[
  {
    "id": "H1",
    "type": "hypothesis",
    "statement": "No scored product publishes a numerical US match rate with a stated denominator, matched population and matching method.",
    "population": "Every product ultimately admitted to the US IQ, across cohorts A–D. Search every otherwise eligible US candidate from the universe sweep before the IQ admission decision, including products with no advertised rate. Join results to admission by stable product ID at Stage 5. Exclude Samba from both frames. Search the named service, aliases and on-product US cases through the cutoff. Do not pool aliases or count a corporate rate that cannot be attributed to the service.",
    "comparator": "For each rate, distinguish (1) the eligible input base and unit, (2) what counts as a successful match, including output population and unit, and (3) how the input was linked to that output. A documented algorithm class and counting rule suffice for method. A numeric sample size is recorded if available but is not required when the denominator is unambiguously defined. Audit evidence, error rates, range, dataset size and label overlap are separate fields.",
    "outcome": "One row per distinct product, metric, case, period and geography. Log the rate as published, numerator and denominator definitions, sample counts if given, exclusions, geographic scope, measurement period or reporting date, source, sponsor and method. A vendor-published example can meet the disclosure test. A second source may supply a definition only through an explicit link to the same metric, version and population. Show products with no numerical claim and unknowns separately. Repeat the test on the full eligibility frame and report any excluded counterexample by name without changing the admitted-frame verdict.",
    "rule_supported": "Within the screened public record, support if there is at least one admitted product with a numeric rate, the product documentation, help/API references, case-study/release searches and alias searches have been completed for every admitted product, and zero qualifying rates were found. No material candidate may remain inaccessible or unclassifiable. Missing fields in a readable claim are evidence about disclosure. This verdict does not establish that an audit is impossible.",
    "rule_contradicted": "One verified publication meeting all three components for any admitted product contradicts H1 immediately, even if other files are blocked or that case is narrow. A seller's publication can establish what it disclosed. Its rate remains attributed to the seller.",
    "rule_unresolved": "Absent a verified counterexample, return unresolved if no admitted product publishes any numeric rate in the screen, a registered search lane is unfinished, a plausible qualifying record cannot be opened after fallbacks, or a material product/version/geography linkage cannot be resolved.",
    "registered_date": "2026-11-11",
    "cutoff": "2026-11-11"
  },
  {
    "id": "H2",
    "type": "hypothesis",
    "statement": "No independent public test reports the accuracy of a scored graph's US person or household links against labelled truth.",
    "population": "Every admitted US product whose sold service creates or resolves person or household links, including a cohort A service only if it sells that resolution task. Search all such US candidates before admission; repeat the analysis for the full eligibility frame as a sensitivity check. Cover public studies available by the cutoff, including historical tests if the tested service/version is identified and its continuity or end is recorded. Report old tests as old tests.",
    "comparator": "A qualifying test uses independently established person or household labels, a defined test population and denominator, and a link-quality metric for that same unit, such as precision, recall, false-match rate or a defined classification accuracy. Keep observed agreement, reach, deterministic-input share, match yield and campaign lift in separate fields. Self-confirming graph labels do not establish truth. Record funding, author interests and access to data. Peer review alone does not establish independence.",
    "outcome": "One row per original study, product/version, US sample or separable US stratum, metric and unit. Record label source, sampled positives/negatives, design, period, holdout or other separation from graph construction, numerator, denominator, uncertainty if stated, exclusions and funding. Record unknowns. Separate independent, financially interested/commissioned, and funding-unclear designs. Report admission-sensitive additions separately.",
    "rule_supported": "Support only as a bounded disclosure claim if there is at least one admitted qualifying graph, every such graph/alias has been searched in a scholarly index, an independent publisher or proceedings index and the product's evidence/library pages with citation follow-up, and no qualifying independent test is found. No material candidate may remain unclassified due to access, design or independence.",
    "rule_contradicted": "One verified independent test satisfying the comparator for one admitted product and US unit contradicts H2, regardless of the value reported, the breadth of the sample or missing records elsewhere. Report the covered product/version, period and limits.",
