# What Can Leave the Room?: What buyers can receive, keep, use again and rebuild from nine data-collaboration products

*Evgeny Popov, No Fluff Advisory · Revised edition · evidence cutoff 9 October 2026 · published 9 October 2026*  
https://nofluffadvisory.com/research/data-collaboration-2026/

**Current status, 9 October 2026.** An earlier edition was published on 9 October 2026 and withdrawn the same day for further review. This revised edition, which answers two critical reviews, was released on 9 October 2026. The evidence cutoff remains 9 October 2026. The update history follows the method section.

## Abstract

An output can leave a clean room and still be hard to use after the deal ends. What matters is what leaves, who receives it and what they may do next. This paper follows fourteen output routes across nine product boundaries — Snowflake, Databricks, AWS Clean Rooms, Google Ads Data Hub, BigQuery clean rooms, InfoSum, LiveRamp, Amazon Marketing Cloud and Meta Advanced Analytics — through public documentation as of 9 October 2026. Public sources describe how outputs are received and some limits on their use, including published policies. They do not settle each buyer’s rights after exit. This study read no private contract and tested no move to another environment. It does not rank vendors; it sets out the assets, terms and tests a buyer needs to check whether a task can continue.

**Five summary claims**

1. Public sources describe how outputs are received and some limits on their use, including published policies. They do not settle what a buyer may keep or reuse after exit, so those answers are recorded as not established — neither granted nor denied.
2. Outputs differ, so each needs its own test. AWS list rules return row-level overlap lists, Snowflake’s clean-room agent can send overlap records to an approved account, and AWS can export a trained model to a named member [10, 7, 11].
3. Keeping an output depends on events the buyer may not control. Databricks outputs last 30 days or until the room is deleted; an InfoSum member that leaves a collaboration can no longer see its event logs [9, 17].
4. Proof must cross the boundary. AWS job logs do not report delivery, a Snowflake activation needs the recipient to import it, and a deletion record covers only its named object [27, 6, 33].
5. The dates are scheduled, not certain. Databricks retires an MCP endpoint on 31 October 2026, Snowflake schedules its legacy clean rooms to close in two steps in 2027, and Publicis–LiveRamp was pending at the cutoff [3, 2, 5].

## 1. The buying decision

*Clean rooms are sold on what stays in. A buyer pays for what comes out, and for what it can still do after the contract ends.*

**In short.** Before signing, name the task you must continue after exit. Then establish which outputs, permissions and dependencies that task needs. An export option alone does not settle whether the task can continue.

Clean rooms promise that raw rows never move and that partners never see each other’s tables. That is the privacy story, and it is often true. The purchase turns on the other side of the boundary. A buyer wants something back: a report, an audience, a model or a list of matched records. It also needs to know what it can still do with that result when the contract ends. *(inference)*

Scheduled dates make the question practical. Snowflake stopped new legacy clean rooms in its web interface on 1 October 2026 and schedules the legacy interface to close and the creation of new rooms through the legacy API to stop on 1 February 2027, and access to existing legacy rooms to end on 1 June 2027 [1, 2]. Databricks schedules an older MCP endpoint to sunset on 31 October 2026 [3]. Publicis agreed on 17 May 2026 to buy LiveRamp; at the cutoff the deal was pending [4, 5]. *(fact; claims C003, C002, C123, C001, C105, C082, C060)*

This paper does not rank vendors. It follows fourteen output routes through public documentation as of 9 October 2026 and records what that documentation shows. Where it is silent, the paper says so and names where it looked. It read published policies where they were linked from the documentation, but no private contract, and it tested no live account, so it cannot say what any buyer’s own terms allow. *(inference)*

> **Before signing.** Name the result you need. Say who will hold it and which task it must support after exit. Ask for three things: the documented rule, the term of use and a test. An export option alone cannot settle all three.

> **How to read the grades.** Every paragraph ends with its evidence grade. Fact: a primary document establishes the event, rule or interface. Attributed: a named party states a plan or position. Vendor-claimed: a vendor’s untested description of how something works. Reported: secondary evidence. Inference: the author’s reading, with its premises in the ledger. A schedule set by a vendor is reported as that vendor’s schedule.

## 2. Four checks for each output

*Receive, retain, reuse and rebuild are four checks on one output. They are not a score for a whole product.*

**In short.** Follow the same object through each check. Documentation shows how outputs are received and some technical rules for keeping and using them. Whether a buyer may keep or reuse them after exit is not settled by the reviewed sources. Whether a task can continue elsewhere was not assessed, because that needs a named task and target environment.

Each check needs its own evidence. If a report can leave but an audience must stay inside an ad system, they get separate rows. Where a source does not state a right, the answer stays open rather than being guessed. *(inference)*

**The four checks and the evidence each needs**  
*Column classifications: inference · inference · inference*

| Check | What it means | Evidence that settles it |
|---|---|---|
| Receive | A receiving mechanism for the named output is documented for the agreed recipient. | A documented mechanism; in a purchase, also an account grant, the destination and a test receipt. |
| Retain | The recipient can keep that output for the agreed period, including after the exit event. | Storage behaviour, expiry and deletion rules, and the applicable retention terms. |
| Reuse | The recipient may use that output for a named later task. | Purpose and licence terms, the permissions needed and a dependency check. |
| Continuity | Another agreed environment can perform a named task with assets the buyer may use. | A named task and environment, an asset list, the dependencies, a successful replay and a stated tolerance. |

> **Figure — Four checks. One output.** *(observed)*  
> Each row is one output route; each column asks one question of it: is there a receiving route, is the output kept through exit, is a later task allowed, and does the task run elsewhere. Public documentation and published policies as of 9 October 2026, observed in documents, not tested in accounts. A cell names the documented technical rule (“Rule”) or published terms (“Terms”) where there are any. “Not established” means the reviewed sources do not settle the question, neither granted nor denied. Continuity was not assessed, because no task and target environment were defined. Not a ranking.  
> Text equivalent: 14 output routes against four checks, observed in public documents. Receiving route: 8 documented and 6 documented with a condition (Snowflake activation to an approved account: recipient must import it; Google Ads Data Hub audience: to linked Google ad accounts; InfoSum collaboration event log: visible while a member; LiveRamp Clean Room owner-approved output: needs the data owner’s express permission; Amazon Marketing Cloud audience: into Amazon Ads, not to the buyer; Meta Advanced Analytics partner-route query data: partner route; needs FAA and CAPI). Kept through exit: not established for all 14; the documented rules and terms are Snowflake analysis result: public terms defer to the customer’s contract; Snowflake activation to an approved account: public terms defer to the customer’s contract; Databricks runner-only output table: source ends at 30 days or room deletion; copying documented; AWS Clean Rooms list-analysis result: source access may end; exported-result retention unstated; Google Ads Data Hub audience: after-exit availability not stated; BigQuery clean rooms query result: contributor may block copying; InfoSum collaboration event log: a member loses access on leaving; owners close; Amazon Marketing Cloud aggregate result: lands in the advertiser’s storage; Amazon Marketing Cloud audience: after-exit availability not stated; Meta Advanced Analytics partner-route query data: 90-day availability in the route. Later task allowed: not established for all 14; the documented rules and terms are Snowflake activation to an approved account: the spec sets the destinations; AWS Clean Rooms list-analysis result: identification needs contributor permission; AWS Clean Rooms trained model artifact: weights can be exported; licence not reviewed; Google Ads Data Hub aggregate result: policy limits some uses; joins in BigQuery documented; Google Ads Data Hub audience: no uses beyond approved activation; BigQuery clean rooms query result: export limits set per listing; InfoSum activation export: permission names the destination; LiveRamp Clean Room owner-approved output: grant scope and duration not stated; Amazon Marketing Cloud aggregate result: restricted sharing; named exceptions; Amazon Marketing Cloud audience: used inside Amazon Ads. Task runs elsewhere: Continuity was not assessed for any route: it needs a named task and a target environment, and none was defined. AWS documents a model-artifact export; replaying it was not assessed.  
> Units: Evidence value per cell; no score. Scope: Fourteen output routes across nine product boundaries; public documentation, cutoff 9 October 2026. Editable SVG: `figures/ladder.svg` [6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26]

Continuity means one named task in a named environment, not a copy of a supplier’s system. Reading a retained report, recomputing it, refreshing it with new data and scoring new records with an exported model are different tasks with different needs. The reviewed sources document output mechanisms, but this study did not establish, for each route, the assets, permissions and dependencies needed to continue a specified task elsewhere. AWS documents a model-artifact export; replaying that model would also need permitted scoring inputs, the same transforms, a compatible runtime, applicable terms and a declared tolerance, and was not assessed. *(inference; claims C103)*

## 3. What each route lets out

*Some approved outputs contain rows, some hold totals, and some can be used only at a named destination.*

**In short.** Row-level lists, identifier exports and trained models leave under named permissions; audiences often stay inside an ad system. None of these routes grants access to a partner’s full source table. Ask which object crosses which boundary, and who grants that right.

At the open end, an AWS Clean Rooms list analysis rule “outputs row-level lists of the overlap” between collaborators’ tables [10]. Snowflake’s clean-room agent supports a template that builds a list of overlapping records with their attributes and exports it to an approved account [7]; ordinary Snowflake analysis results are returned in the workspace [6]. InfoSum’s API carries a permission that lets operators activate export columns to authorised destinations [17, 16]. *(fact; claims C024, C012, C109, C056, C055)*

At the closed end, Ads Data Hub audiences go only to linked Google Ads and Display & Video 360 accounts [14], and Google’s published Ads Data Hub Policies limit some uses of its data, such as building profiles or audiences beyond approved activation [21]. Google also documents joining the results of separate Ads Data Hub queries in BigQuery [22]. Amazon Marketing Cloud returns aggregate results to the advertiser’s S3 bucket, while its audiences go into Amazon Ads [19]. *(fact; claims C038, C049, C124, C125)*

**Ads Data Hub privacy thresholds by mode and data (Google, page updated 25 September 2026)**  
*Column classifications: fact · fact · inference*

| Mode or data | Approximate unique users per result row | Note |
|---|---|---|
| Noise injection | About 20 [13] | A threshold is not a promise that every query above it succeeds. |
| Difference checks | About 50 [13] | Difference checks also compare results across queries. |
| Queries using only click and conversion data | About 10 [13] | An exception within difference-check mode. |
Sources: [13]

**What the documentation lets each route send, and to whom (as of 9 October 2026)**  
*Column classifications: fact · fact · fact · fact*

| Product and route | What can leave | Who receives it | Condition that matters |
|---|---|---|---|
| Snowflake: analysis | Analysis results [6] | The collaborator, in the workspace [6] | Keeping a copy after exit is a terms question |
| Snowflake: activation | Overlapping records with attributes [7] | An approved account, which must import them [6] | Destinations are declared in the collaboration spec [8] |
| Databricks Clean Rooms | Read-only output tables [9] | By default only the notebook runner [9] | Kept 30 days, or until the room is deleted, unless copied [9] |
| AWS Clean Rooms | Row-level overlap lists, trained models, inference output [10, 11] | The member given the matching ability [11] | Logs do not prove delivery [27] |
| Google Ads Data Hub | Privacy-checked aggregates; audiences [13, 14] | The analyst; linked Google Ads and DV360 accounts [14] | Thresholds by mode (table above) |
| BigQuery clean rooms | Query results [15] | Subscribers [15] | Contributors can block copying and export [15] |
| InfoSum | Permissioned identifier exports; aggregates [16, 17] | Authorised destinations [16] | A member that leaves loses the event logs [17] |
| LiveRamp Clean Room | Analytics and activation outputs [18] | As the data owner permits [18] | Each use needs the owner’s express permission [18] |
| Amazon Marketing Cloud | Aggregate results; audiences [19] | Advertiser’s S3 bucket; Amazon Ads [19] | Audiences are used inside Amazon Ads [19] |
| Meta Advanced Analytics (LiveRamp route) | Query data in a configured instance [20] | The partner’s configured instance [20] | Queried data available for 90 days in that route [20] |

“Only aggregates leave” does not describe every reviewed route: some permit approved row-level outputs or model artifacts. That does not imply permission to extract a partner’s unrestricted source table. The useful request is a named list of output objects, each with its recipient, its grant and its retention terms. *(inference)*

## 4. Persistence and exit

*Small documented rules decide what a buyer still holds when access ends.*

**In short.** Outputs can expire early, a model can leave while its training data stays, an activation needs the recipient to act, and a member that leaves can lose its view of the record. Published terms limit some uses, but none reviewed says how long an output may be kept after exit. Test the event that ends access, not just the normal path.

