Errata

Corrected; now readsLiveRamp's own blog materials assert that RampID-to-RampID matching delivers 44% higher match rates than HEM-to-HEM matching, but no methodology, sample, publisher list, or time period is disclosed in those materials, and no independent third-party source corroborates the figure. The mechanistic explanation — that HEM fails on "any" syntactic normalization divergence — overstates a real but partial cause: normalization failures are one documented failure mode, but RampID's graph also matches across multiple email addresses per person and alternative PII, advantages unrelated to normalization.

Corrected; now readsApple's ATT framework launched with iOS 14.5 on April 26, 2021, and correctly requires an explicit OS-level permission prompt before any app can access the IDFA. The remaining statistics, however, contain three distinct errors. First, the "US opt-out rates of approximately 75–80 percent" is a softened figure that obscures the actual early data. Flurry's platform-wide measurement (2.5 million daily US users, May 2021) showed only 4% US opt-in — implying a 96% opt-out rate.

Corrected; now readsThe Pixalate report (May–August 2022, 70+ billion US open-programmatic impressions) did find that 21% of Safari traffic appeared to originate from iCloud Private Relay IP addresses. However, the claim as written materially misrepresents the report's central finding: Pixalate concluded that approximately 90% of that 21% was fraudulent — bad actors spoofing iCloud Private Relay IPs in bid requests, not legitimate users of the service. Legitimate iCloud Private Relay usage accounted for only ~1–2% of Safari traffic.

Corrected; now readsIPv6 prefixes at the granularity assigned by the ISP to a premises (typically /56 or /60 for residential broadband, not /64, which is more commonly a per-device mobile assignment or on-premises subnet) can function as household-level cross-device identifiers, though this is weaker than IPv4 household matching because ISPs and devices increasingly rotate IPv6 prefixes for privacy.

Corrected; now readsVizio's Inscape subsidiary collected viewing data from more than 24 million opted-in US smart TVs as of 2024. In 2024 Inscape announced or expanded partnerships with Comscore, EDO, and Locality for currency-grade TV measurement data. Nielsen and VideoAmp are also established Inscape data licensees — Nielsen under a multi-year deal first expanded in 2022 (renewed again in 2025), VideoAmp under a deal renewed in March 2025 — but neither is evidenced by a 2024-specific press release cited in the claim's footnotes.

Corrected; now readsSamsung TIFA is generated as a hashed PSID. It was rolled out starting March 2019 on all 2016, 2017, 2018, and 2019 Samsung Smart TVs (with 2015 models targeted for end of 2019). Only the user can reset TIFA — via a PSID reset or factory reset — and advertisers cannot trigger a reset. The claim's phrase "manufactured from 2016 onward" is inaccurate: the Samsung developer documentation enumerates specific model-year cohorts (2016–2019) and does not assert open-ended coverage for all future manufacture years.

Corrected; now readsAmazon's advertising identifier is called simply the "Advertising ID" — the acronym "AFAI" (Fire TV Advertising Identifier) does not appear in the cited policy document or elsewhere in Amazon's developer documentation under that name. The identifier is a user-resettable unique identifier available on Fire TV devices running Fire OS 5.2.1.1 or later (correct) and on Fire tablets running Fire OS 5.1 or later (omitted from the claim). Child profiles return the null UUID `00000000-0000-0000-0000-000000000000` (correct).

Corrected; now readsApple tvOS exposes IDFV, scoped to a single developer's apps, as the primary device identifier available without ATT permission; IDFV is deleted if the user removes all apps from a given vendor. However, the claim that tvOS does not provide "a platform-wide cross-publisher advertising ID analogous to RIDA or TIFA" is materially incorrect. tvOS does expose IDFA (Identifier for Advertisers) via ASIdentifierManager — a platform-wide, device-scoped, user-resettable identifier that is functionally analogous to Roku's RIDA and Samsung's TIFA.

Corrected; now readsAmazon Ads and Roku announced a June 2025 partnership to deliver logged-in reach across an estimated 80 million US CTV households via Amazon DSP. The companies attributed this figure to Comscore data and characterized it as representing "more than 80% of US CTV households," though the specific Comscore product and methodology are not disclosed. Early closed-beta tests — conducted by the companies themselves, without published methodology — showed either 40% more unique viewers (per Amazon's press release) or 42% more unique reach (per AdExchanger's attribution to Roku's Kelly MacLean).

Corrected; now readsThe ICO published draft guidance on storage and access technologies in December 2024 (opened for consultation 20 December 2024, closing 14 March 2025) — not final "guidance" issued in January/February 2025. The ICO did not say fingerprinting "violates" GDPR and PECR; it said fingerprinting is "irresponsible" when used without consent and is "subject to the usual consent requirements" — a consent-required framing, not an outright violation finding. Final ICO guidance was not published until April 2026.

Corrected; now readsThe figures of 21,000+ publisher domains and 75% of comScore 100 publishers for ATS adoption are genuine LiveRamp 10-K disclosures and are most clearly attributable to the FY2024 10-K (ramp-20240331, filed May 2024). The same language likely carried forward into FY2025 and possibly FY2026 as business-description boilerplate, but direct verification of the FY2026 10-K text is unavailable due to SEC EDGAR access restrictions.

Corrected; now readsThe six-step sequence is broadly accurate as described in the official documentation, but two sub-claims are not supported by the cited source. (1) The internal Operator sequence: the claim states the Operator "applies a secret salt and hashes," implying salt-then-hash; the actual documentation language is "hashing and adding the secret salt value," listing hashing first, and explicitly declines to specify the internal sequence, calling it "multiple steps" performed solely by the Operator. The directional ordering in the claim is asserted but not supported by the source.

Corrected; now readsThe Trade Desk originated UID2 and has served as its only named Public Operator and interim administrator; IAB Tech Lab formally declined the technical administrator role in March 2022 (CEO Anthony Katsur). However, the claim contains three significant errors. First, the framing of The Trade Desk as "primary Public Operator managing encryption keys and token infrastructure" is materially outdated…

Corrected; now readsAs of mid-2026, EUID (European Unified ID) has The Trade Desk confirmed as its Core Administrator — not an independent third party. The Trade Desk is documented as the rule owner and joint controller under GDPR. EUID documentation uses "currently" language suggesting this arrangement may eventually transition to independent governance, but no such transition has been confirmed or scheduled.

