peer39.com, pixalate.com, adloox.com, channelfactory.com and protected.media returned HTTP 404, iris.tv returned 403, and apostra.com, where scope3.com redirects, served a brand.json without agents (last updated 1 October 2026); AdCP's docs show a DoubleVerify measurement agent only as an example.10direct recordsupported as statedhigh
Two of the 35 registered agents expose AdCP's content-standards tools, whose calibrate_content and validate_content_delivery tasks return a binary pass or fail verdict, and both pass AAO's hosted content_standards storyboard 10 of 10 steps: vastlint, an individual developer's VAST-linting agent (graded on AdCP 3.1.24 on 10 October 2026), and TVIQ LLC's teqblaze-hosted agent (3.1.33, 10 November 2026); neither is a verification vendor and the test runs on a fictional kit.10direct recordsupported as statedhigh
AdCP's compliance catalogue (3.2.9) lists measurement as a preview protocol with no stable baseline and defers a stable measurement-verification specialism 'until the measurement task surface is frozen', and the 3.3.0-beta.5 specialism directory (32 entries) still has none, so AAO's hosted grader has no role under which a verifier's measurement agent can be certified.10direct recordsupported as statedhigh
AdCP's roadmap (v3.2.9) lists a production evidence program for 'consent-based, machine-observed evidence' under the 3.3 working-group cycle, but issue #7068 (opened 30 August 2026 and open on 10 November 2026, proposing Apostra's Interchange as a first but not sole evidence source) was declared not decided for 3.3 on 31 October 2026 and moved out of the 3.3 milestone (Spec Backlog by 8 November); its experimental receipt schema (draft PR #8040) is unmerged and no implementation matrix has been published, so the protocol body documents no production use of any verification verdict.10direct recordsupported with qualificationmedium
AdCP's Known Limitations page (3.2.9) says AdCP 'is not MRC-accredited and does not seek accreditation', is 'the wire and the contract, not the ledger', and leaves GIVT/SIVT filtration, viewability and brand-safety verification to the delivery stack or a vendor such as DoubleVerify, IAS or HUMAN; its FAQ adds that AdCP does not specify attribution or viewability.10direct recordsupported as statedhigh
DoubleVerify said on 17 June 2026 that clients 'can use Anthropic Claude today' to reach its DV Neura Insight Agent over MCP and that a DV Neura Activation Agent would become available in Q3, said on 17 September 2026 that the Activation Agent was 'expected to become available in the months ahead', and claims 'MCP integrations and ADCP support' on its DV Neura page read on 10 November 2026, while no DoubleVerify agent or brand.json appears in AdCP's public records.10vendor assertionsupported as statedhigh
IAS's agentic product, IAS Agent (announced 16 December 2025 for global rollout in early Q1 2026; its Overview Dashboard Insights feature launched in March 2026), is an assistant inside IAS's own interface that recommends brand-safety and suitability settings; the IAS pages read, IAS's 4,749-URL sitemap, AdCP's registry, AAO's public member list and integralads.com/.well-known/brand.json showed no MCP, AdCP or AAMP interface on 10 November 2026.10synthesissupported as statedmedium
HUMAN's documented agent interfaces are an open-source MCP server for its security products, Sightline Cyberfraud Defense and Client-side Defense (24 July 2025), and an open-source Verified AI Agent demonstration of RFC 9421 agent identity (21 July 2025); neither carries an ad-fraud (IVT) verdict into an agent-buying protocol, and HUMAN's AAO member profile, created 8 June 2026, lists no agents.10synthesissupported as statedmedium
Zefr announced on 1 May 2026 an agentic hub, Zain, with what it called the industry's first AdCP-based buying integration for YouTube and an MCP server for contextual insights, and said on 14 September 2026 that it was 'currently running a closed alpha with a set of partners', adding a Rapid Response Blocking Agent that pushes trending content into exclusion lists before delivery; it names no partner, and no Zefr agent or brand.json appears in AdCP's public records.10vendor assertionsupported as statedhigh
For Peer39, Adloox, Pixalate and Channel Factory this review found no agent-protocol interface: their sitemaps read on 10 November 2026 (peer39.com 296 URLs, adloox.com 75 English URLs, pixalate.com 2,470, channelfactory.com 150) list no page on agents, MCP or AdCP beyond AI-content and user-agent topics, their domains serve no brand.json, and none appears in AdCP's registry or among AAO's public member profiles.10synthesissupported as statedmedium
AAMP's Agentic Real Time Framework v1.0 (released 12 November 2025) names fraud detection pre-impression as a use case and lets a container agent attach OpenRTB metric objects (type, value, vendor, for example viewability) to an impression before the auction; its Go reference implementation includes a metrics-addition sample, not a verification vendor's connector.10direct recordsupported as statedhigh
AAMP 2.3 (30 July 2026) frames trust and safety as a vendor approval gate through IAB's Diligence Platform and SafeGuard Privacy, off by default, and as pricing provenance, while the AAMP buyer agent (2.4.2) models brand safety as excluded IAB Content Taxonomy categories and keywords; its example brief lists IAS and DoubleVerify as required brand-safety vendors in a field no code reads.10direct recordsupported as statedhigh
In the agentic sources screened, 'verification' mostly means agent or seller identity rather than ad verification: AdCP's 'Seller verification' walkthrough checks who is authorized to sell, IAB Tech Lab's Agent Registry offers agent 'identity, verification, and disclosure', HUMAN's Verified AI Agent authenticates agents with RFC 9421 signatures, and DoubleVerify's AI Verification (4 November 2025) identifies AI-agent traffic.10synthesissupported with qualificationhigh
AdCP's taxonomy treats AI-native channels such as ChatGPT as a closed loop in which the seller is also the measurement vendor, while on 5 October 2026 IAS announced third-party brand safety and suitability measurement for ChatGPT Ads in a closed pilot with select advertisers and DoubleVerify said it and OpenAI were developing a brand suitability evaluation pilot; neither is an agent-buying protocol integration.10synthesissupported as statedhigh
AAO's public member profiles (159 on 10 November 2026) include DoubleVerify (profile created 19 May 2026) and HUMAN (8 June 2026), both listing no agents, as well as PubMatic, Butler/Till and Apostra, a founding member; no public profile was found for IAS, Zefr, Peer39, Pixalate or Channel Factory, and under shared rule 12 membership earns no deployment credit.10direct recordsupported as statedhigh
No agent interface or agent-buying use of Adloox verdicts was found: the AdCP documentation corpus (read 10 November 2026) never mentions Adloox and names Peer39 only as an example data provider in illustrative signal examples; Peer39's acquisition release refers to 'future-state agentic workflows', and the seller, Scope3, said suitability will matter more as agents take on buying, both statements of intent.10synthesissupported as statedmedium