### Outputs can end before their window

Databricks describes clean-room output tables as temporary and read-only, kept for 30 days unless copied to local storage [9]. “Deleting a clean room removes all output tables and historical data from the output catalog” [9]. By default only the collaborator who runs the notebook can read an output; sharing outputs with all collaborators is a Public Preview option [9]. *(fact; claims C016, C017, C107, C108)*

### A model can leave; its inputs stay

AWS lets a collaboration assign one member the ability to receive trained model exports and, separately, inference output [11]. Since 30 November 2025 AWS has offered synthetic training data in Clean Rooms ML, and it says customers can export the trained model weights [12]. Amazon Marketing Cloud’s Custom Models, announced on 31 October 2025, let customers use their own algorithms to build audiences [26]. *(fact; claims C025, C026, C029, C030, C076)*

An exported model is a transfer mechanism, not a licence. Running it elsewhere needs permitted inputs, the same preprocessing, compatible software and a tolerance agreed in advance. Retraining it needs data the buyer may not be allowed to take. *(inference)*

### Activation needs the recipient

A Snowflake collaboration declares where results may be sent [8]. Ordinary analysis results are returned in the workspace. For collaborator activation, “the receiving collaborator must call VIEW_ACTIVATIONS() and PROCESS_ACTIVATION() to import the results” [6]. A completed activation run is not proof that the recipient imported it. *(fact; claims C010, C013, C109)*

### Leaving is not the same as deletion

InfoSum documents two exits. An owner closes a collaboration; that is irreversible and stops new queries or exports, while previous activity can still be seen. A member that leaves can no longer see the collaboration’s details, including the event logs; an owner must close rather than leave [17]. *(fact; claims C057, C075)*

### Published terms settle some of it

Some limits are public. Google’s Ads Data Hub Policies bar using Google’s data to build or augment profiles or audiences beyond the activation channels Ads Data Hub approves [21]. AWS’s Service Terms bar using Clean Rooms information to identify a person unless the data’s contributor permits it, and say deleting a collaboration ends access to the customer’s content or deletes it from intermediate tables; that does not recall a result already delivered [23]. The Amazon Marketing Cloud Agreement makes query results confidential and restricts sharing, with named exceptions for the relevant agency, the advertiser and service providers acting for the advertiser, all bound to confidentiality [24]. Snowflake’s public clean-room terms defer to the customer’s own agreement [25]. *(fact; claims C124, C126, C127, C128, C129)*

For Databricks, InfoSum, LiveRamp and Meta Advanced Analytics, no public product-specific terms on outputs, retention or post-termination use were found in a bounded search. None of the published terms reviewed states how long a buyer may keep an output after exit, so that answer stays not established. *(inference)*

Keep a record of each result you will need after exit: its owner, location, expiry rule and terms of use. Test the event that ends access, such as room deletion or leaving, as well as the normal expiry. A copy route does not by itself give the right to keep or reuse the copy. *(inference)*

## 5. Proof must cross the boundary

*A job can finish while a transfer fails. A deletion record has a scope.*

**In short.** Each step of an output’s journey needs its own record, joined by one identifier. Platforms document runs and outputs; some document scoped deletion events. No single record shows what every copy became.

AWS is explicit: query and job logs show a query’s status but “don’t report whether query output was delivered” [27]. Since 11 August 2026 a member granted a specific ability can export those logs, once every member approves the change [28]. Snowflake’s Snowsight collaboration view has an Activity tab that records analysis template runs [29]. Its agent observability redacts orchestrator-authored content from consumer records by default [30]. *(fact; claims C031, C028, C119, C004)*

> **Figure — What each record proves** *(illustrative)*  
> Each mark is one documented record, placed on the one step of an output’s journey it covers; records are never joined into a trace. The six-step chain across the top is the author’s illustrative method, not a workflow the five products share. The warm line marks where an output leaves the platform. A record of an operation does not establish the fate of every downstream copy.  
> Text equivalent: Proposed method, illustrative: the job runs; an output exists; recipient imports; destination accepts; a copy is kept; a copy is deleted. An output leaves the platform after “an output exists”. AWS Clean Rooms: query and job logs (the job runs): show a query’s status; do not report delivery; model-output ability (an output exists): a named member may receive model exports; CloudTrail data events (a copy is deleted): record S3 object deletions, if configured; versions may remain. Snowflake: Activity tab (the job runs): records analysis template runs; workspace results (an output exists): analysis results returned in the workspace; activation import (recipient imports): activation route only: the recipient must import. Databricks: output table (an output exists): read-only; the runner reads it by default; table lifetime (a copy is kept): 30 days, or until the room is deleted; a copy is outside this rule. BigQuery: query results (an output exists): for subscribers to a listing; copy and export controls (a copy is kept): a contributor may block copying; table-deletion reason (a copy is deleted): one table: a request, expiry or a query. InfoSum: event logs (the job runs): visible while a member; a leaver loses them; destination export (an output exists): to an authorised destination; closing stops new exports. No record shown covers a destination accepting an output. The records shown cover specific operations and objects. A record of an operation does not establish the fate of every downstream copy.  
> Units: Steps in an evidence chain; no timing. Scope: Proposed method with documented markers for five products. Editable SVG: `figures/proof-chain.svg` [27, 11, 31, 29, 6, 9, 15, 32, 17, 16, 33]

Deletion records exist, with limits. AWS CloudTrail can record S3 object deletions as data events, which are off by default and charged [31]. In a versioning-enabled S3 bucket, a simple delete inserts a delete marker instead of removing the object [33]. BigQuery audit metadata records how a table was deleted [32]. *(fact; claims C110, C111, C112)*

Withdrawal has the same shape. Closing an InfoSum collaboration stops new exports [17]. When an AWS table’s data access budget reaches zero, it can no longer be used in queries, jobs or ML inputs [34]. Google’s Data Manager API can remove every member of an audience [35]. *(fact; claims C057, C032, C042)*

These controls stop new use; the cited pages do not establish deletion of a copy already delivered. So start with one output and one route. Link the run identifier to the output object, the transfer and the recipient’s response, then record what was kept and what was removed. The contract should name who supplies each record and for how long. A hash can show that two files match; it cannot show that the buyer may use either. *(inference)*

## 6. Agent routes, scoped

*An agent is another way to invoke an output route. Its scope and its stage matter more than its label.*

**In short.** Three general surfaces carry a vendor-stated GA label. The advertising routes reviewed are in preview, in closed beta or carry no stated stage; no native InfoSum MCP server and no Ads Data Hub MCP endpoint were found. A route with no stated stage is recorded as such, not as unavailable. A shared protocol carries a request; it does not replace the grants that make it valid.

Three general surfaces carry a GA label: Cortex Agents and MCP servers in Snowflake Native Apps since 7 August 2026 [36], BigQuery’s MCP server since 20 April 2026 [37], and Databricks’ Genie One MCP server since 25 September 2026 [3]. Databricks schedules the former Genie endpoint, /api/2.0/mcp/genie, to sunset on 31 October 2026 [3]. *(fact; claims C006, C086, C099, C105)*

Snowflake’s clean-room agent is an open preview [7], and its AI Gateway for Advertising was announced in preview on 5 October 2026 [38]. Databricks lists Meta’s ads MCP server — Meta’s campaign tools, not Meta Advanced Analytics — in its Marketplace with no stage stated [39]. Amazon’s AMC SQL Generator entered closed beta on 30 September 2026; during the beta it is available in the United States, and Amazon says its agents serve North America, Europe and Far East marketplaces from US-hosted infrastructure [40]. AWS documents an aws-cleanrooms skill that troubleshoots permission errors and missing ML-job logs [41]; Clean Rooms analytics and export through agents were not evaluated here. LiveRamp said in March 2026 it was selectively enabling a limited number of builders to deploy agents [42]. *(fact; claims C011, C005, C020, C047, C096, C116, C106, C113)*

Amazon says the expert tools will reach general availability in early 2027, priced per use through AWS Marketplace, with broader geography before then [40]. That is Amazon’s plan. *(attributed; claims C048, C117)*

> **Figure — Agent routes, their actions and stated stages** *(observed)*  
> Three separate groups: routes whose vendor states a release stage, routes with no stage stated, and named interfaces not found in the recorded search. Labels are listed, not ranked; an unstated stage is not counted as preview or as unavailable. A release label belongs to the interface, not to the product beneath it, and no route grants the right to activate, spend, export or delete.  
> Text equivalent: Release stage stated by the vendor: Snowflake, Native Apps: Cortex Agents and MCP servers, generally available (7 Aug 2026): agents and MCP servers run inside installed apps; a general platform feature, not clean-room specific. Google BigQuery, BigQuery MCP server, generally available (20 Apr 2026): runs queries, reads metadata, lists resources; does not bypass a clean room’s egress rules. Databricks, Genie One MCP server, generally available (25 Sep 2026): answers natural-language questions about data; watch: the old /api/2.0/mcp/genie endpoint is scheduled to sunset on 31 Oct 2026. Snowflake, Clean-room agent, open preview (date not stated): runs approved overlap and overlap-activation templates; can export overlap records to an approved account. Snowflake, AI Gateway for Advertising, preview (5 Oct 2026): governed MCP routes to Meta and TikTok; architecture is vendor-described. Amazon Ads, AMC SQL Generator on the Ads MCP Server, closed beta (30 Sep 2026): writes Amazon Marketing Cloud SQL; US during the beta; agents serve other markets from US-hosted infrastructure; GA planned for early 2027. No release stage stated: Meta, via Databricks, Meta ads MCP server (Marketplace), no stage stated (6 Oct 2026): Meta campaign tools — not Meta Advanced Analytics; listed in Databricks Marketplace. AWS Clean Rooms, aws-cleanrooms skill (AWS MCP), no stage stated (date not stated): troubleshoots permission errors and missing ML-job logs; analytics and export through agents not evaluated. LiveRamp, Agent deployment in its marketplace, no stage stated (Mar 2026): builders deploy agents; selective: a limited number of builders; not the segment builder. No surface found in the recorded search: InfoSum, Native MCP server: API v2 is documented; searched for a native MCP server. Google Ads Data Hub, MCP endpoint: SQL interface is documented; searched for an ADH MCP endpoint.  
> Units: Vendor-stated release stage. Scope: Agent and MCP routes into the reviewed products, cutoff 9 October 2026; not a market survey. Editable SVG: `figures/agent-stages.svg` [36, 37, 3, 7, 38, 40, 39, 41, 42, 17, 43]

MCP and A2A define how a call is authorised. AAMP is IAB Tech Lab’s umbrella framework rather than a transport; AAMP 3.0 introduced the OpenProposal draft, with public comment scheduled to end on 22 October 2026. None of them, by itself, grants advertising permission, records consent or makes an output exportable. Treat an agent route like any other route: name what it may read and write, the account it acts for, how it is revoked and what happens to queued work. *(inference; claims C104, C070, C071, C088)*

## 7. Dates and owners

*Dates matter when a task depends on a route that will close. A change of owner changes who the buyer relies on.*

**In short.** WPP owns InfoSum. Publicis has agreed to buy LiveRamp, but the deal was pending at the cutoff. Databricks and Snowflake have scheduled route retirements between October 2026 and June 2027. Product schedules may change; transaction deadlines and extensions follow the agreement’s stated terms.

> **Figure — What happened, and what is scheduled** *(observed)*  
> Read down. Above the evidence cutoff of 9 October 2026 are events that happened; below it are dates set by named parties, which may move. Status words are the parties’ own terms. Grouped by year; not to scale. Transaction deadlines and extensions follow the agreement’s stated terms.  
> Text equivalent: Before the cutoff, events that happened: 31 January 2024 (ownership), completed: LiveRamp completes the Habu acquisition; 31 July 2024 (platforms), released (GA): AMC on AWS Clean Rooms generally available; 3 April 2025 (ownership), announced: WPP announces its acquisition of InfoSum; 30 November 2025 (platforms), released: AWS Clean Rooms ML adds synthetic training data; 20 April 2026 (agents), released (GA): BigQuery MCP server generally available; 17 May 2026 (ownership), announced: Publicis agrees to buy LiveRamp; 11 June 2026 (ownership), filed: Publicis and LiveRamp file US premerger notices; 16 June 2026 (platforms), private preview: Databricks CustomerLake announced in private preview; 7 August 2026 (agents), released (GA): Snowflake Native App agents and MCP servers; 17 August 2026 (ownership), approved: LiveRamp stockholders adopt the merger agreement; 20 August 2026 (platforms), completed: Snowflake removes legacy connectors, round 1; 10 September 2026 (platforms), completed: Snowflake removes legacy connectors, round 2; 16 September 2026 (ownership), pending: Publicis: the deal remains subject to customary approvals; 22 September 2026 (agents), public comment: AAMP 3.0 introduces the OpenProposal draft; 24 September 2026 (platforms), released (GA): Snowsight collaboration management; 30 September 2026 (agents), closed beta: AMC SQL Generator on the Ads MCP Server (US); 1 October 2026 (platforms), completed: No new legacy Snowflake rooms in the web interface; 5 October 2026 (agents), preview: Snowflake AI Gateway for Advertising announced; 6 October 2026 (agents), listed: Meta ads MCP listed on Databricks Marketplace. After the cutoff, dates set by named parties: 31 October 2026 (agents), scheduled: Databricks: old /api/2.0/mcp/genie endpoint sunsets; By end 2026 (ownership), expected: Publicis expects the LiveRamp deal to close; 31 December 2026 (platforms), reported end: AMC first-party paid-feature offer (trade report); 1 February 2027 (platforms), scheduled: Snowflake: legacy interface closes; legacy-API room creation stops; Early 2027 (agents), planned: Amazon plans expert-tool GA, priced per use; 16 May 2027 (ownership), contract term: Merger outside date (can extend to 16 Aug 2027); 1 June 2027 (platforms), scheduled: Snowflake: access to existing legacy rooms ends.  
> Units: Dates. Scope: Events the paper relies on, January 2024 to June 2027. Editable SVG: `figures/timeline.svg` [44, 45, 4, 46, 47, 5, 19, 12, 48, 49, 50, 29, 1, 51, 2, 37, 36, 52, 40, 38, 39, 3, 53]

WPP announced its acquisition of InfoSum on 3 April 2025 [45]. LiveRamp completed its acquisition of the clean-room company Habu on 31 January 2024 [44]. *(fact; claims C053, C064)*

Publicis agreed on 17 May 2026 to buy LiveRamp for $38.50 a share in cash, a total enterprise value of $2.167 billion [4, 54]. LiveRamp’s stockholders adopted the agreement on 17 August 2026 [47]. The parties filed US premerger notifications on 11 June 2026, and closing also needs CFIUS approval [46]. The outside date is 16 May 2027, extended automatically to 16 August 2027 if only certain regulatory conditions remain [46]. On 16 September 2026 Publicis said the deal “remains subject to customary approvals” [5]. *(fact; claims C082, C078, C061, C090, C091, C092, C060)*

Publicis expects the deal to close before the end of 2026 [5]. LiveRamp told employees it is committed to “an independence charter focused on open access, privacy and pricing controls” [54]. *(attributed; claims C093, C079)*

No closing filing or notice was found in the sources searched through 9 October 2026, so this paper treats LiveRamp as pending. A pending owner is a reason to write access, pricing and data-use terms into the contract. It is not proof that they will change. *(inference; claims C094)*

Snowflake has already removed legacy web-interface connectors in two rounds. On 20 August 2026 it removed the LiveRamp, Acxiom, Merkury and PAIR identity connectors [49]. On 10 September 2026 it removed further identity connectors, including Experian, Acxiom Real ID and The Trade Desk UID 2.0, and activation connectors including Display & Video 360 and LiveRamp; existing configurations were not deleted [50]. *(fact; claims C008, C083, C118, C084)*

PPC Land reported on 2 June 2026 that Amazon made first-party paid-feature signals in AMC free to query through 31 December 2026, while third-party signals such as Experian’s still need a subscription [51]. Amazon’s own page could not be read [53]. *(reported; claims C052)*

## 8. The continuity trade-off

*Keeping work inside the room can be the right choice. The question is what the buyer must still be able to do.*

**In short.** A buyer may need a sound answer, a record of how it was reached and the right to use what it learned — not a copy of every input. Export limits can serve that need while protecting a partner. The case weakens only when the next task needs an asset or grant the buyer cannot keep.