Corrected; now readsYahoo ConnectID is built on more than 205 million authenticated U.S. users (Yahoo's own figure, mid-2024, unaudited). The publisher domain and data platform figures in the claim are stale: by September 2024 Yahoo's own materials reported 55,000+ publisher domains (not ~50,000) and interoperability with 35+ data platforms (not 30+). The cited adtechradar.com November 2025 URL does not contain any of the three specific numbers; the actual source is Yahoo's July 2024 LiveRamp integration press release — a Tier 5/6 self-reported document. All three figures are company-reported with no independent audit on record.

Corrected; now readsTransUnion TruAudience (incorporating Neustar) covers 98% of the U.S. adult population with an identity graph anchored in over 16 billion monthly phone signals (sourced from a March 2023 press release, not the January 2024 one), 1.9 billion total phone numbers for matching (1 billion mobile core), 111 million household-linked IP addresses, and over 125 million U.S. households, with a 99.5% persistency rate.

Corrected; now readsEpsilon CORE ID covers more than 200 million individuals according to company self-reporting, grounded in deterministic offline identifiers (name/address) and transaction records — no independently audited figure exists. Epsilon PeopleCloud (the broader platform of which CORE ID is one component) was reported as embedded in more than half of Publicis Groupe's top 30 advertiser accounts as of 2020–2021 disclosures; attributing this metric to CORE ID specifically overstates the sourcing.

Corrected; now readsGoogle PPID requires publishers to hash or encrypt their identifier such that it is "meaningless to Google." For Google's own demand channels (reservations, Google Demand), Ad Manager re-partitions the identifier into per-publisher partitioned IDs, preventing cross-publisher profiling through Google's own infrastructure. However, for third-party programmatic demand (Authorized Buyers, Open Bidders), Ad Manager shares the raw PPID value directly with buyers exactly as received from the publisher — without partitioning or transformation.

Corrected; now readsPAIR encrypts both advertiser and publisher first-party data three times using three different keys (an advertiser key, a publisher key, and a key shared between advertiser and publisher). No single participating party — advertiser, publisher, or shared-key holder — holds all three keys, preventing any one of them from reverse-decrypting the raw IDs. The claim's assertion that "including Google" cannot decrypt is not supported by the cited source, which does not address Google's own platform-level access.

Corrected; now readsLiveRamp's Interoperable IDs feature allows UID2 tokens to be embedded inside ATS envelopes, enabling publishers using ATS to pass a UID2 alongside their LiveRamp envelope in bid requests. Publishers must contact their LiveRamp representative to enable this feature — it is not on by default — and SHA-256 hashing of the email address is required. The cited source does not characterize this as "bilateral translation" (which implies a conversion layer mapping one ID to the other), nor does it describe the result as simultaneously signaling a RampID and UID2 as twin co-equal identifiers in a single request.

Corrected; now readsIdentity graph full interoperability faces structural headwinds on commercial, legal, and technical dimensions, but characterizing these as categorical "blocks" overstates the evidence. Commercially, providers face disincentives to full bidirectional interoperability because it would erode network-effect moats, but many selective free integrations exist (UID2 and EUID are open-source; RampID is free to access), making "commercially blocked" too strong — "commercially disincentivized toward selective rather than universal interoperability" is more accurate.

Corrected; now readsThe directional content of the claim is plausible — MAID identifiers do decay meaningfully each month, iOS ATT does accelerate that decay, HEM addresses do erode over time, and weekly refresh is a reasonable industry preference for programmatic activation. However, the specific figures (3–7% monthly MAID decay, 8–15% annual HEM decay) are sourced exclusively from GSDSI, a privately held commercial data brokerage, not an industry standards body.

Corrected; now readsOpenRTB 2.6's geo.type field (defined via AdCOM 1.0, List: Location Types) enumerates three source methods — 1 (GPS/Location Services), 2 (IP Address), 3 (User Provided, e.g., registration data) — allowing buyers to assess the reliability of metro, zip, and lat/lon values before making targeting decisions. There is no value 4 ("inferred") in the specification.

Corrected; now readsThe device.ext.ifa_type field (not device.ifa_type) was introduced by IAB Tech Lab in its 2018 OTT IFA guidelines to disambiguate which identifier system populated device.ifa. The original specification defined platform-specific values including rida (Roku), tifa (Samsung), afai (Amazon Fire TV), vida (Vizio), lguid (LG), idfa (Apple), and aaid (Android/Google), and these became widely used in the industry.

Corrected; now readsMany major CTV programmers, including NBCU, Tubi, and Roku, cite VPPA as their reason for not simultaneously passing device identifiers and show-level content information in programmatic bid requests — a practice reflecting a conservative, contested legal interpretation, not a confirmed legal prohibition. Whether VPPA actually covers bid-stream device identifiers as PII remains judicially unresolved: the Second Circuit's 2025 "ordinary person" standard in Solomon v.

Corrected; now readsAmazon has claimed its authenticated graph reaches approximately 90% of US households, but this figure is Amazon's own internal estimate with no independent third-party audit. It was not announced at the 2025 upfront (May 12): Amazon's official upfront recap does not mention it, and the load-bearing Roku partnership that makes 90% achievable was not announced until June 16, 2025. Third-party data (CIRP, eMarketer) puts Prime alone at roughly 80% of US households.

Corrected; now readsLG Ads Solutions signed a multi-year deal with Magnite (announced June 21, 2022) to make ACR data from opted-in LG smart TVs available to the buy side through Magnite's CTV SSP infrastructure. No specific TV-count figure was disclosed in the LG Ads–Magnite announcement. The "more than 20 million opted-in LG smart TVs" figure belongs to a separate 4.5-year licensing deal between LG Ads Solutions and iSpot.tv, not Magnite. Additionally, the cited eMarketer URL ("Samsung Ads Expands Magnite Deal") covers Samsung's distinct partnership with Magnite and provides no corroboration for this LG claim.