    "rule_unresolved": "Absent a qualifying test, return unresolved if the eligible admitted population is empty, a required search is missing, or a plausible candidate's identity, US scope, labels, denominator, methods or financial independence cannot be established. A complete design that plainly measures a different quantity is an ineligible test, not an unknown.",
    "registered_date": "2026-11-11",
    "cutoff": "2026-11-11"
  },
  {
    "id": "H3",
    "type": "hypothesis",
    "statement": "More than half of in-scope interoperable IDs document how a user's choice reaches a recipient of the ID and changes what the recipient may do in at least one applicable jurisdiction.",
    "population": "All current US-available interoperable-ID products in cohort A that pass the buying-task screen at the cutoff, whether or not they later qualify for the IQ. Count one row per distinct product. Code US, EU and UK separately and only where the named service is offered. One documented applicable jurisdiction qualifies the product. Print jurisdiction-specific counts so that an EU path cannot be read as US coverage. A service's EU disclosure cannot establish that an EU-specific variant is itself sold in the US.",
    "comparator": "Trace one documented route for the same product/version, jurisdiction and use: collection and choice at origin, binding of that choice to ID issuance or use, transport in or alongside the bid request, recipient interpretation, and a specified consequence of refusal or withdrawal. Record signal family, version, purpose, issuer/recipient and expiry or renewal if disclosed. An OpenRTB field, TCF or GPP support, login or hashed email alone does not establish the path. ID suppression at origin with no recipient-side choice path is coded separately.",
    "outcome": "For each product, code identity input (for example authenticated, observed or inferred), stated processing basis and separate user choice, transport, recipient action and evidence of deployment. A product-specific implementer specification can document the required path without proving enforcement. Mark yes for a complete path, no for a completed public screen without a complete path, unknown for a material access, applicability or attribution gap. Report Y, N and U and their sum T. Search first-party documentation, issuer specs and recipient/adapter documentation, including restrictive settings. Define E as distinct screened candidates whose eligibility remains unresolved after checking the named service, period, geography and buying task. Publish T and E separately. For sensitivity only, include all E as possible eligible rows. If T+E is positive, report an expanded interval from positive/(T+E) to (positive+U+E)/(T+E), using the positive count in this test. Preserve the primary verdict on confirmed eligible T, disclose if the expanded interval crosses one half, and show contrary as well as supportive product cases where observed.",
    "rule_supported": "If T is positive and Y/T exceeds one half, support the claim about published paths. Unknown rows remain in T. Report the lower and upper possible shares Y/T and (Y+U)/T, and separate prescribed behaviour from observed behaviour.",
    "rule_contradicted": "If T is positive and (Y+U)/T is at most one half, contradict. An exact half contradicts a claim of more than half.",
    "rule_unresolved": "If T is zero, or the interval from Y/T to (Y+U)/T spans the decision boundary (lower at or below one half and upper above it), return unresolved. A wider market-wide inference is outside the test.",
    "registered_date": "2026-11-11",
    "cutoff": "2026-11-11"
  },
  {
    "id": "H4",
    "type": "hypothesis",
    "statement": "For more than half of the scored US web products in the registered comparison, public integration instructions increased the role of third-party-cookie ID sync after Chrome's decision to retain user choice.",
    "population": "From every scored product across A–D that uses browser-based identity on the US open web, include products independently established as offering the same buying task both before and after the Chrome index event. Log entrants, exits and unproved continuity separately. Form the frame via the universe sweep, before examining directions of change. For products whose eligibility cannot be resolved, also print a sensitivity interval that treats each as an unknown eligible product.",
    "comparator": "Register the index event as Google's dated official Chrome decision in 2025 addressing retention of the existing third-party-cookie choice and disposition of a proposed standalone choice prompt. Verify the document and exact date before coding. Compare the latest dated version of an equivalent US Chrome integration guide or changelog in the 365 days before the event with the latest comparable version after it and through the cutoff. A dated archive capture may establish the version visible that day. Compare the same product, workflow, role and deployment context. Independently date each relevant Privacy Sandbox API announcement, deprecation, removal and effective platform/version. Run separate paired comparisons around each verified API retirement announcement and each effective removal applicable to that product. For each event use the latest comparable pre-event and post-event versions under the same rules. Exclude the event day unless timestamps fix the order. Prefer the issuer's dated version for the same guide over archive copies; if ties persist and imply different codes, mark the pair unknown. If no relevant event is verified, no post-event guide exists, or the removal is still prospective, label that secondary contrast unavailable and give the reason. Show opposing or inconsistent directions by API and event, with no pooling or change to the primary Chrome-choice verdict.",