The restrictions are deliberate, and they depend on configuration. Ads Data Hub’s thresholds keep results from describing small groups [13]. BigQuery clean-room contributors can stop subscribers copying or exporting results, and Google warns that analysis rules “might not fully prevent sophisticated queries from extracting unauthorized data”, pointing to approved query templates [15]. Google’s PAIR design returns match rates, not user-level matches [55]. *(fact; claims C036, C035, C114, C044)*

The strongest counter-argument to this paper is that useful learning can survive without a full export. A permitted report, a tested estimate, a frozen model or a written method may be enough for the next decision. Continued governed access can also be a dependency a buyer accepts on purpose, if its terms and price are clear. A closed model can be a sound purchase; an exported one can be a poor exit asset if its inputs, licence or runtime cannot travel. *(inference)*

So test the trade-off directly. Can the buyer still make the next decision? Can it check how the result was reached? Can it refresh the work when needed, and at what cost? A full export is worth having only if it helps answer yes on terms the buyer may use. *(inference)*

## 9. Tools for the buying team

*An output record, a minimum useful exit package, a cost-and-continuity worksheet and thirteen tests. All are proposed; none was run for this paper.*

**In short.** Start with the task to continue, list only the assets it needs, price the exit and the alternative, then agree thirteen tests before signing and run them before full commitment. Preserve every negative-test result: an unexpected success, or a failure for an unrelated reason, needs investigation.

### The output record

Procurement, data engineering and the business owner complete one record per output: the business task and the result it needs; product, route and version; input and output schema; data owner; the account the work runs as; the authorised recipient; destination owner and region; permitted purpose; retention start and end; exit events; copy and deletion terms; code and runtime dependencies; acceptance evidence; and who signs off. Mark which assets the buyer supplied, which a partner supplied and which are joint or derived: one output can depend on several parties’ permissions. *(inference)*

### The minimum useful exit package

**Proposed inventories by output type (no platform is asserted to export every item)**  
*Column classifications: inference · inference*

| Output | What the next task usually needs |
|---|---|
| Report or analysis | The result, query version, input period, metric definitions and uncertainty notes. |
| Model | Permitted artifacts, feature transforms, package versions, evaluation inputs and licence terms. |
| Agent workflow | The approved policy, tool schemas, configurations, permitted memory and action records. |

Classify each dependency as transferable, replaceable, available only through continued access, prohibited from transfer, or unresolved. That gives the buyer a transition plan instead of a single portability colour. *(inference)*

### The cost and continuity worksheet

Ask bidders to quote the same task and the same exit scenario: setup, compute, storage, transfer, licences, labour, running two systems in parallel, revalidation and downtime, with who pays each and whether it is a quote or an estimate. Compare the cost of the exit package with the cost and risk of continued access. Do not build a cross-vendor total from incompatible price units. *(inference)*

### Thirteen tests

**Proposed buyer acceptance tests, each with a deliberate failure case (none performed for this paper)**  
*Column classifications: inference · inference · inference*

| Test | What to do, including the failure case | Evidence that passes |
|---|---|---|
| T01 Output boundary | Request the approved synthetic output, then a disallowed field or row under the same role. | The expected schema arrives; the disallowed request fails for the expected policy reason, not a generic error. |
| T02 Persistence and exit | Copy a permitted result; delete a disposable room or end access; check normal expiry separately. | The named copy stays readable for the agreed period, and its use is allowed by the terms. |
| T03 Inference replay | Load the permitted model elsewhere with fixed preprocessing and held-out inputs; omit one dependency on purpose. | The declared tolerance is met and the missing dependency is visible. Retraining is a separate test. |
| T04 Delivery chain | Link run, object, transfer and destination records through a shared correlation identifier or an auditable mapping; deny one test destination. | Success and failure are distinguishable at each boundary. |
| T05 Revocation | Agree permitted outcomes and maximum delays for new, queued and running work before testing; then revoke a test grant in each state. | Observed behaviour meets the agreed outcomes and delays; residual actions are recorded separately. |
| T06 Audit access | Save the permitted record before exit; test owner and member exits separately. | The buyer keeps the agreed evidence and can read it after access ends. |
| T07 Purpose change | Ask to use the same object for a second, clearly different purpose. | Terms and controls say whether it is allowed; technical availability alone is not a pass. |
| T08 Destination lifecycle | Delete the test object; inspect versions, replicas and backup exceptions in scope. | Deletion records and remaining copies are reconciled; nothing is claimed about unobserved stores. |
| T09 Audience readiness | Track upload acceptance, matching, audience readiness and a paused campaign. | Each state is recorded; upload success is not reported as delivery or impact. |
| T10 Agent identity | Compare the tool’s acting account and grants with the intended user; try a disallowed read and write. | The effective account and denial records match the approved policy. |
| T11 Route migration | Move one workflow off a retiring route (for example the old Genie endpoint or a legacy Snowflake room). | The task still works within tolerance; changed behaviour is recorded. |
| T12 Task rebuild | Run the named task in a second environment using only the agreed exit package. | Quality, freshness, cost and latency meet thresholds set before the test. |
| T13 Commercial exit drill | Request the actual exit schedule, fees, data-use terms and supplier duties for the route. | Named parties accept the deliverables, dates, exceptions and acceptance evidence. |

Use test data and accounts cleared for that use, and record the expected result before running. Synthetic data tests controls and transport; it does not show production match quality, privacy loss or campaign effect. *(inference)*

## 10. What would change a finding

*A finding changes when new proof fills a gap. A new release label alone is not enough.*

**In short.** Each trigger is tied to the row it could change. The search record lists, for each product boundary, what was sought and which documentation and published policies were read.

The task-level test that would change the rebuild column: a documented, permitted transfer that lets another environment perform the same named task within an agreed tolerance. If a buyer can show that, the finding for that route changes, with the source and date recorded. *(inference)*

**Triggers, the row each could change and the evidence needed**  
*Column classifications: inference · attributed · inference · inference*

| Trigger | When | Row it could change | Evidence needed |
|---|---|---|---|
| Genie endpoint sunset | Scheduled by Databricks for 31 October 2026 [3] | Agent routes; test T11 | The replacement service carries the same grants and outputs. |
| Snowflake legacy interface closes; creation of new rooms through the legacy API stops | Scheduled by Snowflake for 1 February 2027 [2] | Snowflake activation | The template and activation route work through the Collaboration API. |
| Access to existing legacy Snowflake rooms ends | Scheduled by Snowflake for 1 June 2027 [2] | Snowflake rows | Dependent routes migrated, or stopped. |
| Publicis–LiveRamp closes, is amended or ends | Publicis expects before the end of 2026; outside date 16 May 2027 [5, 46] | Ownership only | A closing filing and the terms offered after it. |
| AMC first-party paid-feature offer renewed or ends | Reported to end 31 December 2026 [51] | AMC cost | Amazon’s readable terms. |
| AMC expert tools reach GA | Amazon plans early 2027 [40] | AMC agent route | Released scope, geography and pricing. |
| A documented route for a named task | No date announced | The matching continuity cell only | A named task and environment, a documented and permitted transfer, and a replay test. |

### Where we looked

**Search record: documentation and published terms read for each boundary (9 October 2026; about two searches per product for public terms)**  
*Column classifications: inference · inference · inference · inference*

| Boundary | Assets sought | Documentation searched | Blocked or unresolved |
|---|---|---|---|
| Snowflake | Copy, retention and re-run routes for analysis and activation outputs | Collaboration spec, clean-room agent, run-analysis and end-of-life pages; release notes Aug–Oct 2026 [8, 7, 6, 2]; public Data Clean Rooms terms [25] | Public terms defer to the customer agreement; post-exit terms not established from the listed sources |
| Databricks Clean Rooms | Output retention, sharing and export of notebooks or state | Clean Rooms and output-table pages; MLflow; release notes Sep 2026 [56, 9, 57, 3] | No public product-specific terms found; copied-result terms not established from the listed sources |
| AWS Clean Rooms | Result and model export, logs, deletion | Analysis rules, custom-model prerequisites, logging, budgets, troubleshooting [10, 11, 27, 34, 41]; AWS Service Terms section 92 [23] | Published use and deletion terms found; retention after exit not established from the listed sources |
| Google Ads Data Hub | Export and reuse of aggregates and audiences | Privacy checks, audience activation, difference checks and the published Ads Data Hub Policies [13, 14, 22, 21] | Published policy restrictions found; buyer-specific post-exit terms not established |
| BigQuery clean rooms | Copy and export controls, deletion records | Data clean rooms, analysis rules, audit metadata [15, 58, 32] | Subscriber terms set per listing |
| InfoSum | Export permissions, exit and logs | API v2 and activation help pages [17, 16] | No public product-specific terms found; contract terms not established from the listed sources |
| LiveRamp Clean Room | Output permissions and retention | Privacy and governance controls; March 2026 release notes [18, 42] | No public product-specific terms found; retention and exit terms not established from the listed sources |
| Amazon Marketing Cloud | Result retention, audience export, model weights | AMC on AWS Clean Rooms; Custom Models; paid features [19, 26, 53]; Amazon Marketing Cloud Agreement [24] | Agreement found (no retention period); paid-features page unreadable |
| Meta Advanced Analytics | Export schema and retention | LiveRamp configuration guide; Tobiras announcement [20, 59] | No public product-specific terms found; no public Meta contract; Meta help pages behind login |

## 11. Corrections

*This site publishes guides on the same products. Checking them found errors there, in the dossier this paper started from, and — after review — in the first draft of this paper.*

**In short.** 46 rows: 14 on the starting dossier, 14 on this site’s guides and 18 on the first draft. Each states its type, evidence and status.

No Fluff Advisory owns the guides checked here. The starting dossier, produced with an AI deep-research tool, was checked against the guides’ source code and the cited pages. Of its 171 audit rows, 156 matched what the guides say, 7 were recommendations with no guide claim behind them and 8 attributed statements the guides do not make. The guide corrections were merged to the site on 9 October 2026. [60, 61] *(inference)*

A critical review of the first draft then found that it had treated evidence of export and storage as evidence of rights after exit, compared different assets within one figure row, used rebuild in three senses and described some unknown release stages as not available. An earlier edition was published on 9 October 2026 and withdrawn the same day; a second review then asked for the figure to apply its own definitions, for task continuity to be reported as not assessed, and for published use policies to be read. Those changes, and two earlier corrections that this ledger had overstated, are listed below. *(inference)*

**Corrections ledger** — 46 rows (14 to the dossier, 14 to this site’s guides). The full ledger: `data/corrections-ledger.csv`.