Corrected; now readsMicrosoft announced in May 2025 that Microsoft Invest (the DSP formerly known as Xandr Invest) would be shut down on February 28, 2026. The replacement is the Microsoft Advertising Platform, a consolidated self-serve buying tool that uses Copilot as its underlying AI technology — not a product called "Microsoft Invest." The parenthetical in the original claim inverts the naming: Microsoft Invest is the product being killed, not the replacement.

Corrected; now readsHEM and UID2 are significant identity bridges in CTV, but characterizing them as the "two dominant" solutions overstates UID2's market position given active competition from LiveRamp's RampID and other authenticated ID solutions. The LG Ad Solutions figures (165M smart TVs globally, 40M in U.S.) and the named publisher adoptions (FOX, Disney, WBD, NBCUniversal) are factually supported by primary sources, but Disney and WBD adoptions date to 2022–2023, not 2024.

Corrected; now readsThe specific figures of 71% HEM match rate (authenticated CTV) and 47% (open-web) cannot be found in the cited eMarketer source. Direct fetch of the cited eMarketer "FAQ on Identity Resolution" (Jan 7, 2026) confirms the article exists but contains neither figure — only a data-clean-room adoption statistic. The Strategus CTV attribution page likewise contains no 70–85% IP-match figure and no Nielsen 2024 citation. The Nielsen 70–85% IP-match claim cannot be traced to any retrievable Nielsen-published document; it appears to be a practitioner-circulated figure of unknown provenance.

Corrected; now readsVideoAmp holds complete JIC certification as of December 2024 (covering national currency, advanced audiences, and Personified Demos). VideoAmp has not received MRC accreditation for its measurement methodology as of mid-2026; however, it formally initiated the MRC audit process in May 2025, meaning it is actively in the audit pipeline rather than having no MRC engagement. The original claim's framing — relying solely on the December 2024 press release's silence on MRC — omits this materially important subsequent development.

Corrected; now readsiSpot acquired 605 in September 2023 (announced September 13), expanding its total U.S. TV device footprint to 82.7 million by adding 605's Charter Spectrum set-top box data covering 16.6 million homes. The acquisition also brought Inscape ACR data (also licensed by iSpot directly), and attribution partnerships with Catalina, PlaceIQ, and Polk Automotive Solutions.

Corrected; now readsComscore received MRC accreditation on March 20, 2024 for Total Household Rating and Average Audience estimates in national and local TV across all 210 measured local markets — at that point the only measurement vendor accredited for local TV measurement. Age/gender "households with" demographic breaks were explicitly excluded and remained unaccredited at that announcement. Comscore subsequently received MRC accreditation for those "households with" household-level age/gender demographic metrics in April 2025 (national and local).

Corrected; now readsWhat is established: Nielsen holds MRC accreditation for national Big Data + Panel; Comscore holds MRC accreditation for national and local household measurement; iSpot holds MRC accreditation for ad occurrence only. Nielsen does NOT hold JIC currency certification — it publicly refused to participate in the JIC from April 2023, a stance that remained unchanged through the 2025–2026 upfront cycle.

Corrected; now readsThe dominant CTV sales lift architecture chains CTV impressions (tagged with UID2 or HEM) to purchase records (from Circana/NCSolutions, Experian, TransUnion, or LiveRamp) inside a data clean room, computing incremental lift for exposed vs. control households without raw data exchange. Circana announced agreements to acquire NCSolutions and Nielsen's MMM business in August 2024, with an expected Q4 2024 close — but neither deal closed in Q4 2024. The NCSolutions acquisition closed June 2, 2025; the Nielsen MMM acquisition closed August 21, 2025.

Corrected; now readsWebsite visit attribution for CTV uses household IP matching, but the accuracy of this method is far lower than claimed. The CIMM/Go Addressable/Truthset study (November 2025) — benchmarked against ISP ground-truth records across nearly one billion IP records from six major data providers — found IP-to-postal linkages accurate on average only 13% of the time, and IP-to-email linkages accurate 16% of the time, with providers agreeing only 6.4% of the time on postal matches.

Corrected; now readsThe structural mechanism is accurate: CTV builds awareness and consideration days to weeks before purchase, and last-click attribution assigns 100% of conversion credit to the final channel (typically branded search), giving CTV zero measured contribution. However, the specific "30% of branded search conversions trace to a CTV exposure in the prior week" figure is not supported by the cited sources. The Strategus /blog/tv-attribution page contains no such statistic. The AdExchanger 2024 article does not address CTV-to-branded-search attribution at all.

Corrected; now readsIncrementality testing — comparing outcomes between exposed and holdout groups or via geo-testing — is a methodologically defensible attribution approach for CTV because it measures causation rather than correlation and does not require cross-environment individual-level identity bridging. AdExchanger characterized 2024 as "the breakout year of incrementality" in ad measurement broadly, not specifically in CTV. The tvScientific/MetricWorks partnership is a real named CTV incrementality implementation but was announced in July 2022, not 2024, and cannot serve as evidence of 2024 CTV growth.

Corrected; now readsThe 30% CTV return premium figure should be attributed to Analytic Partners' ROI Genome (750+ brands, 45 countries), not to "Digital Applied," which is a content site that does not produce this proprietary benchmark and whose CTV guide does not cite this figure. The Analytic Partners finding also does not specify MMM as its methodology — it derives from a broader commercial analytics database. The Circana acquisition of Nielsen's MMM business was announced in August 2024 but did not close until August 2025, so describing it as a "Q4 2024 acquisition" is inaccurate.

Corrected; now readsACR data derived from the glass-level signal of what displays on a smart TV screen operates independently of the programmatic bid stream and is substantially more difficult to manipulate than bid-stream signals, because it captures pixel and audio fingerprints at the display layer rather than through ad-delivery infrastructure — though the absolute claim that it "cannot be spoofed" is unsupported.

Corrected; now readsRoku, which self-reports approximately 40% U.S. smart TV unit share (Tier 5; not independently verified), uses its ACR data to identify which households saw linear TV campaigns and then activates CTV campaigns targeting incremental viewers not yet exposed — a reach and frequency optimization approach confirmed on Roku's advertising resources page.

Corrected; now readsMPC allows multiple parties to jointly compute a function over their private inputs such that no party learns anything beyond the output — in most deployed protocols this is achieved via secret sharing or garbled circuits rather than literal computation on ciphertext (which is homomorphic encryption, a distinct and costlier method).