Channel Factory markets AERO ('Agentic Execution for ROI') as an agentic system whose agents build brand profiles, curate inclusion and exclusion lists across YouTube and CTV, optimize bidding and report, but its pages name no agent-buying protocol and document no named buyer, transaction or agreement (read 10 November 2026).10vendor assertionsupported as statedmedium
Pixalate sells MCP connectors (Analytics MCP and Enrichment MCP on one remote MCP server) and an 'IQ Agent' that let AI assistants query a client's own IVT reporting and MRT ratings under the client's agreement, with an OpenEPG MCP planned; no AdCP or AAMP implementation, agent-driven buy or agreement giving a Pixalate verdict financial force was found in its documentation index or website sitemap.10vendor assertionsupported as statedmedium
Since AdCP 3.0.0 (released 22 April 2026) the schema has represented five verification-type performance standards (viewability, ivt, completion_rate, brand_safety and attention_score) as rate thresholds that must name a measuring vendor, invalid traffic as a ceiling and the others as floors; the enum is unchanged in 3.2.9 of 9 November 2026.10, 12direct recordsupported as statedhigh
DoubleVerify's blog of 24 July 2026, by its SVP of Product Management, says that 'Today, Butler/Till is using Claude as its buyer agent' to activate media buys through AdCP, connected to DoubleVerify's platform over MCP, and that PubMatic's AgenticOS applies the selected DV protections to inventory packages before the auction; it names the agency but no advertiser, campaign, date range, spend or transaction record.10, 12vendor assertionsupported as statedmedium
Coded against H7's registered rules on 10 November 2026: no demonstrated transaction (state 5) and no identified agreement giving a verification verdict financial force were found; state 1 is met in AdCP since 22 April 2026, in the OpenProposal draft (viewability only) and, for pre-impression fraud and viewability signals, in AAMP's ARTF v1.0; states 2 and 3 are met only by non-verifier implementations (two content-standards agents, Apostra's buyer-side scorecard, an unmerged AAMP seller-agent mapping branch), AAO conformance runs on fictional kits and vendor-asserted connectors; state 4 rests on one vendor blog (DoubleVerify, 24 July 2026) that names the agency but no advertiser, campaign, period or transaction record, describes pre-bid exclusion applied by PubMatic rather than a verdict carried in an AdCP field, and is not corroborated by PubMatic's records, Butler/Till's site or DV's Q3 2026 10-Q. The verifier's reading is that this does not document production use to the standard the shared rules set for vendor sources, so H7 should be coded unresolved on state 4 (a vendor-asserted production claim), not contradicted, with the ambiguity of the registered clause recorded.10, 12inferencesupported with qualificationmedium
Zefr documents agent interfaces but no named agent-driven production use or transaction: it launched Zain on 30 April 2026 (AdExchanger; Zefr's page is dated 1 May) with an AdCP-based YouTube buying integration and an MCP server, said customers had tested it since early April, and on 14 September 2026 described a closed alpha in which unnamed partners activate YouTube campaigns, plus a Rapid Response Blocking Agent; no buyer, campaign or transaction is named.10, 12vendor assertionsupported with qualificationmedium
In AAO's public compliance records for all 35 listed agents read on 10 November 2026, none passed the measurement_terms_accepted storyboard, nine passed measurement_terms_rejected, two passed billing_finality_delivery (a post-SIVT billing-window test naming a fictional measurement vendor; one is the author's company's sales agent), and three passed vendor_metric_accountability, which carries a fictional attention vendor's metric.10, 13direct recordsupported as statedhigh
IAB Tech Lab's OpenProposal (AAMP 3.0), published for comment on 22 September 2026 with the comment period extended from 22 October to 10 November 2026 according to Tech Lab's release page (the repository README still shows 22 October), and not a ratified standard, represents a committed viewable_rate on an MRC basis, a selectable measurement_source vendor (its example offers IAS and DoubleVerify) and a shortfall_remedy of makegood or credit, and has no invalid-traffic or brand-safety guarantee field. On 10 November 2026 none of these fields was on the main branches of the AAMP buyer-agent, seller-agent or iab-agentic-primitives repositories; an unmerged seller-agent branch (feat/openproposal-mapping, last commit 19 October 2026) models all three and maps Google Ad Manager's third-party viewability partner to measurement_source.10, 13direct recordsupported with qualificationhigh
Vendor-reported invalid-traffic and fraud prevalence for periods from 2023 to Q2 2026 spans 0.1% (IAS, optimized CTV, Q2 2026) to 41% among global figures (Pixalate, mobile app, Q2 2026) and to 46.1% with country cuts (Pixalate, UK CTV, Q1 2026), because the vendors measure different bases (open-auction traffic, protected or unprotected campaigns, certified channels) under different definitions (MRC IVT including GIVT, IAS's intent-based IVT, DV's fraud/SIVT violations); none of these figures is an error rate and they cannot be pooled.13synthesiscontradictedhigh
The MRC's content-level brand-safety supplement (September 2018) was sponsored by the 4A's, the ANA and the IAB and approved by the 4A's and the ANA; it makes a false-positive likelihood by service line an optional reporting item ('may include'), but requires services to disclose to users estimates of error where reduced accuracy is expected, confidence levels and standard errors for sample-based determinations and known technical limitations, and lists 'Context and Brand Safety accuracy' among required disclosures; these are disclosures to clients, not published error rates.13direct recordcontradictedhigh
Pixability sells YouTube contextual and brand-suitability segments both inside its managed media service and, under its Self and Self+ models, for activation in the advertiser's own Google Ads or DV360 account; it calls itself a 'YouTube outcomes platform' rather than a 'suitability platform', and its suitability page covers YouTube only. Under the sold-apart-from-media test its Self model is a pre-bid suitability selection product sold apart from Pixability-run media, so Pixability is a cohort C boundary candidate (inclusion segments; boundary with the contextual and curation papers), not a clear listed row (pages read 10 November 2026).13vendor assertioncontradictedmedium
Under the register proposal cohort A has six candidate rows only if HUMAN's advertiser suite and mFilterIt are admitted, and cohorts B and C have five each; but cohort A's six depends on splitting HUMAN into two rows, which the brief's rule against splitting a product to reach six and HUMAN's own description of one Ad Integrity Suite with advertiser and platform offerings (both spanning IVT, viewability and brand safety) put in doubt, while cohort C omits Pixability's and Sightly's self-service suitability products and NewsGuard's page-level ratings, which on the stream's own sold-apart-from-media test are cohort C boundary candidates that could lift cohort C to six or more; candidate rows are not band-eligible rows.13inferencesupported with qualificationmedium