    "outcome": "On each side record third-party-cookie sync as explicitly unavailable/disabled (0), optional/fallback (1), default/recommended (2) or mandatory for the same workflow (3). Missing wording is unknown. First-party storage, generic 'cookies', a browser's ability to set cookies and server transfer alone do not establish third-party ID sync. Use a single documented default workflow or, if several exist, the issuer-designated main US integration. If none can be selected without discretion, code unknown. For T eligible products tally increases I, flat/decreases and unknown pairs U. Report old and new dates, exact guidance, direction and API timeline. A documentation shift alone cannot establish policy causation or actual cookie traffic. Define E as distinct screened candidates whose eligibility remains unresolved after checking the named service, period, geography and buying task. Publish T and E separately. For sensitivity only, include all E as possible eligible rows. If T+E is positive, report an expanded interval from positive/(T+E) to (positive+U+E)/(T+E), using the positive count in this test. Preserve the primary verdict on confirmed eligible T, disclose if the expanded interval crosses one half, and show contrary as well as supportive product cases where observed.",
    "rule_supported": "If the primary event and its date are verified, T is positive and I/T exceeds one half, support. Use all eligible products in T, including unpaired ones, and show the sensitivity to unproved eligibility.",
    "rule_contradicted": "If the primary event and its date are verified, T is positive and (I+U)/T is at most one half, contradict. An exact half contradicts. Report flat and decreasing pairs separately.",
    "rule_unresolved": "Return unresolved if the primary policy/date cannot be verified, T is zero, or missing or incomparable pairs leave the majority undecided. Historical page silence, a changed page topic or a site-wide redesign without like-for-like versions is unknown. If the sensitivity frame reverses or weakens a decisive primary result, disclose the dependence without silently changing its population.",
    "registered_date": "2026-11-11",
    "cutoff": "2026-11-11"
  },
  {
    "id": "H5",
    "type": "hypothesis",
    "statement": "More than half of US CTV household-graph products document both the use of an IP address to assign a device to a household and a timed rule for refreshing or expiring that link.",
    "population": "Every live or beta US cohort D product in the universe that sells household identity for CTV, regardless of its evidence depth or final IQ admission. Exclude Samba. One purchased service per row, distinct from its data feeds and aliases. Record ACR observations, CTV device IDs, residential-IP filtering and other inputs separately.",
    "comparator": "Require two linked disclosures for the same service and edge: the role of IP in household assignment (required, optional or fallback qualifies if explicitly used for the assignment), and a numerical edge refresh, expiry interval or elapsed-time decay rule. The clock must apply to the IP-to-household mapping or its device-to-household edge. An unrelated cookie TTL, file delivery cadence, graph-size count or statement that a feed is updated daily does not qualify.",
    "outcome": "Search provider methods, manuals, data dictionaries and documented integrations. Code both components independently for each edge, including IP as sole input versus in combination. Per product mark yes if at least one edge meets both, no if a completed screen finds no joint disclosure, unknown for material gaps in access or product mapping. Show Y, N, U, and T=Y+N+U. Also report products with only IP, only a timed rule and neither. This is a test of disclosure, not of actual household stability. Define E as distinct screened candidates whose eligibility remains unresolved after checking the named service, period, geography and buying task. Publish T and E separately. For sensitivity only, include all E as possible eligible rows. If T+E is positive, report an expanded interval from positive/(T+E) to (positive+U+E)/(T+E), using the positive count in this test. Preserve the primary verdict on confirmed eligible T, disclose if the expanded interval crosses one half, and show contrary as well as supportive product cases where observed.",
    "rule_supported": "With T positive, support if Y/T exceeds one half even when every unknown is treated as nonqualifying.",
    "rule_contradicted": "With T positive, contradict if (Y+U)/T is at most one half even when every unknown qualifies. An exact half contradicts.",