Corrected; now readsThe market conflation of 'data clean rooms' with 'data collaboration platforms' is a real and documented distinction — Lotame's published content does draw this line, and the LiveRamp/Habu acquisition logic supports the activation-gap argument. However, the specific 23%-to-47% brand adoption figure between 2020 and 2024 is misattributed. It originates from a May 2024 Winterberry Group report ("The Current State and Future Outlook for Data Collaboration Platforms") sponsored by Adobe — not from Lotame analysis or the IAB State of Data 2023, which are cited as sources but do not contain these figures.

Corrected; now readsAWS Clean Rooms launched in preview January 2023, reached GA March 21, 2023; data is analyzed in place in S3 via AWS Glue catalog configured tables (not accurately described as a 'federated SQL architecture'); launch customers named in the GA blog included DISH Media, Fox Corporation, Comscore, and Amazon Marketing Cloud — Omnicom's Annalect does not appear in the cited GA blog post and should be removed or separately sourced; at launch, pricing used CRPU-hours with a 9 CRPU-hour monthly free tier for 12 months, though this free tier has since been discontinued and the current pricing page no longer offers it…

Corrected; now readsSnowflake acquired clean room startup Samooha in December 2023 (described by VentureBeat as its ninth acquisition of that year, though some outlets counted it as the eighth); the technology runs as a Snowflake Native App. In August 2024, Netflix announced partnerships with three clean room providers — Snowflake, LiveRamp, and InfoSum — simultaneously; Snowflake's offering launched immediately while the other two were described as coming in subsequent months. Netflix did not characterize Snowflake as its "first available clean room partner" in its newsroom announcement.

Corrected; now readsNBCU's Audience Insights Hub is a first-party data clean room launched in 2021 on One Platform, built on Snowflake (not AWS Clean Rooms), seeded with 150 million individual IDs and 50 million household IDs per NBCU CDO John Lee's disclosure to AdExchanger (self-reported, not independently audited). Separately, NBCU conducted a proof of concept labeled "Enhanced Audience Insights" — a Peacock + Amazon Ads integration via Amazon Publisher Cloud built on AWS Clean Rooms — that reportedly reached 3.6x more target consumers with 67% lower CPM…

Corrected; now readsAmazon Marketing Cloud is built on AWS Clean Rooms infrastructure and allows Fire TV and Prime Video (including ad-supported content formerly branded Freevee, which was shut down as a distinct service in late 2024) CTV impressions to be joined against Amazon's first-party purchase data. In September 2025 Amazon removed the onboarding/access barrier that had previously required Sponsored Ads advertisers to engage a partner or go through a registration process, giving them direct self-service access within the ads console.

Corrected; now readsThe Disney–Walmart Connect partnership (May 2024) and the Vizio acquisition (completed late 2024, adding Inscape ACR data) are both well-supported. However, two characterizations require correction. First, Walmart Luminate is not precisely "the clean room layer" for Walmart Connect: Luminate (being rebranded Scintilla) is a shopper-insights platform that sits behind Walmart Connect's MTA model and has some clean room functionality, but as of reporting it is a separate, not-yet-fully-integrated layer — and Walmart Connect's actual clean room capability was still in pilot with agency/CPG partners.

Corrected; now readsRetail media clean rooms — with AMC and Walmart Connect as two leading but not exclusive platforms — have materially advanced closed-loop CTV outcome measurement for CPG brands by connecting streaming ad exposure to deterministic purchase data. However, this ecosystem is not a settled "de facto standard": as of 2025, only 48% of U.S.

Corrected; now readsClean room latency varies materially by architecture. Google ADH carries a documented delay of approximately 6 hours (reduced from a prior 24–48 hour ceiling), making it unsuitable for same-day operational monitoring. ETL-pipeline-based clean rooms do face meaningful asynchrony that currently confines most deployments primarily to planning and retrospective measurement.

Corrected; now reads41% of retail media clean room users cite integrating clean rooms into existing optimization and analysis practices as their top hurdle (Skai 2025). The "39% of organizations broadly" figure is not an independent eMarketer finding — it is the same Skai dataset being passed through by eMarketer's FAQ article, which explicitly attributes it to "the 2025 State of Retail Media report." There is therefore only one underlying source, not two independent surveys providing corroboration.

Corrected; now readsThe Supreme Court DENIED certiorari in National Basketball Association v. Salazar (No. 24-994) on December 8, 2025. The case the Supreme Court actually granted cert in to resolve the VPPA "consumer" circuit split is Salazar v. Paramount Global, No. 25-459 (cert granted January 26, 2026), which came up through the Sixth Circuit (the narrow ruling), not the Second Circuit. The Second Circuit's holding in Salazar v. NBA, 118 F.4th 533 (2d Cir. 2024) — that newsletter subscribers qualify as VPPA consumers without a video-specific subscription — is accurately described, as is the circuit split itself.

Corrected; now readsIn February 2026 (announced February 11), California Attorney General Rob Bonta announced a $2.75 million settlement with Disney — the largest CCPA enforcement settlement at that time — arising from Disney's failure to honor consumers' opt-out-of-sale-and-sharing requests across its streaming platforms (Disney+, ESPN+, Hulu) and connected TV apps in a comprehensive, account-wide manner, as required by the CCPA. The violation was an opt-out implementation failure, not a failure of disclosure or consent.

Corrected; now readsVirginia (VDPA), Colorado (CPA), Connecticut (CTDPA), Texas (TDPSA), and Oregon (OCPA) all require businesses broadly to honor consumer opt-outs from targeted advertising, defined as advertising selected based on personal data from consumer activity across nonaffiliated (or unaffiliated) websites and online applications.

Corrected; now readsThe Texas Data Privacy and Security Act (TDPSA, effective July 1, 2024) does not set a revenue or data volume threshold — it applies to any commercial entity that conducts business in Texas or produces products/services consumed by Texas residents and processes personal data, unless the business qualifies as a small business under the SBA definition (which is industry-variable, not a fixed cutoff). Critically, even SBA-qualifying small businesses cannot sell sensitive personal data without prior consumer consent. The TDPSA's applicability structure is among the broadest of any major U.S.