Eight of the nine large platforms document DoubleVerify as a third-party measurement partner on their own pages (Meta, Google, TikTok, Amazon, Pinterest and LinkedIn confirmed in the verifier pass; X and Snap per the profile's browser reads, which this pass could not repeat), and Reddit appears only in DoubleVerify's own release.13synthesissupported with qualificationhigh
Third-party measurement by IAS is documented on all nine large platforms, on the platform's own pages for seven (Meta, YouTube through Display & Video 360, TikTok, Amazon, Snap, Pinterest, LinkedIn) and on IAS's pages only for X and Reddit, whose own help pages could not be read by scripted fetch on 10 November 2026.13synthesissupported as statedmedium
Verve announced on 14 May 2019 that it had implemented Pixalate's IVT detection across its Verve Velocity platform; the author worked at Verve Group from September 2022 to August 2024 (a declared interest), and no source opened shows whether that relationship continued.13vendor assertionsupported as statedmedium
The two verifiers' seller-side lines are defined differently: DoubleVerify's Supply-side is a customer type (publishers, retail media networks, ad networks and exchanges, billed on contracts with minimum guarantees), while IAS's Publisher line is a revenue channel that also includes Publica's CTV ad-serving products; on the last common year, FY2024, the shares are 8.6% (DoubleVerify) and 14.4% (IAS).abstract, 1, 7, 12direct recordsupported as statedhigh
H1a is contradicted on the merged documented grid: 108 of 298 sold units are accredited (36.2%); cohort A has 80/179 (44.7%), B 25/82 (30.5%) and C 3/37 (8.1%). Pooled coverage is below 75% and every cohort is below 50%.abstract, 2, 6, 11, appendixsynthesissupported as statedhigh
The same report contains the only figure in the screened independent record framed as a detection error rate for a named product: treating its crawls as '100% confirmed bot traffic', it reports that IAS's Publisher Optimization 'fr' flag (which IAS's own implementation guide defines as whether the ad represents invalid traffic) marked the declared HTTP Archive bot valid 16% of the time in 2022 to 2024 crawls (17% for declared bots in the executive summary) and the non-declaring URLScan.io bot valid 77% of the time in 2019 to 2024 crawls across over 1,200 websites in North America, Europe, Japan, Singapore, Australia and other countries, without publishing the number of sessions behind either share.abstract, 3independent measurementsupported with qualificationhigh
Smith, Grossman, Torres-Agüero, Sen, Borcea and Chen (arXiv 3 January 2026; ICWSM vol. 20, published 25 May 2026; NSF-funded; one author from DeepSee.io) read DoubleVerify, IAS and Oracle (Moat) brand-safety values exposed by publishers' Google Ad Manager integrations on news articles collected in September 2024 and found 1,416 of 3,219 classified articles (43.99%) unsafe for at least one provider under the authors' own safe-or-unsafe rules, with pairwise Krippendorff's alpha on that label from 0.19 (IAS and DV, 94 Forbes articles) to 0.53 (Oracle and DV, 249 articles).abstract, 3independent measurementsupported as statedhigh
Both full-suite vendors publish protected-versus-unprotected CTV comparisons of the same shape: DoubleVerify reports fraud rates under 1% in protected CTV campaigns against nearly 9% unprotected (release of 7 May 2026), and IAS reports CTV IVT of 0.1% optimized against 9.1% non-optimized in Q2 2026 (21st Media Quality Report); neither publishes the number of campaigns or impressions in each arm.abstract, 3, 12synthesissupported as statedhigh
In the sources screened by 10 November 2026, no settlement paid on a third-party verification verdict was found: the two settlements opened, AdTrader v. Google (US$7 million, approved 1 November 2022, on Google's own invalid-activity detections) and Uber's settlement with Phunware (US$6 million, effective 9 October 2020, resolving Uber's cross-complaint for fraud and other claims), name no verification vendor as their basis.abstract, 4, 12synthesissupported with qualificationmedium
No public evidence found by 10 November 2026 shows verified and unverified impressions of the same inventory clearing at different prices; the closest is publishers' statements to Digiday on 20 December 2017 that one publisher's most and least viewable units differed by only 13% in CPM and that another saw CPMs rise about 20% per 30-point viewability lift, by its own rough estimate, which concern viewability levels, not verification status.abstract, 4, 13synthesissupported as statedlow
On 6 August 2026 DoubleVerify agreed to merge with a subsidiary of Neptune BidCo US Inc., the parent of Nielsen Company (US), LLC, for $13.60 a share in cash; the joint release puts the enterprise value at about $2.15 billion and the proxy the implied equity value at about $2.332 billion; affiliates of Elliott committed $200 million of equity and lenders about $1.8 billion of debt; termination fees are $60 million (DoubleVerify) and $144 million (Parent); and Providence funds holding about 11.8% agreed to vote in favour.abstract, 6, 7, 9, 10, 13direct recordsupported as statedhigh
At 00:11 UTC on 10 November 2026 no closing 8-K, Form 25 or Form 15 was on DoubleVerify's EDGAR filing list (the newest filing was the 8-K Item 7.01 of 9 November) and its investor-relations feed carried no completion release, so the merger was pending and DoubleVerify remained a listed company not owned by Nielsen's parent.abstract, 6, 7, 10, 12direct recordsupported as statedhigh
H3 is supported within the sources screened to 10 November 2026: no independent study reports a scored product's sold-category error rate with a labelled test set and published denominator.abstract, 11, appendixsynthesissupported as statedmedium
H1b is contradicted on the author's interpretation of Peer39's pre-bid accreditation wording and a literal reading of Protected by Mediaocean's CTV display-impression claim.abstract, appendixsynthesissupported as statedmedium
H2 is supported under the author's listed-terms reading of the registered rule: determining effect appears in 2 of 16 readable entries (12.5%), with no qualifying paid settlement or matched verified-versus-unverified inventory price difference found in the screened record.abstract, appendixsynthesissupported as statedmedium
H4 is contradicted at medium confidence under its literal comparator: dated vendor documentation describes verifier tags running through TikTok's Open Measurement SDK integration on TikTok placements.abstract, appendixsynthesissupported as statedmedium
H5 is contradicted narrowly on the author's accepted reading of DoubleVerify's June 2024 structural-separation statement for publisher-client interests as a published safeguard.abstract, appendixsynthesissupported as statedmedium
H6 is contradicted on the author's reading of ISBA's review pledge, named overseer and update trigger as a stated change process for its March 2025 floor-and-tiers guide.abstract, appendixsynthesissupported as statedmedium
H7 remains unresolved on the author's retained interpretation: schemas, connectors and tests are documented, while production claims and partial buyer reporting do not establish the verification-verdict-to-live-agent-decision link; no identified financial-authority agreement was found in the screened record.abstract, appendixsynthesissupported as statedmedium
Machine-readable copy with passages and limitations: data/claim-ledger.csv.