| Id | As first written | Verdict | Change applied | Status |
|---|---|---|---|---|
| D01 | The hub and Snowflake guide display June validation; AMC displays August. | partly supported | Page dates read from source and live pages, not from an older snapshot. | Applied in this paper |
| D02 | A022: the Snowflake guide says only aggregate outputs leave (INCORRECT). | unsupported | Row withdrawn. The overlap-activation fact is kept as C012, sourced to the clean-room agent page. | Applied in this paper |
| D03 | A023: the Snowflake guide lacks versioned connector warnings. | partly supported | Reframed as an omission of the 10 September round (C083, C084). | Applied in this paper; guide fixed (G04) |
| D04 | A044, A129, A134, A161 attribute statements the guides do not make. | unsupported | Rows withdrawn; A044 becomes an addition, not an update. | Applied in this paper |
| D05 | A015 cited the 1 October 2026 release note (SF02) together with the end-of-life page (SF01) for the 2027 milestones. | partly supported | Recorded as a citation-precision note on audit row A015, not a false-source correction. The 2027 milestones stay cited to SNW-S001. | Applied in this paper |
| D06 | A022 cited the clean-room agent page (SF10) and the run-analysis page (SF11) together for overlap activation. | partly supported | Recorded as a citation-precision note, not a false-source correction. | Applied in this paper |
| D07 | C052 graded FACT: AMC paid-feature no-cost period ends 2026-12-31. | unverified | Regraded REPORTED, verdict unverified, narrowed to first-party paid features. | Applied in this paper |
| D08 | C067 graded FACT: web advertisers no longer need to prioritize eight events per domain. | unverified | Regraded REPORTED, verdict unverified; not used in the paper's argument. | Applied in this paper |
| D09 | C005, C020, C054, C066 graded VENDOR-CLAIMED. | partly supported | Regraded FACT for the dated announcement; the capabilities described stay vendor-claimed in the text and limitations. | Applied in this paper |
| D10 | Register titles and URLs for SF04, META01, LR05, LR07. | partly supported | Titles corrected; LR07 re-pointed to SEC EDGAR; the Habu claim re-sourced to LiveRamp's 10-K (LVR-S011). | Applied in this paper |
| D11 | C034: 'Quick Suite' as the former name of Amazon Quick, sourced to a load-testing solution page. | partly supported | Added AWS-S010 and the QuickSight framing (C100). | Applied in this paper; guide fixed (G08) |
| D12 | C068 limitation: the Meta announcement was access-blocked. | partly supported | Limitation updated with the rollout wording. | Applied in this paper |
| D13 | Timeline and agent-route rows omit events from 10 September to 6 October 2026 and earlier GA stages. | supported | Added as C083–C102 with primary sources. | Applied in this paper |
| D14 | A040 graded UNVERIFIABLE; A081 says only that BigQuery documents an MCP server; A124 (graded OUTDATED) recommended renaming Ads Agent without stating each route’s stage. | partly supported | A040 treated as unsupported wording to remove; A081 and A124 completed with the stages the sources state. | Applied in this paper; guides fixed (G06, G09, G11) |
| G01 | LiveRamp guide headings and SEO text read as if Publicis already owns LiveRamp ('LiveRamp after Publicis', 'under Publicis'); one paragraph lists shareholder approval as still outstanding after saying it happened. | unsupported | All ownership wording made pending; the stale shareholder condition removed; the expected year-end close attributed to Publicis. | Merged to the site on 9 October 2026 (pull request 98) |
| G02 | 'the ATS Mobile SDK is fully sunset on 30 September 2026' (future tense after the date). | partly supported | Past tense with the ATS API migration. | Merged to the site on 9 October 2026 (pull request 98) |
| G03 | Snowflake guide does not date the legacy clean-room retirement. | unsupported | Dated migration milestones and the Snowsight replacement added. | Merged to the site on 9 October 2026 (pull request 98) |
| G04 | Connector retirement given as 'August 2026' only; matrix rates Snowflake activation 'Connector-led' without a date. | partly supported | Both rounds dated in the Snowflake guide, the multicloud page and the matrix note. | Merged to the site on 9 October 2026 (pull request 98) |
| G05 | Meta's ads MCP server is 'GA, 16 July 2026' in the Meta guide, 'not a GA feature' in the Snowflake guide and 'developer-available' on the multicloud page. | unsupported | One conservative wording across the three pages, plus the two October routes with their stated stages. | Merged to the site on 9 October 2026 (pull request 98) |
| G06 | 'Agent Bricks (formerly Mosaic AI)'; leftover 'Delta Sharing' after the OpenSharing rename. | unsupported | 'formerly Mosaic AI' removed; OpenSharing used throughout. | Merged to the site on 9 October 2026 (pull request 98) |
| G07 | Matrix rates Databricks multi-party matching 'Partner-led'; guide omits output-table retention. | unsupported | Matrix cell corrected; output-table retention added. | Merged to the site on 9 October 2026 (pull request 98) |
| G08 | 'Amazon Quick Suite, ex-QuickSight'; an AWS footprint 'avoids new vendors, new contracts, and new security reviews'. | unsupported | Name corrected; the overclaim softened to what the documentation supports. Synthetic-data training and model-artifact permissions added. | Merged to the site on 9 October 2026 (pull request 98) |
| G09 | BigQuery clean rooms 'GA in all BigQuery regions'; matrix lists Google agent exposure as 'SQL API only'; Meridian 'not a managed product'. | unsupported | Region limits stated; ADH and BigQuery agent routes separated; Meridian Studio noted. | Merged to the site on 9 October 2026 (pull request 98) |
| G10 | Matrix rates InfoSum activation and segmentation 'Not ideal' and agent exposure 'Not yet'. | unsupported | Opinion labels replaced with what is documented. | Merged to the site on 9 October 2026 (pull request 98) |
| G11 | 'Ads Agent (beta)'; the AMC paid-feature free period stated without scope. | partly supported | Name and stages corrected; the GA plan attributed to Amazon; the free period narrowed and marked as reported. | Merged to the site on 9 October 2026 (pull request 98) |
| G12 | Meta guide: 'event priority and an 8-slot limit'; 'engaged-view'. | unsupported | Terminology updated; the eight-event change marked as reported. | Merged to the site on 9 October 2026 (pull request 98) |
| G13 | Matrix row 'Consent & privacy enforcement' treats access permissions as consent; 'Most … fail because …' prevalence claims stated as fact. | unsupported | Row renamed to access and purpose controls with a note; prevalence claims marked as the author's experience. | Merged to the site on 9 October 2026 (pull request 98) |
| G14 | Hard-coded 'Last reviewed June 2026' label beside September review stamps on four guides. | partly supported | Review stamps updated to the October 2026 re-review. | Merged to the site on 9 October 2026 (pull request 98) |
| R01 | Headline: every product lets something leave; portability is four rights. | partly supported | Receive, retain, reuse and rebuild are now four checks per output; post-exit rights are "not established from the reviewed sources" where no terms were read. | Applied in this edition |
| R02 | Figure 1 compared different assets across one product row (for example AWS overlap lists, then model weights). | partly supported | Figure 1 rebuilt with one row per product and output route, the same object across all four checks, and separate values for documented mechanism, conditional configuration and terms not established. | Applied in this edition |
| R03 | Rebuild meant a useful result, a complete workflow and complete model and agent state in different places. | partly supported | Rebuild now means performing a named task in another environment; inference replay, retraining and full migration are separate tests. C103 narrowed and given a per-boundary search record. | Applied in this edition |
| R04 | "The run alone delivers nothing" (Snowflake). | partly supported | Scoped to activation; ordinary analysis given its own row (C109). | Applied in this edition |
| R05 | Databricks: outputs kept 30 days unless copied. | partly supported | Early-deletion rule and output modes added (C107, C108); test T02 now covers room deletion. | Applied in this edition |
| R06 | No product documents deletion evidence from a recipient. | partly supported | Section 5 retitled "Proof must cross the boundary"; deletion records described with their scope (C110–C112). | Applied in this edition |
| R07 | Advertising agent routes are not generally available. | partly supported | Figure 3 is now categorical; unstated stays unstated (C104 rewritten). | Applied in this edition |
| R08 | Agent routes omitted AWS Clean Rooms operational MCP support and the Databricks Genie endpoint sunset. | partly supported | Both added with their narrow scope (C105, C106); the sunset is in the timeline and test T11. | Applied in this edition |
| R09 | AMC expert tools available in the United States only. | partly supported | Both statements kept and attributed (C096, C116, C117). | Applied in this edition |
| R10 | Every claim carries a verbatim passage; verbatim matching shown as verification. | partly supported | Inference rows now carry premises, reasoning and search scope; compound claims split; thin passages extended or claims narrowed; counts generated from structured fields. | Applied in this edition |
| R11 | Source access statuses inconsistent with their header; stale limitations kept beside successful reads. | partly supported | A single access_status field (read, unavailable, read elsewhere, not reopened) drives every count; dossier and validation access dates stored separately; stale limitations cleared. | Applied in this edition |
| R12 | First draft labelled published 9 October 2026 before release. | partly supported | Edition history now records the publication and withdrawal; this revision is a separate release with its own date. | Applied in this edition |
| R13 | The counter-argument fails once learning must survive a supplier change. | partly supported | Section 8 rewritten around the task the buyer must continue, with an output record and a minimum useful exit package. | Applied in this edition |
| R14 | Figure 1 graded storage locations, availability windows and initial destinations as conditional retention or reuse. | partly supported | Each cell now shows the documented rule separately; retention and reuse are “not established” unless the sources settle them. | Applied in this edition |
| R15 | No documented way to redo the same task elsewhere was found (C103). | partly supported | C103 restated: this study did not establish, per route, what a specified task needs to continue elsewhere. Column renamed Continuity; AWS shows artifact export documented, replay not assessed. | Applied in this edition |
| R16 | Missing answers were said to sit in contracts the study did not read. | partly supported | Both added (C124, C125); search record covers published policies; “terms not public” replaced with “not established from the listed sources”. | Applied in this edition |
| R17 | Several excerpts omitted part of their claims; AWS agent scope, the AAMP comment object and Snowflake dates were imprecise; tests T01, T04 and T05 lacked pass criteria. | partly supported | Excerpts extended or claims narrowed; AWS scope is troubleshooting documented, analytics and export not evaluated; the OpenProposal draft named as the comment object; Snowflake February and June milestones separated; pass criteria added. | Applied in this edition |
| R18 | C128 said AMC bars sharing query results with third parties; C103 kept a not-found conclusion about re-running a task; two AWS labels in Figure 1 dropped their qualifications; Figure 2 said no product documents a destination’s acceptance or every downstream copy. | partly supported | C128 and C103 rewritten; Figure 1 labels restored to their qualifications; Figure 2 narrowed to the records shown; R15’s repair completed in every edition. | Applied in this edition |

## Method, disclosure and downloads

This study reads public sources, including published use policies linked from the documentation, up to 9 October 2026. It began from a research package produced with an AI deep-research tool — 171 audit rows on this site’s guides, 82 claims and 91 sources — which was checked against the guides’ source code and the cited pages, and then revised after a critical review of the first draft. The ledger shows which claims rest on a source and which are the author’s findings. Each sourced claim has a passage, a locator where needed, a link and a check date; each author finding states its premises, reasoning and, for a not-found result, where it searched. A not-found result does not mean the route cannot exist. Release labels state what a vendor says is ready; they do not show what a buyer’s account can do, and a launch post shows that a claim was made on a date, not that it is true. Validation and review added 23 sources and 46 claims. No live account was tested and no private contract was read.

**Interests and disclosure.** The author has worked at Samba TV, a TV measurement and ad data company, since May 2025. He co-leads the Signals & Measurement working group of AgenticAdvertising.org, which publishes AdCP, one of the agent protocols in the background ledger; that organization is an interested source, and the affiliation is disclosed so readers can weigh the protocol discussion. No Fluff Advisory, the author’s own practice, publishes the platform guides checked in section 11.

**Update history**

- **9 October 2026** — First edition published on this site, then withdrawn the same day for revision after a critical review.
- **9 October 2026** — First revision: four checks per output instead of four rights; Figure 1 rebuilt with one output per row; Snowflake activation and Databricks room deletion corrected; scoped deletion records; agent stages kept categorical; the Genie endpoint sunset added; inference rows given premises; source access in one field; two overstated corrections reclassified.
- **9 October 2026** — Second revision after a second review: Figure 1 now separates documented technical rules from what is not established; task continuity reported as not assessed; Google’s published Ads Data Hub Policies and BigQuery join documentation added; excerpt repairs; pass/fail criteria for tests T01, T04 and T05.
- **9 October 2026** — Revised edition released on this site.
- **9 October 2026** — Corrections after a check of the live edition: the AMC sharing restriction now carries its named exceptions; the task-continuity claim restated; two AWS labels in Figure 1 restored to their qualifications; a universal statement in Figure 2 narrowed.
- **9 October 2026** — Figures redrawn to one design: Figure 1 as a route ledger, Figure 2 as a proof-boundary map, Figure 3 as a grouped route atlas and Figure 4 as a dated record with the evidence cutoff marked. No finding changed; figure text equivalents now follow the figure data.