Corrected; now readsEUID (European Unified ID) was created by The Trade Desk and LiveRamp (announced February 28, 2022) as a Europe/UK-specific variant of UID2 to address GDPR and UK data-protection requirements. EUID's participant documentation explicitly requires that consent be freely given (prohibiting cookie walls), specific, informed, unambiguous, demonstrated by clear affirmative action, and accompanied by withdrawal rights — all accurately described in the claim.

Corrected; now readsLiveRamp generated $812.9 million in revenue in fiscal year 2026 (ended March 31, 2026) — reported as $813 million in summary disclosures — at a 71 percent GAAP gross margin ($575 million GAAP gross profit) and 73 percent non-GAAP gross margin. The Publicis Groupe all-cash acquisition at $38.50 per share, representing a 29.8 percent premium to the May 15, 2026 close, $2.546 billion equity value, and $2.167 billion enterprise value, is accurately stated.

Corrected; now readsUID2 is a free, open-source framework that enables publishers and DSPs to adopt email-based pseudonymous targeting without LiveRamp, creating genuine competitive pressure on one layer of LiveRamp's onboarding business — but it is not a clean fee-bypass of LiveRamp's HEM onboarding, as UID2 and LiveRamp are interoperable and address partly distinct use cases (authenticated first-party email tokens vs. offline CRM-to-online-ID matching across device graphs).

Corrected; now readsThe underlying incident is real and well-sourced, but the claim contains two attributional errors that materially misrepresent the record. First, the "cannot feasibly decrypt every UID2 in incoming bid requests" formulation is not attributed to TTD's SVP Waseem Basheer; that concept was expressed by TTD's Chief Strategy Officer Samantha Jacobson. Basheer's actual quotes concern matching difficulty when signals are corrupted and the difficulty of detecting problems absent an anomaly signal.

Corrected; now readsThe TrueData acquisition (November 4, 2025, BusinessWire confirmed), the Future Today CTV partnership (June 5, 2025, 1.5 billion monthly CTV ad impressions, more than 50 million monthly viewers), and the Adaptive Identity probabilistic characterization are all accurate. Two issues require correction. First, TrueData is accurately described as US-based and deterministic, but the phrase "U.S.-first" does not appear in any source — sourcing calls it "US-based" or "top-tier US-based." Second, and more materially…

Corrected; now readsTransUnion acquired Neustar for $3.1 billion, closing on December 1, 2021 (not 2022), integrating the Neustar OneID into TruAudience. The combined graph spans 16 billion monthly phone signals, 125 million U.S. household records, and 10 billion daily real-time digital interactions across 250-plus media owner partnerships (all self-reported figures from a March 2023 TransUnion press release). TransUnion's FY2025 revenue was approximately $4.58 billion ($4,576 million), representing 9 percent full-year growth (not 13 percent); the 13 percent figure applies only to Q4 2025 quarterly revenue.

Corrected; now readsYahoo ConnectID reaches more than 205 million authenticated U.S. users. Publisher domain coverage grew from approximately 50,000 (April 2024) to approximately 55,000 (late 2024) — "approximately 55,000" is accurate as of late 2024 but overstates what Yahoo's April 2024 press release actually said ("nearly 50,000").

Corrected; now readsThe Roku/Confidential Publisher Match element is wrong by two years. Roku was named the first CTV publisher in Google's Confidential Publisher Match at Google's NewFronts 2026 (March 23, 2026), not NewFronts 2024. The product did not exist at NewFronts 2024; what Google announced at NewFronts 2024 was its intention to contribute the PAIR protocol to IAB Tech Lab.

Corrected; now readsSamsung Ads operates from a base of 77 million active Samsung Smart TVs (per Samsung's own 2025 materials) drawing on proprietary ACR data. An ARF Dash Study, reported August 27-28, 2025 in MediaPost, placed Samsung at 45.54 million advertising-reach smart TV households vs. Amazon Prime Video at 45.48 million. The "60%+ U.S.

Corrected; now readsThe Council of Trent met in three discontinuous periods — 1545–47 (Paul III, 13 sessions), 1551–52 (Julius III, 6 sessions), and 1562–63 (Pius IV, 6 sessions, peak ~280 bishops) — working simultaneously on doctrinal definition and disciplinary reform.

Corrected; now readsThe Society of Jesus, founded by Ignatius of Loyola and approved in 1540, remains a living institution and is, per the Annuario Pontificio 2026 / Annuarium Statisticum Ecclesiae 2024, the largest men's religious order in the Catholic Church, with about 18,711 members worldwide.

Corrected; now readsFrom about 1965 Mass attendance in majority-Catholic nations fell roughly four percentage points per decade more than in Protestant countries (cumulating to ~20 points by 2010, per an NBER study), and French priests fell from ~41,000 in the mid-1960s to ~15,000 by 2010 - but whether Vatican II caused this or whether secularization, affluence, and Humanae Vitae did is genuinely contested.

Corrected; now readsPer the 2025 Annuario Pontificio / 2023 Annuarium Statisticum Ecclesiae, there were about 1.406 billion Catholics; Africa (~281 million, ~20%) grew ~3.3% with priests rising ~2.7%, the Americas hold ~47.8%, Asia ~11%, while Europe's share and priest numbers declined (priests -1.6%) - a Global-South demographic shift against a still-Rome-centralized institution.

Corrected; now readsThe major men's institutes are in decline — Jesuits ~13,768 (down from ~36,000 in 1965), Salesians ~13,694 (positioned to overtake the Jesuits as the largest), Franciscans OFM ~15,130, Capuchins ~11,092, Benedictines ~7,558; religious sisters fell to 589,423 in 2023 (−9,805 in a year, driven by Europe and the Americas).

Corrected; now readsAs of mid-2026 Fiducia Supplicans remains in force, applied unevenly by geography (parts of the West vs. effectively a dead letter in Africa); Pope Leo XIV has made no move to repeal or expand it — a status partly inferred from the absence of any contrary act.

Corrected; now readsThe German Synodal Way's successor body, a 'Synodal Conference' of 27 diocesan bishops and 27 ZdK lay representatives with equal voting rights, had its statutes approved by the ZdK (late 2025) and adopted by the German Bishops' Conference plenary on Feb. 24, 2026; the statutes were submitted to Rome for recognitio ad experimentum, with Vatican approval not granted as of early 2026.