How the products are scored

Each row is a named product serving one buying task. Company-wide evidence does not automatically transfer between products, and aliases create no extra row. Cohort A measures viewability, general and sophisticated invalid traffic (GIVT and SIVT), and safety or suitability across at least two environments. Cohort B sells impression IVT measurement or filtration; cohort C sells content safety or suitability for verification or blocking. Platform first-party measurement, researchers and standards bodies are described separately .

The author made every inclusion, cohort and verdict call. The product register records each inclusion reason, and the corrections ledger logs every change after the verifier pass, starting with the register decisions of 10 November 2026, taken before any cell was scored. The sweep had proposed two rows for HUMAN's Ad Integrity Suite, which would have brought cohort A to the size that bands need; its verifier objected that the brief forbids splitting a product to reach it, and HUMAN describes one suite . The suite is scored once, in cohort B, where its primary output and accredited scope sit. Channel Factory's ViewIQ entered cohort C on terms of 17 August 2026 that sell it apart from media ; Pixability and Sightly entered on the same test, pending the second wave's check of their buying task .

The dimensions, anchors and masks were fixed on 9 November 2026: cohort A takes all twelve dimensions, B ten and C eleven, and n/a (not applicable) comes only from the mask.

Group and weightDimensionWhat it asksCohorts
Depth, 0.45Method disclosureHow each sold metric is determined, as publishedA, B, C
Depth, 0.45Independent detection evidencePublished error-rate tests with a labelled test set and a denominatorA, B, C
Depth, 0.45Invalid-traffic methodGIVT and SIVT scope, pre-bid and post-bid, per environment soldA, B
Depth, 0.45Safety and suitability methodFramework and dated edition, content-level or placement-level, languages and mediaA, C
Depth, 0.45Verdict uses and authorityHow far the verdict travels toward settlement, from report only to a documented settlementA, B, C
Reach, 0.25Environment coverageLive environments among desktop, mobile web, in-app, CTV, audio and digital out-of-homeA, B, C
Reach, 0.25Platform accessDocumented partnership on the nine large platformsA, C
Reach, 0.25Measured scaleImpressions or transactions measured in a stated 12-month periodA, B, C
Reach, 0.25US availability and useDocumented US availability and production useA, B, C
Standing, 0.30AccreditationAccredited units as a share of sold units, on MRC's recordsA, B, C
Standing, 0.30Interest disclosure and safeguardsRevenue by customer type, sell-side products, a conflict or separation policyA, B, C
Standing, 0.30Dispute and correction processA dispute route with time limits, a correction policy, documented correctionsA, B, C

A 0 requires the anchor's documented absence or failure, never an allegation. Missing evidence is n/e. Cells record availability, date, rationale, sources and evidence grade; vendor technical documentation is capped at 4 and marketing at 3. The masks alone determine n/a. Announced or retired products are listed, and admission needs four counted cells.

Claude agents scored the calibration rows with separate refute-mode verifiers. An OpenAI model scored the added run-1 rows and then self-checked each in a refute-mode pass. A supplemental consistency review inspected the added rows and led to the Protected dispute-process correction. The author checked the work and retained the primary scores. Only verified and qualified cells count; an unverified cell is treated as n/e. A scorer's self-check is not a blind second score. There is no human second scorer and no inter-rater statistic.

Cross-row interpretations after extraction

Cross-row interpretations identified as R1 to R8 were made after extraction on 10 November 2026. They are post-extraction interpretations, not rules fixed before research: R1, an environment counts as sold where the vendor says it measures or protects advertisers' traffic there; R2, off-platform ad networks such as Pangle and Audience Network are in-app inventory, not platforms; R3, one H1a unit per product, metric family and environment, pre-bid and post-bid counted once; R4, OpenPath clause 4.4 gives anchor 4 only where the product is MRC-accredited for the relevant metric and documented as utilized by The Trade Desk or its Ad Partners; record the evidence grade for both conditions and apply the applicable cap; R5, a vendor's own volume statement with a unit and period scores at grade D; R6, sources a scorer opens are registered before they are cited; R7, one standard for every row, whatever the vendor's size; R8, Pixability and Sightly are scored only if sold apart from managed media, which the sweep verifiers confirmed. The original registered anchors remain distinguishable from those later interpretations.

The Verdict Quotient

The VQ combines documented capability within a cohort. It averages counted cells within Depth, Reach and Standing, skipping n/e, n/a and unverified cells. Raw is their weighted mean at 0.45/0.25/0.30, renormalised if a group is missing. F is the mean Raw score of admitted cohort rows, including thin records. Shrunk = (k × Raw + K × F) / (k + K), with counted-cell total k and prior strength K = 4. VQ = 100 × Shrunk / 5, with halves rounded up only for display. Accuracy and legal safety remain outside the index.

K is the author's heuristic rather than a fitted statistical parameter. Shrinkage moderates thin records' observed averages without imputing missing cells. The VQ borrows the Contextual Quotient's band offsets but uses a different field mean; no validation follows from their resemblance.

Bands come from the unrounded offset 20 × (Shrunk − F), and equal values share a band.