**Claim ledger** — 128 atomic claims; 126 supported as stated, 2 unverified (graded reported). Each source-based row carries a supporting passage and a check date; the 3 inference rows carry premises and reasoning instead. The full ledger: `data/claim-ledger.csv`.

| Id | Claim | Grade | Verdict | Passage | Sources |
|---|---|---|---|---|---|
| C001 | Snowflake schedules final loss of access to legacy provider-consumer clean rooms for 2027-06-01. | fact | supported | “All legacy Provider and Consumer clean rooms will no longer be accessible.” | [2] |
| C002 | Snowflake schedules the legacy clean-room web interface to become inaccessible on 2027-02-01. | fact | supported | “The web application user interface will no longer be accessible” | [2] |
| C003 | The 2026-10-01 release stops creating new clean rooms in the legacy UI. | fact | supported | “no longer create new clean rooms” | [1] |
| C004 | Native App agent observability redacts orchestrator-authored content from consumer records by default. | fact | supported | “redacts orchestrator-authored content” | [30] |
| C005 | Snowflake announced an AI Gateway for Advertising in preview on 2026-10-05. | fact | supported | “AI Gateway for Advertising, currently in preview” | [38] |
| C006 | Snowflake announced GA support for Cortex Agents and MCP servers in Native Apps on 2026-08-07. | fact | supported | “Cortex Agents and MCP servers in Snowflake Native Apps are now generally available.” | [36] |
| C007 | Snowflake describes a Meta loop that sends governed conversion signals onto Meta's delivery and optimization systems. | vendor-claimed | supported | “directly onto Meta’s delivery and optimization systems” | [77] |
| C008 | The 2026-08-20 clean-room release removes listed legacy identity connectors including LiveRamp. | fact | supported | “the LiveRamp, Acxiom, Merkury, and PAIR third-party identity connectors have been removed.” | [49] |
| C009 | Cross-region Snowflake secure sharing uses data replication. | fact | supported | “cross-region data sharing utilizes Snowflake data replication functionality” | [78] |
| C010 | Snowflake Collaboration API specifications explicitly declare permitted activation destinations. | fact | supported | “activation_destinations: # Optional: Where results can be sent” | [8] |
| C011 | Snowflake documents its clean-room agent as an open preview. | fact | supported | “Preview Feature — Open” | [7] |
| C012 | The Snowflake clean-room agent can activate overlapping records to an approved account. | fact | supported | “exports it to an approved account” | [7] |
| C013 | For collaborator activation in Snowflake, the receiving collaborator must import the results; ordinary analysis results are returned in the workspace. | fact | supported | “the receiving collaborator must call `VIEW_ACTIVATIONS()` and `PROCESS_ACTIVATION()` to import the results.” | [6] |
| C014 | Databricks announced CustomerLake in private preview on 2026-06-16. | fact | supported | “CustomerLake is now available in Private Preview.” | [48] |
| C015 | Databricks proposed consumption-based pricing for CustomerLake. | vendor-claimed | supported | “value-aligned consumption model” | [48] |
| C016 | Databricks clean-room output tables are temporary and read-only. | fact | supported | “temporary read-only tables” | [9] |
| C017 | Databricks documents a 30-day output-table storage window; retention requires copying. | fact | supported | “If you want to keep an output table for more than 30 days, you must copy it to local storage.” | [9] |
| C018 | Databricks Clean Rooms supports collaboration without exposing other collaborators’ underlying raw data. | fact | supported | “Collaborators cannot see the data in other collaborators’ tables, views, or volumes” | [56] |
| C019 | Databricks MCP services are governed through Unity Catalog and Unity Gateway. | fact | supported | “Databricks-provided and registered MCPs use Unity Catalog permissions. Use Unity Gateway to control access, manage credentials, and monitor usage.” | [80] |
| C020 | Databricks announced Meta’s ads MCP server in its Marketplace on 2026-10-06. | fact | supported | “now available in Databricks Marketplace” | [39] |
| C021 | Databricks describes Meta-side rules checked before each ads MCP tool execution. | vendor-claimed | supported | “checks every tool call” | [39] |
| C022 | Unity Catalog documentation describes lineage tracking and activity logging for auditing. | fact | supported | “tracking lineage as data and AI assets are used, logging activity for auditing” | [82] |
| C023 | On 2026-03-27, the Agent Bricks page expanded to cover all Databricks AI capabilities. | fact | supported | “umbrella for all Databricks AI capabilities” | [83] |
| C024 | AWS Clean Rooms list analysis can output row-level overlap lists. | fact | supported | “outputs row-level lists of the overlap” | [10] |
| C025 | AWS custom-model collaborations assign the ability to receive trained model exports, and separately inference output, to a named member. | fact | supported | “Assign `CAN_RECEIVE_MODEL_OUTPUT` or `CAN_RECEIVE_INFERENCE_OUTPUT` abilities to the member that will receive trained model exports or inference output, respectively.” | [11] |
| C026 | AWS controls receipt of inference output separately from receipt of model artifacts. | fact | supported | “Assign `CAN_RECEIVE_MODEL_OUTPUT` or `CAN_RECEIVE_INFERENCE_OUTPUT` abilities to the member that will receive trained model exports or inference output, respectively.” | [11] |
| C027 | AWS Clean Rooms SQL and PySpark pricing uses CRPU consumption. | fact | supported | “AWS Clean Rooms measures compute capacity in Clean Rooms Processing Unit (CRPU)-hours” | [84] |
| C028 | AWS added an explicit query-analysis-log export capability on 2026-08-11. | fact | supported | “To export logs, your membership needs the `CAN_EXPORT_QUERY_ANALYSIS_LOG` member ability.” | [28] |
| C029 | AWS announced synthetic dataset generation for Clean Rooms ML on 2025-11-30. | fact | supported | “You can start using privacy-enhancing synthetic dataset generation through AWS Clean Rooms today.” | [12] |
| C030 | AWS states that trained model weights may be exported from the synthetic-data workflow. | fact | supported | “export the trained model weights” | [12] |
| C031 | AWS query/job logs do not establish whether output was delivered. | fact | supported | “don't report whether query output was delivered” | [27] |
| C032 | When an AWS Clean Rooms table's data access budget reaches zero, the table can no longer be used in SQL queries, PySpark jobs or ML input channels. | fact | supported | “When the budget reaches zero, the table can't be used in SQL queries, Pyspark jobs, nor as part of ML input channels derived from the table.” | [34] |
| C033 | AWS Entity Resolution documents provider-led matching among its matching options. | fact | supported | “data service provider-led matching” | [85] |
| C034 | AWS documentation uses Amazon Quick as the current name and identifies Quick Suite as the former name. | fact | supported | “Amazon Quick (formerly Amazon Quick Suite)” | [86] |
| C035 | BigQuery clean-room contributors can restrict copying and exporting query results. | fact | supported | “Data contributors can configure additional controls to prevent copying and exporting query results that subscribers obtain.” | [15] |
| C036 | Ads Data Hub noise injection requires approximately 20 unique users per result row. | fact | supported | “Noise injection requires approximately 20 unique users per result row.” | [13] |
| C037 | Ads Data Hub difference checks require approximately 50 unique users per row, and approximately 10 where a query uses only click and conversion data. | fact | supported | “Difference checks require approximately 50” | [13] |
| C038 | ADH supports audience activation into linked Google Ads and Display & Video 360 accounts. | fact | supported | “audiences, which can then be shared with your linked Google Ads and Display & Video 360 accounts.” | [14] |
| C039 | Google documents a remote BigQuery MCP server. | fact | supported | “You can use the BigQuery remote MCP server to perform tasks such as running queries, getting metadata, and listing resources.” | [43] |
| C040 | BigQuery DP enforcement is GA while parameter-driven DP budgeting remains preview. | fact | supported | “Parameter-driven privacy budgeting for differential privacy is in preview … differential privacy enforcement in BigQuery data clean rooms is now generally available (GA)” | [58] |
| C041 | GA4 BigQuery Export exports raw events from Google Analytics properties to BigQuery. | fact | supported | “You can export all of your raw events from Google Analytics properties” | [87] |
| C042 | Google added RemoveAllAudienceMembers to Data Manager API on 2026-07-30. | fact | supported | “Added the `RemoveAllAudienceMembers` method to the `IngestionService` to remove or replace all audience members” | [35] |
| C043 | Google describes Meridian as open-source, with its code fully visible and available to modify. | fact | supported | “Because Meridian is open-source, its underlying code is fully visible and available for you to modify.” | [88] |
| C044 | Google’s PAIR description says the advertiser and publisher each receive the match rate of their own audience list, with no user-level matches or PII sent back. | fact | supported | “the advertiser and the publisher both receive the percentage match rate of their respective first-party audience lists, but no user-level matches or Personally Identifiable Information (PII) will be sent back.” | [55] |
| C045 | Amazon announced direct AMC access for sponsored ads advertisers on 2025-09-16. | fact | supported | “instant, self-service access for sponsored ads advertisers directly within their Amazon Ads account.” | [89] |
| C046 | Amazon’s 2026-09-29 announcement uses the name Amazon Ads Agent. | fact | supported | “Amazon Ads Agent” | [90] |
| C047 | AMC SQL Generator expert tool launched through MCP in closed beta on 2026-09-30. | fact | supported | “Expert tools are now available in closed beta on the Amazon Ads MCP Server” | [40] |
| C048 | Amazon says expert-tool GA is planned for early 2027 with usage-based pricing. | attributed | supported | “General availability (early 2027): Pay-per-use via AWS Marketplace” | [40] |
| C049 | AMC on AWS Clean Rooms returns aggregate-analysis results to the advertiser’s S3 bucket; audience queries return results to Amazon Ads’ S3 bucket. | fact | supported | “Run a SQL query with AMC data and the advertiser's data that return the results to the advertiser's S3 bucket using aggregate analysis.” | [19] |
| C050 | Amazon describes AMC as a no-cost clean room for eligible advertisers. | vendor-claimed | supported | “available at no cost” | [91] |
| C051 | Prime Video viewership insights were announced in open beta on 2025-11-11. | fact | supported | “Amazon Marketing Cloud (AMC) now includes Prime Video viewership signals in open beta” | [92] |
| C052 | PPC Land reported on 2026-06-02 that Amazon made AMC first-party paid-feature signals free to query through 2026-12-31, while third-party signals such as Experian still need a subscription. | reported | unverified | “are available at no additional cost, effective immediately through December 31, 2026 … third-party paid features signals - such as those from Experian - continue to require a subscription fee” | [51] [53] |
| C053 | WPP announced its acquisition of InfoSum on 2025-04-03. | fact | supported | “acquisition of InfoSum” | [45] |
| C054 | InfoSum announced Beacons across AWS, Azure and Google Cloud on 2025-11-12. | fact | supported | “Reimagined infrastructure will deploy directly into customer environments across Amazon Web Services (AWS), Google Cloud, and Microsoft Azure” | [93] |
| C055 | InfoSum's activation documentation describes configuring an export of a segment to a chosen destination. | fact | supported | “To configure an export of that segment to your desired Destination” | [16] |
| C056 | InfoSum API permissions separately expose identifier-export authorization. | fact | supported | “allows operators to activate export columns from this dataset.” | [17] |
| C057 | InfoSum documents that closing a collaboration it owns is irreversible and stops new queries or exports, while previous activity can still be seen. | fact | supported | “can no longer be used for new queries or exports but previous activity can still be seen” | [17] |
| C058 | InfoSum describes redaction, rounding and noise injection as privacy techniques. | vendor-claimed | supported | “redaction thresholds, rounding, and noise injection” | [94] |
| C059 | WPP announced that WPP Media replaces GroupM on 2025-05-28. | fact | supported | “WPP Media replaces GroupM” | [95] |
| C060 | As of its 2026-09-16 financing announcement, Publicis still described the LiveRamp acquisition as subject to approvals. | fact | supported | “As a reminder, the acquisition of LiveRamp remains subject to customary approvals and is expected to close before the end of 2026.” | [5] |
| C061 | LiveRamp stockholders adopted the merger agreement on 2026-08-17. | fact | supported | “stockholders adopted the Merger Agreement” | [47] |
| C062 | LiveRamp clean-room controls require data-owner permission for analytics and activation uses. | fact | supported | “Activation or Analytics only allowed with the express permission of the data owner.” | [18] |
| C063 | LiveRamp’s March 2026 notes describe agent-powered access to its platform. | vendor-claimed | supported | “enabling AI agents to operate across identity, segmentation, activation, and measurement” | [42] |
| C064 | LiveRamp completed its acquisition of Habu on 2024-01-31. | fact | supported | “On January 31, 2024, the Company completed the acquisition of Habu, a data clean room software provider” | [44] |
| C065 | LiveRamp set the ATS Mobile SDK sunset effective 2026-09-30. | fact | supported | “LiveRamp has made the decision to sunset the ATS Mobile SDK effective September 30, 2026.” | [97] |
| C066 | dentsu announced Tobiras using Meta Advanced Analytics on 2024-11-21. | fact | supported | “Tobiras, a new solution that combines an advertiser’s first party data with Meta’s Advanced Analytics data” | [59] |
| C067 | Meta’s help page, as indexed, says web advertisers no longer need to prioritize eight conversion events per domain; a change date was not found. | reported | unverified | “You no longer need to prioritize 8 conversion events per domain” | [98] [114] |
| C068 | Meta’s 2026-03-03 indexed announcement uses engage-through attribution terminology. | fact | supported | “it will be known as engage-through attribution.” | [99] |