Corrected; now readsThe Catholic Church is a communion of 24 churches sui iuris — the Latin Church plus 23 Eastern Catholic churches across five ritual families — the largest being the Ukrainian Greek Catholic Church (~5.5 million, Byzantine) and the Syro-Malabar Church of Kerala (~4.5 million, East Syriac), governed under a separate Code of Canons of the Eastern Churches (1990).

Corrected; now readsPope Leo XIV traveled to Türkiye for the 1,700th anniversary of Nicaea and signed a joint declaration with Ecumenical Patriarch Bartholomew I in Istanbul on 29 November 2025, pledging to continue toward 'full communion between our sister Churches,' rejecting 'absorption or domination,' and proposing a common Easter date.

Corrected; now readsA Pew Research Center study released 21 January 2026 (spring 2024 survey, 6,200+ adults, six countries) found Catholic shares down 9+ points in every country since 2013-14: Argentina 71%→58%, Brazil and Chile to 46%, Colombia down 19 points to 60%, Mexico to 67%, Peru steady at 67%.

Corrected; now readsUS dioceses are consolidating parishes on a large scale driven by priest shortage, falling attendance, aging congregations, and abuse-litigation finances: >2,000 Northeast/Rust-Belt parishes closed in two decades; Baltimore merging 61 parishes into 21, Peoria 156 to 75 by 2027, Buffalo ~34% of parishes.

Corrected; now readsInsurers are withdrawing from or litigating out of abuse coverage: a New York appeals court ruled (April 23, 2024) that insurers need not pay for the Archdiocese of New York's claims tied to systemic failures, and carriers have broadly narrowed terms and excluded decades-old incidents, pushing dioceses toward self-funding.

Corrected; now readsPew's 2023-24 Religious Landscape Study finds that for every 1 person who joins the Catholic Church in the US, about 8.4 leave — the worst net retention of any major US tradition; 45% of former Catholics cite abuse scandals as a reason.

Corrected; now readsThe number of US Catholic priests fell more than 40% from 1970 to 2024 (59,192 to 33,589, CARA), with a median priest age over 70 and ~1 in 4 US priests foreign-born.

Corrected; now readsIn 2008 the Obama campaign matched its national voter file (VoteBuilder/VAN) against online interactions, an email list of roughly 13 million, and Facebook friend-graph data — an early voter-file/digital-identity fusion. (Earlier '190M voters / 23M emails' figures were dropped as unsupported by the cited sources.)

Corrected; now readsIn 2018 Telaria reported CTV at ~30% of political ad spend on its platform (peak 37% in September), up from ~1% in 2016. (A widely cited SpotX '+900%' figure belongs to the 2020 cycle — since April 2020 — not 2018, and was removed.)

Corrected; now readsIn 2020 the Biden campaign reserved a record ~$60M in digital media (within a ~$280M buy), centered on CTV but also covering YouTube, ESPN, streaming audio/podcasts, and online gaming — not a CTV-exclusive figure.

Corrected; now reads2024 political CTV spend was projected at ~$1.5–1.56B mid-cycle but came in at ~$2.3B actual (AdImpact final) — about 21% of the record ~$11.1B cycle and the second-largest medium after broadcast.

Corrected; now readsA September 2020 Mozilla Foundation review called streaming political-ad transparency a 'Wild West'; CTV's 'public file' is a patchwork of voluntary platform archives of uneven completeness. (Date corrected from 2024 to 2020.)

Corrected; now readsThe November 2017 Comscore–L2 partnership productized a national-voter-file-to-TV-measurement fusion — an early example, not the first (Rentrak/i360 2014, Rentrak/Targeted Victory 2014, and Nielsen/Sinclair 2015 predate it).

Corrected; now readsMadhive launched a local CTV political marketplace (Feb 27, 2024) across 400+ congressional districts and all 210 DMAs, with a political data layer drawing on Data Trust, TargetSmart, Experian, and L2 (corrected from a mis-listed 'Tunnl').

Corrected; now readsState comprehensive privacy laws leave voter/political data outside their reach: the CCPA excludes voter data as 'publicly available' records and campaigns (527s) as non-'businesses' — not via an express CCPA political carve-out, which is a feature of successor laws. Net effect: voter targeting proceeds without commercial consent mechanisms.

Corrected; now readsWashington's Fair Campaign Practices Act ad-disclosure mandate survived Meta's First Amendment challenge: a $35.2M penalty for 822 violations, affirmed by the Washington Supreme Court (per curiam, June 18, 2026), with six of nine justices affirming liability. (Date corrected from June 19; plurality nuance added.)

Corrected; now readsAs of mid-2026 roughly 30 states (Public Citizen — Maryland the 30th, May 13, 2026) have enacted laws regulating deepfakes/AI in election communications; California's AB 2839 was struck down in full (Kohls v. Bonta, E.D. Cal., Aug 29, 2025), not partially.

Corrected; now readsAdImpact's June-2026 update projects 2026 political CTV spend at ~$2.7B (revised up from $2.48–2.5B in September 2025) and broadcast at ~$5.6B (revised up, not down) — CTV remains the only major channel projected to grow over the 2024 presidential cycle.

Corrected; now readsPolitical CTV ad spend grew roughly 800% from 2020 (~$260M) to 2024 (~$2.34B actual); the widely cited '~600%' measures 2020 to eMarketer's 2024 forecast of ~$1.56B. 2022 was ~$1.09B; CTV's share of political programmatic budgets rose 19%→30% (2020→2022).

Corrected; now readsBorrell's local-TV revenue data show political advertising rising from 9% of local-station ad revenue in 2008 to 17% in 2012. (An earlier '$1.1B broadcast vs. $200M local cable' 2012-presidential figure was removed as uncorroborated by the cited source.)

Corrected; now readsThe pre-agentic stack standardized the message format but left DSP-SSP-data integrations hand-built, producing an 'ad tech tax' in which roughly one-quarter of programmatic ad spend was effectively wasted in the supply chain (ANA/TAG TrustNet, 2024).