BandOffset, 20 × (Shrunk − F)
Most documented+12 or more
Well documented+1 or more, below +12
Partly documented−8 or more, below +1
Least documentedBelow −8

Band eligibility requires 6 counted cells and one in each applicable group. A cohort needs 6 eligible rows for bands; rows missing a group show which one, without a band. Chapter 11 reports the tests below.

TestWhat it changesWhat is reported
ReweightingDepth, Reach and Standing at equal thirds, 0.60/0.20/0.20, 0.30/0.45/0.25 and 0.30/0.20/0.50, with Raw and F recomputed across the fixed cohortHow far quotients and bands move
Prior strengthK = 0 (Raw), 2 and 8How far quotients and bands move
One-cell deletionEach counted cell turned into n/e in turn, with k and eligibility recomputedQuotient change, admission loss and eligibility loss, separately
Missing-evidence scenariosEvery applicable n/e cell of one row filled with 1, then 4, with F and every other row frozenRows whose band changes; imputed cells never create admission or eligibility
EnvelopeThe same fills at the scale's ends, 0 and 5Reported apart, never used to flag a band
CoverageCounted cells as a share of applicable dimensionsReported for every row

A band changed by one-cell deletion or a missing-evidence scenario is marked sensitive. Such ordering supports no buyer recommendation. A dated adverse-record 0 remains supported by its archive unless superseded.

The review of the finished paper

The finished paper was submitted to an outside AI review. Review findings and any resulting changes must be recorded in the corrections ledger before publication.

The corrections ledger holds every substantive change made after the verifier pass, with what questioned it, what was checked, the verdict, the change and how it was verified.

Corrections ledger

Every substantive change made after the verification and challenge passes, including withdrawn claims. Already reflected in the chapters; reproduced so the difference between first research and final text can be audited.