| C069 | Robyn documents calibration using causal experiment results. | fact | supported | “Calibration of average effect size with causal experiments” | [100] |
| C070 | MCP specifies transport-level authorization for access to MCP resources. | fact | supported | “The Model Context Protocol provides authorization capabilities at the transport level” | [101] |
| C071 | A2A requires authorization checks on requests. | fact | supported | “Servers MUST return an authorization error when the authenticated client lacks required permissions” | [102] |
| C072 | AdCP documents advertising task interfaces over MCP and A2A. | fact | supported | “Planning-time tasks use MCP and A2A transports” | [103] |
| C073 | IAB Tech Lab presents AAMP as a portfolio/program of agentic advertising protocols. | fact | supported | “AAMP is an IAB Tech Lab umbrella initiative for bringing agentic AI into the advertising ecosystem” | [104] |
| C074 | IAB Tech Lab lists PAIR 1.1 as finalized in July 2025 and OPJA as deprecated. | fact | supported | “PAIR version 1.1 was finalized in July 2025. … OPJA if therefore a deprecated standard and no longer supported.” | [105] |
| C075 | InfoSum says a company that leaves a collaboration can no longer see its details, including the event logs; the owner must close rather than leave. | fact | supported | “Leaving the collaboration will mean the leaving company will no longer be able to see any details about the collaboration including the event logs.” | [17] |
| C076 | Amazon announced AMC Custom Models for Audiences on 2025-10-31, which lets customers use their proprietary modeling algorithms to create custom audiences. | fact | supported | “enables customers to leverage their proprietary modeling algorithms to create custom audiences” | [26] |
| C077 | MLflow on Databricks records traces of an application's inputs, outputs and metadata. | fact | supported | “inputs, outputs, and metadata” | [57] |
| C078 | LiveRamp’s filed communication specifies cash consideration of $38.50 per share upon the proposed transaction. | fact | supported | “We announced that LiveRamp has entered into a definitive agreement to be acquired by Publicis, a global leader in communications, for $38.50 per share in cash” | [54] |
| C079 | LiveRamp says it is committed to an independence charter addressing open access, privacy and pricing controls after the transaction. | attributed | supported | “We are committed to maintaining operational neutrality and abiding by an independence charter focused on open access, privacy and pricing controls” | [54] |
| C080 | LiveRamp’s documented Meta/Facebook Advanced Analytics connection requires an FAA instance and CAPI support. | fact | supported | “You must have a Facebook Advanced Analytics instance configured and Conversions API (CAPI) support prior to initiating” | [20] |
| C081 | LiveRamp documents a 90-day queried-data availability window in Facebook Advanced Analytics. | fact | supported | “Queried data is available for 90 days in Facebook Advanced Analytics.” | [20] |
| C082 | Publicis’s 2026-05-17 announcement specifies total enterprise value of $2.167 billion for the proposed LiveRamp transaction. | fact | supported | “total enterprise value of $2.167 billion” | [4] |
| C083 | On 2026-09-10 Snowflake removed a second set of legacy web-interface identity connectors, including Acxiom Real ID, Experian and The Trade Desk UID 2.0. | fact | supported | “the Acxiom - Real ID, Acxiom Real ID Transcoding, LiveRamp, Merkury, Experian, and The Trade Desk UID 2.0 connectors” | [50] |
| C084 | Snowflake's 2026-09-10 connector removal did not delete existing connector configurations. | fact | supported | “This change does not delete existing connector configurations.” | [50] |
| C085 | Snowflake made managing clean-room collaborations in the Snowsight interface generally available on 2026-09-24. | fact | supported | “Managing collaborations in the Snowsight UI is now generally available.” | [29] |
| C086 | BigQuery's MCP server became generally available on 2026-04-20. | fact | supported | “This feature is Generally Available (GA).” | [37] |
| C087 | Query templates for BigQuery data clean rooms became generally available on 2026-08-11. | fact | supported | “Query templates for data clean rooms are generally available (GA).” | [37] |
| C088 | IAB Tech Lab’s AAMP 3.0, released on 2026-09-22, introduced the OpenProposal specification, in public comment until 2026-10-22. | fact | supported | “The new specification will be in Public Comment until October 22, 2026” | [52] |
| C089 | IAB Tech Lab says AAMP 3.0 work includes standardizing agent authentication and authorization. | attributed | supported | “standardizing agent authentication and authorization protocols” | [52] |
| C090 | Publicis and LiveRamp filed US premerger (HSR) notifications on 2026-06-11. | fact | supported | “On June 11, 2026, Publicis and LiveRamp filed their respective Premerger Notification and Report Forms” | [46] |
| C091 | CFIUS approval is a condition of closing the Publicis–LiveRamp merger. | fact | supported | “the approval of the Committee on Foreign Investment in the United States” | [46] |
| C092 | The merger agreement sets an outside date of 2027-05-16, automatically extended to 2027-08-16 if only certain regulatory conditions remain unmet. | fact | supported | “May 16, 2027 (the “Outside Date”) … the Outside Date, which will be automatically extended to August 16, 2027 if certain regulatory closing conditions remain the only conditions not satisfied or waived” | [46] |
| C093 | Publicis said on 2026-09-16 that it expects the LiveRamp acquisition to close before the end of 2026. | attributed | supported | “expected to close before the end of 2026” | [5] |
| C094 | No filing or announcement of the Publicis–LiveRamp closing was found in the sources searched through 2026-10-09. | inference | supported |  | [5] [47] [46] |
| C095 | LiveRamp announced on 2026-08-04 that marketers can access de-duplicated attribution and incrementality measurement from Meta in the LiveRamp Clean Room. | fact | supported | “Marketers can access de-duplicated attribution and incrementality measurement from Meta in the LiveRamp Clean Room.” | [112] |
| C096 | During the closed beta, Amazon's AMC expert tools are available in the United States. | fact | supported | “During closed beta, expert tools are available in the United States.” | [40] |
| C097 | Google announced Meridian Studio, a Google Cloud-based enterprise platform for building Meridian models, on 2026-05-28 without publishing a release stage. | fact | supported | “Meridian Studio: A Google Cloud-based enterprise platform to build and scale models, featuring native agentic AI” | [108] |
| C098 | Google said on 2026-09-10 that GeoX is generally available globally in Meridian. | fact | supported | “Now, GeoX is generally available globally in Meridian” | [109] |
| C099 | Databricks made the Genie One MCP server generally available on 2026-09-25. | fact | supported | “The Genie One MCP server is now generally available” | [3] |
| C100 | AWS documents Amazon Quick as having evolved from Amazon QuickSight. | fact | supported | “Amazon Quick evolved from Amazon QuickSight.” | [113] |
| C101 | Google says standard Universal Analytics properties stopped processing hits from 2023-07-01. | fact | supported | “Standard Universal Analytics properties stopped processing hits” | [110] |
| C102 | The A2A changelog dates version 1.0.1 2026-05-26; GitHub published the release on 2026-05-28. | fact | supported | “1.0.1 (2026-05-26)” | [111] |
| C103 | Task continuity was not assessed, because the study did not define a specific task and target environment for each route. | inference | supported |  | [9] [11] [15] [13] [17] [8] [18] [26] [20] |
| C104 | Three routes in Figure 3 carry a vendor-stated GA label. | inference | supported |  | [36] [37] [3] |
| C105 | Databricks scheduled the former Beta endpoint /api/2.0/mcp/genie to sunset on 2026-10-31. | fact | supported | “is deprecated and will be sunset on October 31, 2026” | [3] |
| C106 | AWS documents an aws-cleanrooms skill, used through AWS MCP, that troubleshoots permission errors and missing CloudWatch logs for custom ML jobs. | fact | supported | “Missing CloudWatch logs for custom ML model jobs – CloudWatch logs not appearing for custom model training or inference jobs.” | [41] |
| C107 | Deleting a Databricks clean room removes all its output tables and historical data from the output catalog. | fact | supported | “Deleting a clean room removes all output tables and historical data from the output catalog.” | [9] |
| C108 | By default a Databricks clean-room output table is accessible only to the collaborator who runs the notebook; sharing output with all collaborators is in Public Preview. | fact | supported | “By default, an output table is accessible only to the collaborator who runs the notebook” | [9] |
| C109 | In a Snowflake collaboration, ordinary analysis results are returned in the workspace. | fact | supported | “Results are returned in the workspace.” | [6] |
| C110 | AWS CloudTrail can record S3 object-level events such as deletions as data events, which are off by default and charged. | fact | supported | “By default, trails and event data stores do not log data events. Additional charges apply for data events.” | [31] |
| C111 | In a versioning-enabled S3 bucket, a simple DELETE inserts a delete marker instead of permanently deleting the object. | fact | supported | “When versioning is enabled, a simple `DELETE` cannot permanently delete an object” | [33] |
| C112 | BigQuery audit metadata records how a table was deleted (for example a delete request, expiry or a DDL query). | fact | supported | “Describes how the table was deleted.” | [32] |
| C113 | LiveRamp said in March 2026 that it was selectively enabling a limited number of builders to deploy agents. | fact | supported | “we are selectively enabling a limited number of builders to deploy agents” | [42] |
| C114 | Google warns that BigQuery analysis rules might not fully prevent sophisticated queries from extracting unauthorized data, and points to query templates to restrict execution to approved queries. | fact | supported | “might not fully prevent sophisticated queries from extracting unauthorized data” | [15] |
| C116 | Amazon says AMC agents can serve advertisers across North America, Europe and Far East marketplaces from US-hosted infrastructure. | fact | supported | “Agents can serve advertisers across North America, Europe and Far East marketplaces from U.S.-hosted infrastructure.” | [40] |
| C117 | Amazon says broader geography for the expert tools will be available before general availability in early 2027. | attributed | supported | “Broader geography will be available before general availability in early 2027.” | [40] |
| C118 | The 2026-09-10 Snowflake release also removed legacy web-interface activation connectors, including Google Display & Video 360, LiveRamp and The Trade Desk. | fact | supported | “the Google Display & Video 360, Google Display & Video 360 - PAIR, LiveRamp, The Trade Desk - UID 2.0, The Trade Desk - CRM, and Yahoo DSP connectors have been removed from the provider and consumer web interface.” | [50] |
| C119 | Snowflake's Snowsight collaboration view has an Activity tab that records analysis template runs. | fact | supported | “A new Activity tab gives you an audit trail of the analysis template runs in a collaboration.” | [29] |
| C120 | Amazon covers the cost of the AMC expert tools during the closed beta. | fact | supported | “Costs are covered during the closed beta” | [40] |
| C121 | Databricks made the Unity Gateway API, which manages model and MCP services, generally available on 2026-09-16. | fact | supported | “Unity Gateway API and developer tools are generally available” | [3] |
| C122 | Google says Universal Analytics 360 properties with a processing extension had it end on 2024-07-01. | fact | supported | “received a one-time processing extension ending on July 1, 2024” | [110] |
| C123 | Snowflake schedules the end of new legacy clean-room creation through the Provider and Consumer API on 2027-02-01. | fact | supported | “new legacy clean rooms can no longer be created through the Provider and Consumer API” | [2] |
| C124 | Google’s Ads Data Hub Policies prohibit using Google’s proprietary data to build or augment user profiles or audience lists beyond the audience activation channels Ads Data Hub approves. | fact | supported | “Attempt to use Google's proprietary data to build, edit or augment user profiles or to build, edit or augment audience lists for any uses beyond those permitted in the audience activation channels approved by Ads Data Hub” | [21] |
| C125 | Google documents joining the results of multiple Ads Data Hub aggregation queries in BigQuery. | fact | supported | “You can join results (in BigQuery) from multiple aggregation queries (from Ads Data Hub).” | [22] |
| C126 | AWS Service Terms prohibit using AWS Clean Rooms, or information obtained from it, to identify a person, unless the data’s third-party contributor permits it. | fact | supported | “You may not use AWS Clean Rooms or any information obtained from your use of AWS Clean Rooms to identify a person … unless otherwise permitted by the applicable third-party contributor of the data.” | [23] |
| C127 | AWS Service Terms say deleting a Clean Rooms collaboration or its resources ends access to the customer’s content or deletes it from intermediate tables, as the documentation describes. | fact | supported | “will result in termination of access to Your Content or deletion of Your Content from intermediate tables, as described in the applicable service documentation” | [23] |
| C128 | The Amazon Marketing Cloud Agreement makes query results confidential and restricts sharing them, with named exceptions for the relevant agency, the advertiser and service providers acting for the advertiser, each required to keep them confidential. | fact | supported | “The Query Results are confidential information, and you may not share them with Third Parties, except that you may share the Query Results with your relevant agency (if Customer is an Advertiser or Representative)” | [24] |
| C129 | Snowflake’s public Data Clean Rooms terms defer to the customer’s agreement for the Snowflake Service. | fact | supported | “Use of Snowflake Data Clean Rooms is governed by your agreement for use of the Snowflake Service.” | [25] |
Downloads: `data-collaboration-2026.pdf`, `plain/`, `data/claim-ledger.csv` / `.json`, `data/source-register.csv` / `.json`, `data/corrections-ledger.csv` / `.json`, `figures/*.svg`.