Corrected; now readsPremium CTV inventory is transacted heavily via curated private deals: PMPs account for roughly 47% of CTV spend versus roughly 15% via the open exchange, programmatic guaranteed made up about a quarter of CTV buys in 2024, and over 90% of US CTV ad spend in 2025 ran through programmatic pipes.

Corrected; now readsDisney's DRAX (Disney Real-Time Ad Exchange) biddable/self-service platform reported biddable transactions up 70% since launch and 8,000+ video advertisers at the CES 2026 cycle, with a goal to automate over half its ad business within five years and Disney Select first-party audiences activatable programmatically via DRAX (clean room announced October 2025).

Corrected; now readsRoku is exploring strategic options including a full sale per Reuters (June 12, 2026); market cap ~$19.4B; shares jumped ~20% on the news; Reuters cited discussions with at least one US media company, no final decisions — a live going-concern flag for Roku's ad/identity platform and any agentic posture.

Corrected; now readsShoppable CTV is positioned in 2026 as a critical retail-media channel (platforms incl. Roku, Amazon Freevee, YouTube TV, Hulu, Peacock, Tubi, Samsung TV+, Netflix Ads, Disney+, Paramount+; Pinterest links Pins to shoppable actions in Roku after acquiring tvScientific in Dec 2025); Google documents shoppable CTV and agentic commerce together — but autonomous agent purchase from a CTV ad is aspirational-to-piloted, not a shipped production flow.

Corrected; now readsOne analyst source claims disputes on agent-initiated transactions run roughly 2.4x the rate of comparable human card-not-present transactions, weighted toward 'did not authorize' and 'not as described' rather than fraud — but this figure is single-source and could not be independently corroborated.

Corrected; now readsAs of 2026 shoppable CTV (T-commerce) still completes purchases via QR code or send-to-phone hand-off to mobile checkout, not via a TV-resident agent completing an AP2 Payment Mandate; an end-to-end agentic CTV purchase is announced/aspirational, not a shipping product in any source reviewed.

Corrected; now readsIAB Tech Lab CEO Anthony Katsur has called AdCP 'deeply flawed' and 'resource-consuming,' said the industry needn't have 'another industry trade group,' and that AdCP 'went far in trying to standardize something that's unproven' vs. IAB's OpenRTB-based ARTF.

Corrected; now readsMiQ announced a Titan OS partnership on Nov 18, 2025 (live Q1 2026), becoming the first to license Titan OS log-level smart-TV data and the only managed-service partner with direct access across Europe, feeding into Sigma.

Corrected; now readsIn January 2023 Samba TV transitioned its US managed-media-sales business to MiQ and became MiQ's multi-year household-level ACR data and measurement supplier; Samba ACR is chipset-integrated across 24 top smart-TV brands.

Corrected; now readsAt acquisition, Pathlabs retained its brand and operates as an independent business unit under MiQ, led by CEO William Lapointe reporting to US co-CEOs Neer and Madorsky.

Corrected; now readsMiQ Sigma's marketed integrations include retail/CPG data partners NielsenIQ, Numerator and Circana, alongside TTD, DV360, AWS, Samba, Experian and Databricks.

Corrected; now readsNo credible evidence exists that Bridgepoint has exited MiQ as of June 2026; an aggregator 'Rocket Lab' March 2026 deal entry appears to be a data error conflating the late-March 2026 Adsmovil acquisition with an unrelated entity.

Corrected; now readsMiQ's India/APAC presence is organic (no comparable acquisition): MiQ Digital India had ~413 employees in 2025, runs a Bengaluru Centre of Excellence for engineering/data science where MiQ says Sigma was built, and APAC is led from Singapore by Jason Scott.

Corrected; now readsMiQ states its Sigma Trading Agent reduces cross-DSP/SSP audience activation from ~6 hours to ~30 seconds — a self-reported, single-source vendor claim attributed to a MiQ sales lead with no disclosed methodology, sample, or independent verification.

12 categories not 11

Now readsGARM launched June 2019; Sept 2020 framework defined 11 categories (+Misinformation 2021 = 12), a floor, and low/med/high tiers.

MRC viewability does not transfer

Now readsMRC 50%/2s viewability is a web/in-browser construct; mechanics do not transfer cleanly to full-screen CTV.

IAS now private

Now readsIAS taken private by Novacap, completed Dec 23 2025; the 'public-market duopoly' framing is now partly stale (DV public, IAS private).

dismissal is 2026, with appeal

Now readsX antitrust suit dismissed with prejudice March 26-27 2026 (not 2025); X appealed to 5th Circuit.

TCF v2.3; narrowed not outlawed

Now readsEU CTV brand safety gated by consent law: TCF litigation (Jan 7 2026 partial IAB Europe win; TCF v2.3 mandatory Feb 2026) + DSA.

DPDP Rules notified, not pending

Now readsIndia DPDP Rules notified Nov 13 2025; full compliance May 13 2027; near-categorical ban on targeted ads to under-18s.

Carlyle letter-count attribution: 173 letters = Norton's 1883 first edition, not Slater's definitive 1964 edition; total above 190 - corrected across the corpus

Now readsEmerson–Carlyle: forty-year correspondence (173 letters in Norton's 1883 first ed.); American agent

Lincoln-meeting date -> early February 1862 (c. Feb 1-4), not 'February 28' (doc 06)

Carlyle correspondence: 173 (89/84) = Norton 1883 first ed.; Slater 1964 is the definitive fuller edition; total >190 (doc 05, Key_Claims 5.1, Sources)

Harvard Board of Overseers term -> 1867-1879 per the Historical Register of Harvard (doc 07, Key_Claims 3.6/7.3); LL.D. firmly 1866

William James 1903 centenary: 'an address' as one of several speakers, not necessarily 'the principal address' (doc 08)

marriage date 1935->1939 (Doc 01)

Centre founding 1962->1963 (Doc 06)

'medium is the mother' article misattribution -> John Dennis Anderson (Doc 01)

'we shape our tools' de-attributed from McLuhan (Doc 08)

The verification panel refuted this claim as first written; it was corrected before publication. That run recorded the claim ID but not the wording it replaced.

Now readsChesterton triggered McLuhan's conversion; he was received into the Catholic Church on 25 March 1937.