IDOriginal claim or scoreChallengeEvidence checkedVerdictChange madeReasonAffected
M01Workstream U proposed two rows for HUMAN: the Ad Integrity Suite for Platforms (cohort B) and the Ad Integrity Suite for advertisers and agencies (cohort A), which would have given cohort A six rows.U's verifier objected: the brief forbids splitting one product to reach a cohort's size, and HUMAN describes one Ad Integrity Suite with advertiser and platform offerings, both spanning invalid traffic, viewability and brand safety.HUMAN's releases of 16 June 2026 (advertisers and agencies) and 5 October 2026 (platforms); MRC's letter of 14 April 2026 (pre-bid and post-bid GIVT and SIVT in four environments; display and video viewability in three). structuralOne row, HUMAN Ad Integrity Suite, scored in cohort B, where its primary output and its accredited scope sit; the advertiser edition's brand-safety scope is recorded as an attribute. Cohort A keeps five rows and is shown without bands.One product, one row (the brief's rule); the cohort follows the primary output.data/product-register.json; cohorts A and B; chapter 11
M02Workstream U listed Channel Factory's ViewIQ as bundled with managed media, so not scored.The deep profile found terms of 17 August 2026 for a self-service software platform (free accounts and paid plans billed monthly) whose classification produces inclusion and exclusion lists the buyer runs in Google Ads or Display & Video 360.Channel Factory's Terms and Conditions (17 August 2026), read in the deep profile and checked by its verifier. contradictedViewIQ admitted to cohort C on the sold-apart-from-media test.The cohort rule turns on whether the product is bought apart from media; the terms show it is.data/product-register.json; cohort C
M03Workstream U listed Pixability and Sightly as media-bundled suitability services, not scored.U's verifier contradicted the listing: Pixability sells YouTube contextual and suitability segments under its Self and Self+ models for activation in the advertiser's own Google Ads or DV360 account, and Sightly sells a self-service list builder for YouTube suitability, both apart from managed media.Pixability's home, brand-suitability and self-serve pages; Sightly's home and self-service pages (10 November 2026). contradictedBoth admitted to cohort C, conditional on the wave-two product sweep confirming the buying task.Same sold-apart-from-media test as M02.data/product-register.json; cohort C
M04Anura was proposed as an invalid-traffic product for cohort B.Anura Ad Tag Protect protects a publisher's or advertiser's own accounts and sites; cohort B covers impression-level invalid-traffic detection sold to the buy side or trading platforms.Anura's Ad Tag Protect and Ad Fraud Detection pages (10 November 2026). supported with qualificationListed, not scored, as a publisher account-protection tool.The cohort definition fixed in brief v3 (9 November 2026).data/product-register.json
FR-01Verdicts are schema fields in agent buying; no transaction and no binding agreement were found.The protocol-state figure remains unsynchronised with the reviewed narrative. HUMAN ad-fraud is assigned a vendor connector for a server covering other security products; IAS Agent gets a connector despite no documented protocol interface; Pixalate is grouped under all-none despite its Analytics/Enrichment MCP documentation. The AdCP content-standards interoperability row also says none while Chapter 10 records passing tests for non-verifier agents. Separate creative-verdict tests from verification-product tests. Rebuild the figure, source data, HTML and CSV from one adjudicated state table; changing only the title cannot fix the cells.Primary sources reopened 11 November: https://www.humansecurity.com/learn/blog/connect-ai-to-security-data-with-human-mcp-server/ (What You Can Do: Sightline Cyberfraud Defense and Client-side Defense); https://integralads.com/insider/ias-agent-always-on-campaign-insights-built-for-action/ (Signal UI feature); https://docs.pixalate.com/mcp/overview (Products and Connect your assistant). Pack data/figure-data.json#protocols and Chapter 10 tables show the contradictions. These establish connector scope, not a transaction or financial authority. contradictedapplied: protocol states adjudicated once in scripts/derivations.mjs (HUMAN and IAS Agent no connector; Pixalate's MCP server added; AdCP content-standards tests by non-verifier agents shown)Final outside review (OpenAI), 11 November 202610-agents
FR-02The ± flag here records a band change when missing evidence is filled with the stated scenario values.The publication rule requires a sensitivity warning for either a single-cell deletion or a missing-evidence fill. All eight C rows change band or lose eligibility under a deletion, yet the visible matrix flags only five scenario-sensitive rows. Peer39, Proximic and Zefr are incorrectly unflagged. Add deletion sensitivity as a distinct visible field or combine it into the publication flag and regenerate text, matrix, SVG and data.Pack data/scoring-rules.json composite_index.missing_rule and Chapter 13 require deletion-sensitive publication labels. data/vq-missing-evidence.csv records five Peer39 band changes, two Proximic band changes, and one Zefr change. The 23 total includes six Mantis eligibility losses. This is a directly reproduced internal rule violation. contradictedA published sensitivity flag must cover a band change or eligibility loss under any single-cell deletion, as well as a band change under either missing-evidence scenario. The table separates these tests; every cohort C row is deletion-sensitive under the current scores. Scenario sensitivity alone applies to five rows.Final outside review (OpenAI), 11 November 202611-vq
FR-03Pooled coverage is 107 of 296 (36.1%).The arithmetic reproduces the packed units, but the grid omits Zefr YouTube GIVT while counting equivalent net-of-GIVT reporting as a separate accredited family for DoubleVerify and IAS. Zefr's ADH letter and its own accredited reporting include net-of-GIVT impressions. Apply one rule to all three: if sold GIVT-filtered reporting creates a family unit, add Zefr's YouTube unit to both numerator and denominator; otherwise remove analogous inferred units consistently. Do not hard-code a corrected pooled total until all unit repairs are merged. H1a remains contradicted on these narrow variants.https://mediaratingcouncil.org/sites/default/files/Accreditation%20Letters/Zefr/GI110625_Zefr%20Google%20ADH%20Accreditation%20Letter.pdf, pp. 1–2, explicitly includes GIVT filtration and Total Impressions (Net of GIVT). Compare data/json/h1a-units.json: Zefr YouTube has content-level and video-viewability units but no GIVT; the DV and IAS rows include YouTube GIVT. Chapter 13 R3 requires a common product/family/environment grid. contradictedapplied: Zefr's accredited YouTube GIVT unit added to the H1a gridFinal outside review (OpenAI), 11 November 202602-accreditation
FR-04The merged documented grid records 107 accredited units out of 296 sold (36.1%):Sightly's documented generic-web/programmatic SmartLists unit is acknowledged in its accreditation cell but omitted from the raw H1a grid because of the former validator vocabulary. The same grid already permits descriptive generic-web environment labels for Zefr. This is an inconsistent denominator rule. Add the documented device-unstated unit using a canonical label, or publish an explicit common exclusion policy and corresponding sensitivity. Repair raw units, derived checks, figure data, abstract, appendix, claim synthesis and plain edition together.https://www.sightly.com/self-service-media/, SmartLists for Programmatic, documents content-level inclusion/exclusion lists; reopened 11 November. Pack capability-matrix.json Sightly/accreditation explicitly records 0/2 if generic web is admitted, but data/json/h1a-units.json contains only its YouTube row. Zefr already has device-unstated web labels. This is a known documented omission, distinct from an unknown total denominator. contradictedapplied: Sightly's device-unstated web content-level unit addedFinal outside review (OpenAI), 11 November 202602-accreditation
FR-05Mantis remains in cohort C under the author's inclusion ruling.The admission rule requires a sold live/beta product, yet every numeric Sentinel cell is unconfirmed and the paper says separate purchase and live/beta availability remain unresolved. An author inclusion ruling does not establish the required facts. Either document those gates or list Sentinel outside the primary scored cohort and recompute the field and every dependent band. Holding all other cell scores fixed and excluding Mantis changes the cohort field from 37.42 to 39.05; Channel Factory and Peer39 both become Partly documented. Keep any exception in a clearly labelled secondary analysis.https://www.mantissolutions.com/for-advertisers-and-agencies, Mantis Sentinel (Monitoring Tag), reopened 11 November: it describes a tag and outputs but supplies neither standalone purchasing terms nor a live/beta status. Pack data/scoring-rules.json selection_rules.scored_set requires all admission conditions. capability-matrix.json has six unconfirmed numeric Mantis cells; independent recomputation in mechanical-checks.json and REPORT.md. No closure or nonavailability is inferred from the evidence gap. contradictedapplied: Mantis Sentinel listed, not scored; cohort C recomputed (Channel Factory and Peer39 now Partly documented); text updatedFinal outside review (OpenAI), 11 November 202611-vq