## References

1. Oct 1, 2026: Snowflake Data Clean Rooms updates | Snowflake Documentation. Snowflake, 2026-10-01. https://docs.snowflake.com/en/release-notes/2026/other/2026-10-01-dcr (SNW-S002; accessed 2026-10-09)
2. End of life for legacy Provider and Consumer Data Clean Rooms | Snowflake Documentation. Snowflake, n.d.; dated retirement milestones. https://docs.snowflake.com/user-guide/cleanrooms/pc-eol (SNW-S001; accessed 2026-10-09)
3. September 2026 (Databricks platform release notes). Databricks, Entries dated 2026-09-16 and 2026-09-25. https://docs.databricks.com/aws/en/release-notes/product/2026/september (DBX-S010; accessed 2026-10-09)
4. Publicis to acquire LiveRamp to accelerate data co-creation for smarter agents | Publicis Groupe. Publicis Groupe, 2026-05-17. https://www.publicisgroupe.com/en/news/press-releases/publicis-to-acquire-liveramp-to-accelerate-data-co-creation-for-smarter-agents (LVR-S009; accessed 2026-10-09)
5. Publicis Groupe successfully prices EUR 500 million of bond issue | Publicis Groupe. Publicis Groupe, 2026-09-16. https://www.publicisgroupe.com/en/news/press-releases/publicis-groupe-successfully-prices-our-500-million-of-bond-issue (LVR-S001; accessed 2026-10-09)
6. Run analysis and activation in Snowsight | Snowflake Documentation. Snowflake, n.d. (living documentation). https://docs.snowflake.com/en/user-guide/cleanrooms/collab-ui-run-analysis (SNW-S011; accessed 2026-10-09)
7. Snowflake Data Clean Rooms Agent | Snowflake Documentation. Snowflake, n.d. (living documentation). https://docs.snowflake.com/en/user-guide/cleanrooms/dcr-agent (SNW-S010; accessed 2026-10-09)
8. Collaboration specification | Snowflake Documentation. Snowflake, n.d. (living documentation). https://docs.snowflake.com/en/user-guide/cleanrooms/spec-collaboration (SNW-S009; accessed 2026-10-09)
9. Create and work with output tables in Databricks Clean Rooms | Databricks on AWS. Databricks, n.d.; page updated 2026-10-06. https://docs.databricks.com/aws/en/clean-rooms/output-tables (DBX-S002; accessed 2026-10-09)
10. List analysis rule - AWS Clean Rooms. Amazon Web Services, n.d. (living documentation). https://docs.aws.amazon.com/clean-rooms/latest/userguide/analysis-rules-list.html (AWS-S001; accessed 2026-10-09)
11. Custom ML modeling prerequisites - AWS Clean Rooms. Amazon Web Services, n.d. (living documentation). https://docs.aws.amazon.com/clean-rooms/latest/userguide/custom-model-prerequisites.html (AWS-S002; accessed 2026-10-09)
12. AWS Clean Rooms launches privacy-enhancing synthetic dataset generation for ML model training | AWS News Blog. Amazon Web Services, 2025-11-30. https://aws.amazon.com/blogs/aws/aws-clean-rooms-launches-privacy-enhancing-synthetic-dataset-generation-for-ml-model-training/ (AWS-S005; accessed 2026-10-09)
13. Privacy checks in Ads Data Hub  |  Google for Developers. Google, n.d.; page updated 2026-09-25. https://developers.google.com/ads-data-hub/guides/privacy-checks (GOO-S002; accessed 2026-10-09)
14. Build and manage audiences  |  ADH for Marketers  |  Google for Developers. Google, n.d. (living documentation). https://developers.google.com/ads-data-hub/marketers/guides/audience-activation (GOO-S003; accessed 2026-10-09)
15. Share sensitive data with data clean rooms  |  BigQuery  |  Google Cloud Documentation. Google, n.d.; page updated 2026-10-07. https://docs.cloud.google.com/bigquery/docs/data-clean-rooms (GOO-S001; accessed 2026-10-09)
16. Activate Data to a Destination – InfoSum. InfoSum, n.d. (living documentation). https://support.infosum.com/hc/en-us/articles/20496960672018-Activate-Data-to-a-Destination (INF-S003; accessed 2026-10-09)
17. InfoSum Developer API | InfoSum API Guide. InfoSum, n.d. (API v2 documentation). https://infosum.github.io/infosum-docs/api/v2/ (INF-S004; accessed 2026-10-09)
18. Understanding LiveRamp Clean Room Privacy and Governance Controls. LiveRamp, n.d.; page last modified 2026-03-16. https://docs.liveramp.com/connect/en/understanding-liveramp-clean-room-privacy-and-governance-controls.html (LVR-S003; accessed 2026-10-09)
19. Plan your advertising campaigns with Amazon Marketing Cloud on AWS Clean Rooms, now generally available | AWS News Blog. Amazon Web Services, 2024-07-31. https://aws.amazon.com/blogs/aws/plan-your-advertising-campaigns-with-amazon-marketing-cloud-on-aws-clean-rooms-now-generally-available/ (AMC-S004; accessed 2026-10-09)
20. Configure Facebook Advanced Analytics. LiveRamp, Last modified 2026-03-05. https://docs.liveramp.com/connect/en/configure-facebook-advanced-analytics.html (LVR-S008; accessed 2026-10-09)
21. Ads Data Hub Policies. Google, 2024-01-16 (last updated). https://developers.google.com/ads-data-hub/resources/policies (GOO-S015; accessed 2026-10-09)
22. Difference checks | Ads Data Hub. Google, n.d.; page updated 2026-05-18. https://developers.google.com/ads-data-hub/guides/legacy-diff-checks (GOO-S016; accessed 2026-10-09)
23. AWS Service Terms (section 92, AWS Clean Rooms). Amazon Web Services, Last updated 2026-10-01. https://aws.amazon.com/service-terms/ (AWS-S014; accessed 2026-10-09)
24. Amazon Marketing Cloud Agreement. Amazon Ads, Last updated 2024-10-11. https://advertising.amazon.com/marketing-cloud/terms/AMC_Terms.html (AMC-S009; accessed 2026-10-09)
25. Snowflake Data Clean Rooms Terms. Snowflake, Last updated 2024-04-30. https://www.snowflake.com/en/legal/other/data-clean-rooms/terms/ (SNW-S015; accessed 2026-10-09)
26. Create Machine Learning Models with AMC Custom Models | Amazon Ads. Amazon Ads, 2025-10-31. https://advertising.amazon.com/resources/whats-new/amc-custom-models-for-audiences (AMC-S008; accessed 2026-10-09)
27. Analysis logging in AWS Clean Rooms - AWS Clean Rooms. Amazon Web Services, n.d. (living documentation). https://docs.aws.amazon.com/clean-rooms/latest/userguide/query-logs.html (AWS-S006; accessed 2026-10-09)
28. Document history for the AWS Clean Rooms User Guide - AWS Clean Rooms. Amazon Web Services, n.d.; individual entries dated. https://docs.aws.amazon.com/clean-rooms/latest/userguide/doc-history.html (AWS-S004; accessed 2026-10-09)
29. Sep 24, 2026: Snowflake Data Clean Rooms updates. Snowflake, 2026-09-24. https://docs.snowflake.com/en/release-notes/2026/other/2026-09-24-dcr (SNW-S014; accessed 2026-10-09)
30. Oct 2, 2026: Snowflake Native Apps: Observability for Cortex Agents | Snowflake Documentation. Snowflake, 2026-10-02. https://docs.snowflake.com/en/release-notes/2026/other/2026-10-02-native-apps-agent-observability (SNW-S003; accessed 2026-10-09)
31. Logging data events. Amazon Web Services, n.d. (living documentation). https://docs.aws.amazon.com/awscloudtrail/latest/userguide/logging-data-events-with-cloudtrail.html (AWS-S011; accessed 2026-10-09)
32. BigQueryAuditMetadata.TableDeletion.Reason. Google Cloud, n.d.; page updated 2025-07-02. https://docs.cloud.google.com/bigquery/docs/reference/auditlogs/rest/Shared.Types/BigQueryAuditMetadata.TableDeletion.Reason (GOO-S014; accessed 2026-10-09)
33. Deleting object versions from a versioning-enabled bucket. Amazon Web Services, n.d. (living documentation). https://docs.aws.amazon.com/AmazonS3/latest/userguide/DeletingObjectVersions.html (AWS-S012; accessed 2026-10-09)
34. Associating a configured table to a collaboration - AWS Clean Rooms. Amazon Web Services, n.d. (living documentation). https://docs.aws.amazon.com/clean-rooms/latest/userguide/associate-configured-table.html (AWS-S007; accessed 2026-10-09)
35. Release notes  |  Data Manager API  |  Google for Developers. Google, Entries through 2026-07-30. https://developers.google.com/data-manager/api/reference (GOO-S007; accessed 2026-10-09)
36. Aug 7, 2026: Snowflake Native Apps: Cortex Agents and MCP servers (*General availability*) | Snowflake Documentation. Snowflake, 2026-08-07. https://docs.snowflake.com/en/release-notes/2026/other/2026-08-07-native-apps-agents-mcp-ga (SNW-S005; accessed 2026-10-09)
37. BigQuery release notes. Google Cloud, Entries dated 2026-04-20 and 2026-08-11. https://docs.cloud.google.com/bigquery/docs/release-notes (GOO-S011; accessed 2026-10-09)
38. Building a Governed Path for AI Agents in Advertising. Snowflake, 2026-10-05. https://www.snowflake.com/en/blog/ai-agents-in-advertising/ (SNW-S004; accessed 2026-10-09)
39. Meta’s ads MCP server comes to Databricks: Put your customer intelligence to work in advertising campaigns | Databricks Blog. Databricks, 2026-10-06. https://www.databricks.com/blog/meta-ads-mcp-databricks (DBX-S005; accessed 2026-10-09)
40. Amazon Ads MCP Server adds AMC SQL generation | Amazon Ads. Amazon Ads, 2026-09-30. https://advertising.amazon.com/en-ca/resources/whats-new/expert-tools-bring-amc-sql-generation-to-agent (AMC-S003; accessed 2026-10-09)
41. Troubleshooting AWS Clean Rooms. Amazon Web Services, n.d. (living documentation). https://docs.aws.amazon.com/clean-rooms/latest/userguide/troubleshooting.html (AWS-S013; accessed 2026-10-09)
42. March 2026 Platform Release Notes | LiveRamp. LiveRamp, March 2026; exact day not found. https://liveramp.com/our-platform/march-2026-platform-release-notes (LVR-S004; accessed 2026-10-09)
43. Use the BigQuery MCP server  |  Google Cloud Documentation. Google, n.d. (living documentation). https://docs.cloud.google.com/bigquery/docs/use-bigquery-mcp (GOO-S004; accessed 2026-10-09)
44. LiveRamp Holdings, Inc. Annual report on Form 10-K for the fiscal year ended 31 March 2024. LiveRamp Holdings, Inc. (SEC EDGAR), 2024-05-22. https://www.sec.gov/Archives/edgar/data/733269/000073326924000018/ramp-20240331.htm (LVR-S011; accessed 2026-10-09)
45. WPP acquires InfoSum in major investment in its AI-driven data offer | WPP. WPP, 2025-04-03. https://www.wpp.com/en/news/2025/04/wpp-acquires-infosum-in-major-investment-in-its-ai-driven-data-offer (INF-S001; accessed 2026-10-09)
46. LiveRamp Holdings, Inc. Definitive proxy statement (DEFM14A) on the proposed merger with Publicis. LiveRamp Holdings, Inc. (SEC EDGAR), 2026-07-06. https://www.sec.gov/Archives/edgar/data/733269/000110465926080505/tm264528-7_defm14aseq1.htm (LVR-S010; accessed 2026-10-09)
47. LiveRamp Holdings, Inc. — Form 8-K: shareholder adoption of merger agreement. LiveRamp filing / US SEC EDGAR, 2026-08-17. https://www.sec.gov/Archives/edgar/data/733269/000110465926097907/tm2623310d1_8k.htm (LVR-S002; accessed 2026-10-09)
48. Databricks Enters the Marketing Industry with CustomerLake: Agentic Customer Data Platform (CDP) - Databricks. Databricks, 2026-06-16. https://www.databricks.com/company/newsroom/press-releases/databricks-enters-marketing-industry-customerlake-agentic-customer (DBX-S001; accessed 2026-10-09)
49. Aug 20, 2026: Snowflake Data Clean Rooms updates | Snowflake Documentation. Snowflake, 2026-08-20. https://docs.snowflake.com/en/release-notes/2026/other/2026-08-20-dcr (SNW-S007; accessed 2026-10-09)
50. Sep 10, 2026: Snowflake Data Clean Rooms updates. Snowflake, 2026-09-10. https://docs.snowflake.com/en/release-notes/2026/other/2026-09-10-dcr (SNW-S013; accessed 2026-10-09)
51. Amazon now lets AMC users query 1P paid features for free until end of 2026. PPC Land, 2026-06-02. https://ppc.land/amazon-now-lets-amc-users-query-1p-paid-features-for-free-until-end-of-2026/ (TRD-S003; accessed 2026-10-09)
52. IAB Tech Lab Introduces AAMP 3.0 to Standardize the RFP-to-Buy Process for Agentic Advertising. IAB Tech Lab, 2026-09-22. https://iabtechlab.com/press-releases/iab-tech-lab-introduces-aamp-3-0-with-openproposal/ (STD-S006; accessed 2026-10-09)
53. Paid features overview | Amazon Ads API. Amazon Ads, n.d. (official documentation excerpt). https://advertising.amazon.com/API/docs/en-us/guides/amazon-marketing-cloud/datasources/paid_features_overview (AMC-S007; accessed 2026-10-09)
54. LiveRamp Holdings, Inc. Schedule 14A (DEFA14A): employee FAQ on the proposed Publicis transaction. LiveRamp, 2026-05-18. https://www.sec.gov/Archives/edgar/data/733269/000110465926063026/tm2614904d5_defa14a.htm (LVR-S007; accessed 2026-10-09)
55. Engage your first-party audience in Display & Video 360. Google, 2022-10-11. https://blog.google/products/marketingplatform/360/engage-your-first-party-audience-in-display-video-360/ (GOO-S009; accessed 2026-10-09)
56. What is Databricks Clean Rooms? | Databricks on AWS. Databricks, n.d. (living documentation). https://docs.databricks.com/aws/en/clean-rooms (DBX-S003; accessed 2026-10-09)
57. MLflow on Databricks | Databricks on AWS. Databricks, n.d. (living documentation). https://docs.databricks.com/aws/en/mlflow/ (DBX-S009; accessed 2026-10-09)
58. Restrict data access using analysis rules  |  BigQuery  |  Google Cloud Documentation. Google, n.d. (living documentation). https://docs.cloud.google.com/bigquery/docs/analysis-rules (GOO-S005; accessed 2026-10-09)