Amazon '~5x O&O margin' figure not in the cited Digiday source — removed; qualitative self-preferencing retained.

Yahoo 'no REDS-equivalent log feed' wrong — the partner-gated Yahoo Measurement Feed exists; reframed as conditional, not absent.

Yahoo 'first DSP to give full LLD to ANA' — superlative true only for IAB Data Transparency Labels; corrected to 'now provides'.

ADM audiences 'pushed to The Trade Desk' unconfirmed in cited sources — hedged as unverified.

$26.8B is a Q2 2025 figure, not Q3; the Q3 '21.8% over $123B' footnote framing was an author reconstruction — reattributed.

'~7 poems published in lifetime' (0-3)

'lifelong white-dress recluse rarely seen outside' popular image (0-3)

'over 1,000 poems' as the total (0-3)

'died at 55 in the house where she was raised' as stated (1-2)

Corrected; now readsAPNIC reached its last-/8 austerity phase on 15 April 2011; RIPE NCC reached its last /8 on 14 September 2012 and full exhaustion on 25 November 2019; LACNIC entered exhaustion on 10 June 2014; ARIN's free pool hit zero on 24 September 2015; AFRINIC entered IPv4 Exhaustion (Soft Landing) Phase 1 on 31 March 2017 (announced 2 April 2017), with Phase 2 following on 13 January 2020.

Corrected; now readsLegacy (pre-RIR) allocations total roughly 82 /8-equivalents — about 37% of routable IPv4 space (Richter et al. 2015) — including the 40–49 single-organization Class A /8s of the classful era (RFC 1466 records 49 by May 1992), free, contract-less, and overwhelmingly to US institutions; the legacy space then became the principal supply side of the transfer market (e.g., MIT's reported 2017 sale of ~8 million addresses from 18.0.0.0/8 to Amazon).

Corrected; now readsA CGNAT vendor worked example (6WIND) puts the cost of serving 8,000 subscribers via CGNAT at ~$31k TCO ($6k for 127 public IPs, ~$5k hardware, ~$20k licensing; 63 subscribers per public IP) versus ~$392k to buy 8,000 addresses at the then-current $49/IP; broker commentary (Prefix Broker) counters that CGNAT is “not always more cost-effective” once carrier-grade hardware and logging equipment are priced, and that outright purchase is “generally equally expensive.

Corrected; now readsSpain measures ~10% and Italy ~17% IPv6 adoption — among the lowest in Western Europe — while fixed-line address sharing is widespread: in Spain via MásOrange (MásMóvil/Yoigo/Pepephone), Digi, Finetwork and Orange/Jazztel (DS-Lite), with incumbent Movistar assigning public IPv4 on fixed fiber and CGNAT-ing only mobile; in Italy via Sky WiFi's MAP-T (~16:1 sharing), with TIM fixed-line CGNAT practitioner-reported but not regulator-verified.

Corrected; now readsOpenRTB's tmax field caps how long an exchange waits for a bid including internet latency; SSPs typically set 80–120ms (Google historically ~100ms), a buyer responding late is dropped, and with 25–30 DSPs per auction engineers target ~70ms to protect tail latency — making cross-internet round-trip a structural 'latency tax.

Corrected; now readsHeader bidding — placing JavaScript in the page header so a publisher solicits simultaneous bids from multiple SSPs before calling the ad server — emerged in 2014 and was named/popularized by a 2015 AdExchanger article, replacing the sequential 'waterfall' and breaking Google's 'last look' advantage.

Corrected; now readsPublic-cloud egress fees (≈$0.09/GB AWS internet egress, ≈$0.12/GB GCP premium, $0.01/GB cross-AZ; ~6–12% of cloud bills) make high-volume remote bidding costly; because each bid request evaluated carries a real per-request cost, DSPs impose a queries-per-second (QPS) ceiling and historically processed only ~20–30% of the available bidstream.

Corrected; now readsTikTok launched TikTok TV apps in 2020–2021 across Fire TV, Google/Android TV, Samsung, and LG; reporting characterizes the earlier CTV app as having failed to gain traction.

Corrected; now readsThe data clean room — a neutral environment where two parties match audiences and exposure-to-conversion data without seeing each other's raw user-level records — is the mechanism that makes social-to-CTV legally and commercially possible (brand loads customer list, publisher loads exposure logs, room returns only aggregated overlap/reach/incremental conversions).

Corrected; now readsAmazon's ad business grew 24% YoY to $17.2B in Q1 2026 (>$70B TTM); Amazon DSP holds ~14% DSP share, is the fastest-growing top-five DSP (~24% YoY), and has direct CTV deals with Netflix, Roku (80M+ households), Disney, Samsung, Tubi and Comcast.

narrowed from an achieved threshold to a stated trajectory, and re-cited to the Netflix Q1 2026 earnings transcript

Now readsNetflix completed its transition to a fully in-house Netflix Ads Suite by 2025, and told investors programmatic buying was on track to exceed 50% of its non-live advertising business, with programmatic inventory volume up more than a third heading into Q2 2026.

sourcing upgraded from a single trade blog to Linux Foundation and Anthropic primary announcements

Now readsMCP was donated by Anthropic to a Linux-Foundation-hosted Agentic AI Foundation in December 2025, co-founded with Block and OpenAI.

confirmed the Google remedies ruling has NOT issued; retained as a live forecast trigger

Now readsJudge Brinkema ruled Google illegally monopolized publisher ad-server and ad-exchange markets; remedy closing arguments concluded Nov 21 2025; she is weighing structural (AdX divestiture) vs behavioral-only remedies, with no AdX buyer identified; most observers expected a ruling by mid-2026.

overclaimed 'mainstream' for polemical-reading thesis — corrected to 'major scholarly position'

Now readsA major scholarly position (not unanimous) reads the Fourth Gospel's strenuous subordination of John as polemical response to a persistent Baptist-venerating community.

overclaimed Mandaean 'only surviving community' without geographic precision — corrected

Now readsMandaeans are the only surviving community for whom John is central (as chief prophet/teacher, not forerunner), historically in southern Iraq and Khuzestan.

Corrected; now readsThe Centre did NOT close in 1984; Gil Henriques directed it 1981-1985 and the University of Geneva closed it c.1985 (one source says 1986).