    "rule_unresolved": "Return unresolved if T is zero or the bounds straddle the majority threshold (lower at or below one half and upper above it). A blocked candidate method, an ambiguous interval or uncertain product scope remains unknown if it could change the joint disclosure.",
    "registered_date": "2026-11-11",
    "cutoff": "2026-11-11"
  },
  {
    "id": "H6",
    "type": "hypothesis",
    "statement": "Since the start of 2022, at least one public enforcement outcome involving an identified in-scope identity product has a separately documented change to that product's data or operations.",
    "population": "All identity services in A–D identified by the universe sweep, including listed and historical identity products with documented activity in the period. Include Samba in the separate legal screen only, with its interest flag. Exclude adjacent services listed solely for boundary context. Cover actions from 1 January 2022 through the cutoff by the FTC, US state attorneys general including DC, EU data protection authorities and the UK ICO where the service was sold. Build a fixed entity/alias and authority checklist before extracting actions. Screen federal and state public enforcement indexes, European and national authority records, ICO records and linked court dockets. Record jurisdiction and coverage gaps.",
    "comparator": "Require two propositions: (a) an entered operative order, consent decree, final administrative decision or judgment identifies the product or unambiguously identifies its service through the respondent and activity; (b) a distinct dated passage or exhibit documents an implemented change to collection, retention, deletion, onward transfer, linkage or opt-out under that disposition. Prefer a regulator compliance record, filed sworn certification or independent audit. A seller statement must be attributed and its implementation corroborated by such a record. A requirement and proof of compliance may be in one docket if they are distinct records. A stayed, vacated or reversed disposition is coded with its posture and cannot alone qualify.",
    "outcome": "One row per docket or authority action with linked records by date. Code inquiry, allegation, proposed settlement, procedural ruling, operative outcome, appeal, voluntary dismissal, specific obligation, effective date and implementation evidence separately. Record vendor response and outstanding duties. One authority's jurisdiction and one product's obligation cannot be extended to all products of the respondent. Report policy changes whose connection to the action is unproved. This table sits outside the IQ.",
    "rule_supported": "One verified action-product pair meeting both propositions supports H6 immediately, regardless of incomplete records elsewhere. Name the product, jurisdiction, obligation, implemented change and evidence date.",
    "rule_contradicted": "Contradict the public-record claim only if the complete registered authority-by-entity screen is logged, all plausible dockets and implementation leads have been resolved, and there is no qualifying pair. State the exact public sources, jurisdictions and dates screened; the result cannot establish that no changes occurred in private.",
    "rule_unresolved": "Absent a verified qualifying pair, return unresolved if a required authority or product lane is incomplete, a plausible docket is blocked, product attribution is ambiguous or a credible implementation lead cannot be settled. An operative order on its own proves an obligation. A complaint, announcement or voluntary dismissal on its own proves neither the underlying allegation nor implementation.",
    "registered_date": "2026-11-11",
    "cutoff": "2026-11-11"
  },
  {
    "id": "H7",
    "type": "hypothesis",
    "statement": "In the public records screened, identity support in agent-driven advertising protocols reaches at most schema representation.",
    "population": "AdCP, AAMP and other publicly versioned agent-buying protocols discovered in the standards and universe screens by the cutoff. Include an additional protocol if its issuer describes agent-mediated buying of advertising and provides a published interface; do not select by identity support. Include product-specific connectors for those protocols. Freeze protocol names, issuers and release lines before coding maturity. Search release histories as well as current versions and report retired evidence as historical. Count products neither by working-group membership nor by author affiliation.",
    "comparator": "Track five states separately for each protocol/version, identity signal and operation: schema representation; implemented connector; demonstrated interoperability; documented production use; demonstrated transaction. An identity signal identifies, resolves or constrains an advertising person, household, device, identifier namespace or publisher audience by its identity basis. An account credential, media-buy ID, research-only synthetic schema example or an audience name with no identity semantics does not establish a beyond-schema state.",