59. Tobiras Enhances Meta Analytics for Growth. dentsu, 2024-11-21. https://www.dentsu.com/news-releases/dentsu-launches-tobiras-with-metas-advanced-analytics-technology-to-win-in-the-algorithmic-era (MTA-S001; accessed 2026-10-09)
60. Enterprise Data Collaboration Playbook — No Fluff Advisory. No Fluff Advisory / Evgeny Popov, Last validated 2026-06-06. https://nofluffadvisory.com/services/playbooks/multicloud-data-orchestration/ (NFA-S000; accessed 2026-10-09)
61. Clean Room & Platform Fit Guide — No Fluff Advisory. No Fluff Advisory / Evgeny Popov, n.d.; observed 2026-10-09. https://nofluffadvisory.com/services/playbooks/multicloud-data-orchestration/platform-fit/ (NFA-S010; accessed 2026-10-09)
62. Snowflake Platform Fit for Data Collaboration | No Fluff Advisory. No Fluff Advisory / Evgeny Popov, Reviewed September 2026 (lastValidated 2026-09 at origin/master 918e2091). https://nofluffadvisory.com/services/playbooks/multicloud-data-orchestration/platform-fit/snowflake/ (NFA-S001; accessed 2026-10-09)
63. Databricks Platform Fit for Data Collaboration | No Fluff Advisory. No Fluff Advisory / Evgeny Popov, Reviewed September 2026. https://nofluffadvisory.com/services/playbooks/multicloud-data-orchestration/platform-fit/databricks/ (NFA-S002; accessed 2026-10-09)
64. AWS Platform Fit for Data Collaboration | No Fluff Advisory. No Fluff Advisory / Evgeny Popov, Reviewed September 2026. https://nofluffadvisory.com/services/playbooks/multicloud-data-orchestration/platform-fit/aws/ (NFA-S003; accessed 2026-10-09)
65. Google GMP, ADH & Advanced Marketing Analytics Platform Fit | No Fluff Advisory. No Fluff Advisory / Evgeny Popov, Reviewed September 2026. https://nofluffadvisory.com/services/playbooks/multicloud-data-orchestration/platform-fit/google-gmp-adh/ (NFA-S004; accessed 2026-10-09)
66. InfoSum Platform Fit for Data Collaboration | No Fluff Advisory. No Fluff Advisory / Evgeny Popov, Reviewed September 2026; June labels also present. https://nofluffadvisory.com/services/playbooks/multicloud-data-orchestration/platform-fit/infosum/ (NFA-S005; accessed 2026-10-09)
67. LiveRamp Platform Fit for Data Collaboration | No Fluff Advisory. No Fluff Advisory / Evgeny Popov, Reviewed September 2026; June labels also present. https://nofluffadvisory.com/services/playbooks/multicloud-data-orchestration/platform-fit/liveramp/ (NFA-S006; accessed 2026-10-09)
68. Amazon Marketing Cloud Platform Fit | No Fluff Advisory. No Fluff Advisory / Evgeny Popov, Reviewed September 2026 (lastValidated 2026-09); hard-coded June 2026 market-context label also present. https://nofluffadvisory.com/services/playbooks/multicloud-data-orchestration/platform-fit/amazon-marketing-cloud/ (NFA-S007; accessed 2026-10-09)
69. Meta Advanced Analytics Platform Fit | No Fluff Advisory. No Fluff Advisory / Evgeny Popov, Reviewed September 2026; June labels also present. https://nofluffadvisory.com/services/playbooks/multicloud-data-orchestration/platform-fit/meta-advanced-analytics/ (NFA-S008; accessed 2026-10-09)
70. Semantic & Agent-Ready Data Collaboration — No Fluff Advisory. No Fluff Advisory / Evgeny Popov, n.d.; observed 2026-10-09. https://nofluffadvisory.com/services/playbooks/multicloud-data-orchestration/agent-ready/ (NFA-S011; accessed 2026-10-09)
71. Multi-Cloud Data Collaboration Orchestration | No Fluff Advisory. No Fluff Advisory / Evgeny Popov, Reviewed 2026-09-15. https://nofluffadvisory.com/services/playbooks/multicloud-data-orchestration/platform-fit/multicloud-orchestration/ (NFA-S009; accessed 2026-10-09)
72. The Evolution of Holding Companies: Control, client value and three paths for marketing work. No Fluff Advisory / Evgeny Popov, 2026-10-06; evidence cutoff 2026-10-05. https://nofluffadvisory.com/research/holding-companies-2026/ (NFA-S101; accessed 2026-10-09)
73. Adtech and Martech Were Never Two Industries. They Were Two Copies of the Customer. — No Fluff Advisory. No Fluff Advisory / Evgeny Popov, 2026-07-05. https://nofluffadvisory.com/writing/the-agentic-cdp-two-industries-one-table/ (NFA-S102; accessed 2026-10-09)
74. The Loop Closed Inside the Wall — No Fluff Advisory. No Fluff Advisory / Evgeny Popov, 2026-07-21. https://nofluffadvisory.com/writing/the-loop-closed-inside-the-wall/ (NFA-S103; accessed 2026-10-09)
75. Proof and Story, in plain language. No Fluff Advisory / Evgeny Popov, August 2026; exact publication day not found. https://nofluffadvisory.com/research/contextual-2026/plain/ (NFA-S104; accessed 2026-10-09)
76. Paying Attention, in plain language. No Fluff Advisory / Evgeny Popov, October 2026; cutoff 2026-10-01. https://nofluffadvisory.com/research/attention-measurement-2026/plain/ (NFA-S105; accessed 2026-10-09)
77. Snowflake context and Meta campaigns: close the signal loop. Snowflake, 2026-07-21. https://www.snowflake.com/en/blog/snowflake-meta-campaigns-governed-conversion-signals/ (SNW-S006; accessed 2026-10-09)
78. Share data securely across regions and cloud platforms | Snowflake Documentation. Snowflake, n.d. (living documentation). https://docs.snowflake.com/en/user-guide/secure-data-sharing-across-regions-platforms (SNW-S008; accessed 2026-10-09)
79. About Secure Data Sharing | Snowflake Documentation. Snowflake, n.d. (living documentation). https://docs.snowflake.com/en/user-guide/data-sharing-intro (SNW-S012; accessed 2026-10-09)
80. MCPs | Databricks on AWS. Databricks, n.d.; page updated 2026-10-08. https://docs.databricks.com/aws/en/agents/agent-framework/mcp-services (DBX-S004; accessed 2026-10-09)
81. Connecting customer context to measurable ROI with agentic marketing | Databricks Blog. Databricks, 2026-10-01. https://www.databricks.com/blog/connecting-customer-context-measurable-roi-agentic-marketing (DBX-S006; accessed 2026-10-09)
82. What is Unity Catalog? | Databricks on AWS. Databricks, Updated 2026-09-11. https://docs.databricks.com/aws/en/data-governance/unity-catalog/ (DBX-S007; accessed 2026-10-09)
83. March 2026 | Databricks on AWS. Databricks, Entry 2026-03-27; page update label 2026-03-25. https://docs.databricks.com/aws/en/release-notes/product/2026/march (DBX-S008; accessed 2026-10-09)
84. AWS Clean Rooms Pricing. Amazon Web Services, n.d. (current price model). https://aws.amazon.com/clean-rooms/pricing/ (AWS-S003; accessed 2026-10-09)
85. What is AWS Entity Resolution? - AWS Entity Resolution. Amazon Web Services, n.d. (living documentation). https://docs.aws.amazon.com/entityresolution/latest/userguide/what-is-service.html (AWS-S008; accessed 2026-10-09)
86. MCP Server integration - Distributed Load Testing on AWS. Amazon Web Services, n.d. (living documentation). https://docs.aws.amazon.com/solutions/latest/distributed-load-testing-on-aws/mcp-server-integration.html (AWS-S009; accessed 2026-10-09)
87. BigQuery Export - Analytics Help. Google, n.d. (living documentation). https://support.google.com/analytics/answer/9358801?hl=en (GOO-S006; accessed 2026-10-09)
88. Meridian  |  Google for Developers. Google, Updated 2026-07-09. https://developers.google.com/meridian/docs/basics/meridian-introduction (GOO-S008; accessed 2026-10-09)
89. Direct Access to AMC for Sponsored Ads Advertisers | Amazon Ads. Amazon Ads, 2025-09-16. https://advertising.amazon.com/resources/whats-new/direct-access-to-amc-for-sponsored-ads-advertiser (AMC-S001; accessed 2026-10-09)
90. Introducing Amazon Ads Agent, your AI-powered ad platform | Amazon Ads. Amazon Ads, 2026-09-29. https://advertising.amazon.com/library/news/amazon-ads-agent (AMC-S002; accessed 2026-10-09)
91. Amazon Marketing Cloud: Discover advertising insights | Amazon Ads. Amazon Ads, n.d. (current product page). https://advertising.amazon.com/solutions/products/amazon-marketing-cloud?ref_=a20m_us_wn_amc_091625_p_amc (AMC-S005; accessed 2026-10-09)
92. AMC launches Prime Video insights | Amazon Ads. Amazon Ads, 2025-11-11. https://advertising.amazon.com/resources/whats-new/unboxed-2025-amc-launches-prime-video-insights (AMC-S006; accessed 2026-10-09)
93. InfoSum introduces Beacons | Press Release | InfoSum. InfoSum, 2025-11-12. https://www.infosum.com/blog/infosum-introduces-beacons-for-cross-cloud-data-collaboration (INF-S002; accessed 2026-10-09)
94. Privacy-first data collaboration | InfoSum. InfoSum, n.d. (current product page). https://www.infosum.com/platform/privacy (INF-S005; accessed 2026-10-09)
95. WPP Media launches as fully integrated, AI-powered media company | WPP. WPP, 2025-05-28. https://www.wpp.com/en/news/2025/05/wpp-media-launches-as-fully-integrated-ai-powered-media-company (INF-S006; accessed 2026-10-09)
96. LiveRamp Completes Acquisition of Habu | LiveRamp, Inc.. LiveRamp, 2024-01-31. https://investors.liveramp.com/news-releases/news-release-details/liveramp-completes-acquisition-habu (LVR-S005; accessed 2026-10-09)
97. About the Android ATS Mobile SDK. LiveRamp, n.d.; effective date 2026-09-30. https://developers.liveramp.com/authenticatedtraffic-api/docs/ats-mobile-sdk-for-android (LVR-S006; accessed 2026-10-09)
98. About Meta's Aggregated Event Measurement - Facebook. Meta / Robyn maintainers, n.d. (official help excerpt). https://www.facebook.com/business/help/721422165168355 (MTA-S002; accessed 2026-10-09)
99. Simplifying Ad Measurement for a Social-First World - Facebook. Meta / Robyn maintainers, 2026-03-03. https://www.facebook.com/business/news/click-attribution (MTA-S003; accessed 2026-10-09)
100. Key Features | Robyn. Robyn project (Meta open source), n.d. (living documentation). https://facebookexperimental.github.io/Robyn/docs/features/ (MTA-S004; accessed 2026-10-09)
101. modelcontextprotocol/docs/specification/2026-07-28/basic/authorization/index.mdx at main · modelcontextprotocol/modelcontextprotocol · GitHub. MCP project, Specification path 2026-07-28; repository main. https://github.com/modelcontextprotocol/modelcontextprotocol/blob/main/docs/specification/2026-07-28/basic/authorization/index.mdx (STD-S001; accessed 2026-10-09)
102. Overview - A2A Protocol. A2A project, Version 1.0.1; released on GitHub 2026-05-28 (STD-S007). https://a2a-protocol.org/v1.0.1/specification/ (STD-S002; accessed 2026-10-09)
103. GitHub - adcontextprotocol/adcp: Docs and reference implementation for the Ad Context Protocol · GitHub. AdCP project / AgenticAdvertising.org, n.d.; repository main at access. https://github.com/adcontextprotocol/adcp (STD-S003; accessed 2026-10-09)
104. AAMP Agentic Advertising Management Protocols from IAB Tech Lab. IAB Tech Lab, n.d.; page updated 2026-09-22. https://iabtechlab.com/standards/aamp-agentic-advertising-management-protocols/ (STD-S004; accessed 2026-10-09)
105. Data Clean Rooms Guidance - IAB Tech Lab. IAB Tech Lab, Updated 2025-07-22. https://iabtechlab.com/datacleanrooms/ (STD-S005; accessed 2026-10-09)
106. Publicis Acquires LiveRamp To Advance New Agentic AI Push. Forrester, 2026-05-20. https://www.forrester.com/blogs/publicis-acquires-liveramp-to-advance-new-agentic-ai-push/ (TRD-S001; accessed 2026-10-09)
107. Amazon Marketing Cloud launches Prime Video viewership signals for advertisers. PPC Land, n.d.; reports event of 2025-11-11. https://ppc.land/amazon-marketing-cloud-launches-prime-video-viewership-signals-for-advertisers/ (TRD-S002; accessed 2026-10-09)
108. The “What” and “Why” on Meridian: Ads DevCast E6. Google, 2026-05-28. https://ads-developers.googleblog.com/2026/05/the-what-and-why-on-meridian-ads.html (GOO-S010; accessed 2026-10-09)
109. Drive profitable growth with new data and measurement tools. Google, 2026-09-10. https://blog.google/products/ads-commerce/data-strength-updates/ (GOO-S012; accessed 2026-10-09)
110. Google Analytics 4 has replaced Universal Analytics. Google (Analytics Help), n.d. (living help page). https://support.google.com/analytics/answer/11583528 (GOO-S013; accessed 2026-10-09)
111. Release v1.0.1 · a2aproject/A2A. A2A project (GitHub), 2026-05-28. https://github.com/a2aproject/A2A/releases/tag/v1.0.1 (STD-S007; accessed 2026-10-09)
112. LiveRamp Adds Meta to Cross-Media Intelligence. LiveRamp, 2026-08-04. https://liveramp.com/news/liveramp-adds-meta-to-cross-media-intelligence (LVR-S012; accessed 2026-10-09)
113. What is Amazon Quick?. Amazon Web Services, n.d. (living documentation). https://docs.aws.amazon.com/quick/latest/userguide/what-is.html (AWS-S010; accessed 2026-10-09)
114. Meta's removal of Aggregated Event Measurement (AEM) and its implications for advertisers. DEPT (Kasper Koletzki Kuipers), 2023-05-31. https://www.dept.global/en-dk/insight/metas-removal-of-aggregated-event-measurement-aem-and-its-implications-for-advertisers/ (TRD-S004; accessed 2026-10-09)