FR-06R4, The Trade Desk's OpenPath clause 4.4 gives anchor 4 on verdict authority to any product MRC accredits for invalid traffic or viewabilityR4 drops the contract's utilization condition and turns accreditation alone into financial authority. The HUMAN and Pixalate rationales document use as well as scope, but Protected is raised on the class ruling alone. Accreditation is necessary for that clause, not sufficient proof of the full class condition. Preserve the qualifier and apply the same utilization-source grade to every row; vendor-only proof cannot borrow grade A from its accreditation letter to evade the grade-D cap.https://assets.thetradedesk.com/assets/documents/OpenPath-Publisher-Terms-and-Conditions-v.10.31.24.pdf, p. 4, clause 4.4: the class is MRC-accredited services used by TD or its Ad Partners. The short operative qualifier is “utilized by TD and/or its Ad Partners”. Compare HUMAN/verdict_authority and Pixalate/verdict_authority with Protected/verdict_authority in the matrix. contradictedR4, OpenPath clause 4.4 gives anchor 4 only where the product is MRC-accredited for the relevant metric and documented as utilized by The Trade Desk or its Ad Partners; record the evidence grade for both conditions and apply the applicable capFinal outside review (OpenAI), 11 November 202613-method
FR-07The supported reading therefore needs an explicit population and coding ruling before publication [219, 222, 234, 225, 242, 230, 88].The owning chapter still treats the H2 ruling as pending, while the abstract and appendix assert the author has made it. That leaves incompatible descriptions of the final verdict. Keep the listed-terms operational-rule reading and explicitly preserve the wider-population sensitivity; do not silently redefine the entire registered population.Internal source: data/hypotheses.json H2 population includes public contracts, while rule_supported refers to listed readable terms. Chapter 4 presents both codings; the abstract records the author's retained listed-terms reading. OpenPath clause independently confirms a counterexample but cannot settle the population choice. contradictedThe author retains H2 as supported under the registered listed-terms operational rule: 2/16 readable listed entries, or 12.5%. The broader population and treatment of review-qualified or trigger-only clauses remain reported sensitivities, not an unconditional claim across all public contracts [219, 222, 234, 225, 242, 230, 88].Final outside review (OpenAI), 11 November 202604-money
FR-08An alternative reading is that the signal incorporates accredited post-serve data, so the inference needs the author's explicit approval [16, 150, 49, 367].This still requests approval that the final abstract already attributes to the author. Preserve the alternative reading while describing the call as settled.https://www.peer39.com/verification, Viewability, and the Adloox letter of 29 June 2026, p. 1, were reopened. The vendor wording supports the ambiguity; no MRC finding resolves it. The final abstract explicitly records the author's interpretation. supported with qualificationAn alternative reading is that the signal incorporates accredited post-serve data. The author retains the broader-scope interpretation for H1b; it is an author judgement, not an MRC finding [16, 150, 49, 367].Final outside review (OpenAI), 11 November 202602-accreditation
FR-09it certifies a measurement process against MRC's standards and does not test accuracyThe categorical claim that accreditation does not test accuracy overstates the justified distinction. MRC expressly describes accuracy/functionality testing in pre-bid audits. What the public letters do not supply is an independently reproducible sold-category error-rate estimate of the kind H3 defines.https://mediaratingcouncil.org/sites/default/files/News/MRC%20Statement%20on%20pre%20bid%20IVT%20requirements%20and%20processes.pdf, p. 6, says audit scope includes file/database production, accuracy, functionality and availability. This does not establish a published product-wide miss rate. contradictedit certifies a scoped measurement process against MRC's standards; the public accreditation letter does not supply the sold-category error-rate estimate defined in H3Final outside review (OpenAI), 11 November 202605-platforms
FR-10Accreditation certifies that a measurement process meets a standard; it is not a test of accuracy.The decision-tool row repeats the same overly categorical accuracy claim. Align it with the audit scope rather than suggesting MRC audits never test accuracy.Same primary MRC statement, 11 April 2025, p. 6: https://mediaratingcouncil.org/sites/default/files/News/MRC%20Statement%20on%20pre%20bid%20IVT%20requirements%20and%20processes.pdf. The distinction is a scoped audit versus the public labelled-set error-rate evidence requested in H3. contradictedAccreditation certifies conformance of a scoped measurement process; the public letter does not provide the labelled-set product error rate requested in H3.Final outside review (OpenAI), 11 November 202612-next
FR-11scores as of 11 December 2026The quotient figure caption presents a future freeze as an observed score date, contrary to the November-read caveat. Use the actual scoring/read period and keep the planned freeze separate.Internal chronology: the final pack was built and supplied on 11 November 2026; Chapters 00 and 13 say the 11 December recheck has not happened. figures/fig-quotient.json and its generated caption carry the premature date. supported with qualificationscores from November 2026 evidence; planned freeze recheck on 11 December 2026Final outside review (OpenAI), 11 November 202611-vq
FR-12https://assets.thetradedesk.com/assets/documents/legal/OpenPath-Publisher-Terms-and-Conditions-v.10.31.24.pdfThe registered OpenPath source URL failed to open; the same current primary PDF is available without /legal/. Repair its canonical source-register entry and regenerate numbered references and evidence links. This is a retrievability defect, not a failure of the contract claim.Opened successfully on 11 November: https://assets.thetradedesk.com/assets/documents/OpenPath-Publisher-Terms-and-Conditions-v.10.31.24.pdf, p. 4 clause 4.4. The registered /legal/ route returned an error in this review. supported with qualificationhttps://assets.thetradedesk.com/assets/documents/OpenPath-Publisher-Terms-and-Conditions-v.10.31.24.pdfFinal outside review (OpenAI), 11 November 2026references
FR-13| 14 Aug 2023 | Scibids | Acquired by DoubleVerify (campaign optimization) | Acquisition (completed) | $121.4m | [275, 274, 26] |The acquisition table retains the earlier Scibids price without its basis even though the FY2025 10-K gives adjusted consideration of $121.1m net of cash. Distinguish the initial figure from the later adjusted figure. Update duplicated claim-ledger/profile values as well, rather than silently mixing gross, net and preliminary values.https://www.sec.gov/Archives/edgar/data/1819928/000110465926020499/dv-20251231x10k.htm, Note 4 Scibids Technology SAS, printed p. 84: $121.1m net of cash including a $0.3m measurement-period decrease; table total $121.070m. Reopened 11 November. supported with qualification| 14 Aug 2023 | Scibids | Acquired by DoubleVerify (campaign optimization) | Acquisition (completed) | $121.1m net of cash, adjusted in the FY2025 10-K | [275, 274, 26] |Final outside review (OpenAI), 11 November 202607-market
FR-14The finished paper went to one outside AI review before publication; its findings and the changes they led to are in the corrections ledger.The review pack describes final-review findings and resulting changes as already logged, but its corrections ledger contains only M01–M04 register decisions. The current review has not yet been applied. Retain accurate draft wording now, then log each substantive disposition and change before changing to completed-history wording.Internal primary record: data/corrections-ledger.json contains four entries, dated before this final review, mirrored verbatim in Chapter 13. This review delivers new findings; they cannot already be in that frozen ledger. contradictedThe finished paper was submitted to an outside AI review. Review findings and any resulting changes must be recorded in the corrections ledger before publication.Final outside review (OpenAI), 11 November 202613-method
FR-15Provisional, rewritten from the findings: An internal drafting instruction remains in the public subtitle. Remove the editorial prefix while retaining the substantive description.Exact text on line 3 of the reading edition; the HTML title/subtitle should be regenerated from the same canonical source. No external fact is asserted. supported with qualificationEditorial drafting prefix removed from the public subtitle.Final outside review (OpenAI), 11 November 2026front-matter
FR-16| Escalated | Escalated IVT scoring | fraud | 46 (cohort too small) | 1 | 2 | 3 | n/a | 2 | 3 | n/a | n/e | 3 | 1 | 3 | 1 | 9/10 |Cross-row method calibration is inconsistent: Escalated scores 1 despite its rationale explicitly recording an hourly list cadence; the same anchor is described as signals/models only without thresholds or cadence, while Zefr gets 3 partly because cadence is present. Escalated's rationale also demands thresholds/model validation for anchor 3, which only requires a method summary for at least half the sold metrics. Re-open the scored metric coverage and apply the same boundary to all rows, including the illuma alternative. Do not automatically promote a score from the existence of cadence alone.Pack capability-matrix.json Escalated/method_disclosure, Zefr/method_disclosure and dimension anchors; https://escalated.io/features.html (list updates and integration features) and https://escalated.io/integrations.html (API, pixel and blocking routes), reopened 11 November. The matrix itself admits the cadence while imposing an extra threshold requirement. This is a calibration finding, not proof that the product deserves a particular replacement score. contradictedmethod_disclosure 1 -> 3, grade D; rationale rewritten to the registered anchorFinal outside review (OpenAI), 11 November 202611-vq
Machine-readable copy: data/corrections-ledger.csv.