    "outcome": "Record version, field/path, meaning, caller, recipient and evidence for each state. A versioned functioning connector with a verifiable code path that handles an identity field in the buying workflow suffices for state 2; a type definition, unexecuted pseudocode or prose promise does not. For interoperability record two implementations and their trace; for production identify a deployment and evidenced use; for a transaction document the actual buy, identity role and whether sandbox or live. A sandbox identity-linked exchange can establish interoperability. It cannot on its own establish a live transaction. Supporting an audience selection alone does not prove identity crossed a boundary.",
    "rule_supported": "Support as a bounded disclosure conclusion only if at least one identity schema field is verified, every included protocol's specifications, official repositories/releases, connector listings, test records and named implementation/production leads are screened, and none establishes any state 2–5. All credible beyond-schema leads must be accessible and resolved. Scope each observation to its implementation, version and date.",
    "rule_contradicted": "Any one verified in-scope example of any state 2–5 contradicts the schema-only ceiling, even if there is no transaction, other protocols lack a connector or separate records are blocked. Never infer all lower states from a higher state; document each independently.",
    "rule_unresolved": "Without a verified beyond-schema example, return unresolved if no qualifying identity field is established, discovery/coverage is incomplete, an eligible protocol's version or purpose cannot be established, or a credible implementation lead cannot be opened/classified. An unsubstantiated production announcement stays a vendor assertion, with the implementation question open.",
    "registered_date": "2026-11-11",
    "cutoff": "2026-11-11"
  },
  {
    "id": "H8",
    "type": "hypothesis",
    "statement": "More than half of the US graph-based offers marketed as cookie replacements since 2020 reuse a named predecessor's core identity linkage.",
    "population": "Screen all A–D candidate, listed and historical identity services outside Samba for a dated issuer claim from 1 January 2020 through the cutoff that a new, renamed or relaunched US graph-based offering substitutes for third-party-cookie identity. One distinct successor offer per buying task, counted once at its first qualifying launch or relaunch. Determine eligibility from launch material independently of predecessor visibility, graph continuity and IQ admission. Include an eligible offer even if no predecessor is identified. Record launch pitches for non-graph alternatives as a separate descriptive comparison.",
    "comparator": "Compare each successor at launch to the predecessor explicitly identified by technical migration documents, version history, service mapping, filing or an equivalent dated first-party record. Reuse requires an explicit record of continued use of the same underlying person/household/device linkage graph or edge set, plus corroboration via an operational artifact such as a migration map, persistent namespace mapping or edge lineage record. Changed token formats or delivery routes can coexist with graph continuity. The same owner, staff, source-data class or probabilistic/deterministic label cannot by itself establish reuse.",
    "outcome": "Per eligible successor record purchase task, launch date, cookie-replacement statement, predecessor, identity unit, linkage inputs and rules, identifier/edge continuity, and what changed in the collection, rights or activation path. Code R for evidenced reuse, N for a documented independent graph construction or replacement of core links (or a documented first build without inherited links), and U if continuity is indeterminate. A hybrid qualifies for R only for documented retention of the predecessor's primary linkage for that successor's main US task; unresolved mixes stay U. T=R+N+U. Report term counts and examples plus the effect of unknowns. Distinguish continued linkage from a new activation wrapper and from any inference about deliberate relabelling. Define E as distinct screened candidates whose eligibility remains unresolved after checking the named service, period, geography and buying task. Publish T and E separately. For sensitivity only, include all E as possible eligible rows. If T+E is positive, report an expanded interval from positive/(T+E) to (positive+U+E)/(T+E), using the positive count in this test. Preserve the primary verdict on confirmed eligible T, disclose if the expanded interval crosses one half, and show contrary as well as supportive product cases where observed.",
    "rule_supported": "With T positive, support if R/T exceeds one half, regardless of unknowns. Limit the inference to registered US offers and the launch periods examined.",
    "rule_contradicted": "With T positive, contradict if (R+U)/T is at most one half. An exact half contradicts. Document any named product whose lineage differs from the aggregate.",
    "rule_unresolved": "Return unresolved if T is zero or unknown lineages place the lower share at or below one half and the upper share above it. A missing ancestor, shared owner, renamed product or seller's 'new graph' tagline without technical evidence does not resolve lineage.",
    "registered_date": "2026-11-11",
    "cutoff": "2026-11-11"
  }
]