Limits

Public records leave executed contracts, recovered credits, logs and vendor test sets largely inaccessible. The vendor/commissioned screen found no sold-category error rate on a labelled holdout with a denominator ; the pricing screen found no matched comparison of verified and unverified impressions . Advertisers supplied logs to some studies, including the ANA's 2023 study . Sparse public disclosure can understate a product's capability.

Status-sensitive facts were read on 9 and 10 November 2026 and are dated in the text: MRC's list carries no update stamp, and DoubleVerify's sale was pending at the last check . The evidence cutoff is 11 November 2026; those facts are read again on 11 December 2026, and later changes are to be logged separately.

Absences are bounded to the logged source screen and date. Failed access establishes no absence. The legal record excludes sealed and nonpublic matters and does not cover every state docket (Chapter 9).

Agents misread sources. Separate refute-mode verifiers checked the original Claude research files; an OpenAI model completed later sweep checks and chapter fact-checks and self-checked its added scoring rows. The author checked the combined work. These checks can share model or source assumptions. Corrections are logged, scores have no human second scorer, and the US dimension counts documented US availability only.

Interests

The author has worked since May 2025 at Samba, the TV measurement and advertising-data company formerly branded Samba TV. Nielsen competes with Samba in TV measurement, and on 6 August 2026 DoubleVerify agreed to merge with a subsidiary of Neptune BidCo US Inc., Nielsen's parent ; the merger was pending at the last check, on 10 November 2026 . The paper says so wherever the transaction is discussed at length, and DoubleVerify is scored under the same published rules as every other vendor.

He co-leads the Signals & Measurement working group of AgenticAdvertising.org, which maintains AdCP and whose scope touches verification signals, so workstream J's findings on AdCP are a declared interest. No Fluff Advisory, his firm, operates a signals agent and a sales agent in that organisation's registry ; in records read on 10 November 2026, the sales agent is one of two that passed a storyboard testing billing after a post-SIVT window, which names a fictional measurement vendor . Comments he filed on IAB Tech Lab's OpenProposal, the AAMP 3.0 draft , are registered as his own work, an interest rather than evidence, and none concerns verification fields .

Earlier he was an executive vice president at Verve Group, an advertising platform whose supply the verifiers measure, from September 2022 to August 2024; Verve announced in May 2019 that it had implemented Pixalate's invalid-traffic detection, and no source opened shows whether that continued . He has also worked at Hearts & Science, an agency that bought verification for its clients, and at Lotame. His essays arguing that a verification verdict "moves nothing" and that advertising has no adjudicator for disputed conversion counts are rival framings tested by H2, not evidence.

No vendor funded, reviewed or saw the paper, and no vendor or third party was contacted while it was prepared, so no correction invitation could be mistaken for editorial control. The paper, its data and its review prompts name no company as a consulting client. Every vendor is scored under the same published rules.

Appendix

Appendix: hypotheses, products and downloads

The registered hypotheses and how they read

The hypotheses and their decision rules were registered on 9 November 2026. The readings below apply to November evidence, before the planned 11 December 2026 freeze recheck. The author's 11 November rulings retain H1b, H5, H6 and H7's chapter readings and settle H2 on its listed-terms rule. Their alternative interpretations remain part of the record.

IDHypothesisReadingBasisChapter
H1aAt least three quarters of sold units hold MRC accreditationContradicted108/298 pooled (36.2%); A 80/179 (44.7%), B 25/82 (30.5%), C 3/37 (8.1%); unknown denominators remain unknown 2
H1bProduct accreditation claims stay within listed scopeContradicted, author's interpretationPeer39's pre-bid wording and Protected's CTV display claim exceed scope on the retained readings; no MRC finding is inferred 2
H2Third-party verdicts rarely determine paymentSupported on the listed-terms rule2/16 readable entries (12.5%); wider-contract and qualified-clause sensitivity retained; no qualifying paid settlement or matched price difference found in the screen 4
H3No independent public sold-category error rate is established for a scored productSupported within the screenNo qualifying labelled-set test with a published denominator found; IAS's publisher-tool candidate is outside the scored task 3
H4Every documented platform route uses platform-supplied dataContradicted, medium confidenceTikTok's dated measurement-kit route meets the verifier-executed comparator; platform control and route freshness qualify it 5
H5Seller revenue or interested ownership has no published safeguardContradicted, narrowlyAuthor accepts DoubleVerify's separation statement for publisher clients; stricter policy reading retained 7
H6No maintained successor floor-and-tiers framework existsContradicted on the author's process readingISBA's pledge, overseer and trigger accepted; formal-procedure reading unresolved; product implementation clause unestablished 8
H7Verification verdicts stop at schemas, without production use or financial authorityUnresolvedConnectors and tests exceed schema-only use; production evidence lacks the verdict-to-decision link and an identified financial agreement 10

The coded questions

IDQuestionWhere answered
Q1Which letters, proceedings, filings and reports concern the products, and what did vendors answer?Chapter 9, outside the quotient
Q2What does each product measure, through which method, at which stage, and with what action and coverage?Chapter 1, with platform routes in Chapter 5 and CTV and in-app detail in Chapter 6
Q3How has the market consolidated since 2021?Chapter 7, keeping agreements, approvals, completions, rebrands and retirements distinct
Q4Which standards and versions govern the services and assurance?Chapter 2, with CTV and in-app specifications in Chapter 6

The product universe

The register records each company and purchased service, its owner, cohort, status and inclusion reason. Listed and historical records are described without receiving a zero score. The matrix compares products within their registered buying task. Mantis Sentinel is listed, not scored: the final review found neither separate purchase terms nor a live or beta status for it.

Downloads

The datasets expose the source and claim registers, unit grid, scoring rules, cells and derived checks. Read the status dates and access limits alongside the data. The December freeze is a scheduled recheck, not a completed review in this November draft.

Downloads

The two other editions, then the data as CSV and JSON, UTF-8. The data files are the canonical files every chart, matrix and count in the paper was built from.

References

Numbered sources

The 1069 sources cited in this paper, numbered in order of first citation. The full register of 1458 source records, with access dates, access limits and reliability notes, is in data/source-register.csv. Click any number in the text to open the same record.

1069 numbered sources