{
  "published_before_scoring": "2026-11-11",
  "cutoff": "2026-11-11",
  "purpose": "Register the evidence standard for the Identity Quotient. The unit is one named product serving one buying task. The IQ measures how much usable documentation a buyer can inspect in public for that task at the cutoff.",
  "selection_rules": {
    "universe": "Use the seed register, the ten research streams and the dated U sweep to enumerate candidate products, aliases, comparators and historical entries. Record all screened candidates and bounded inclusion or exclusion reasons. This is a reasoned US market sample. EU and UK context is recorded separately where pertinent.",
    "scored_set": "Require a documented in-scope US route, a named live or beta product and at least four counted cells for admission. At the pre-score freeze, establish buyable service, operator, ownership, aliases, cohort or listed/historical disposition and US status. Separate a general graph from an independently purchased CTV service only on product-level evidence. A feature, integration, wrapper, prior name or duplicate sales channel does not create another row. Samba TV, clean rooms, generic curation, CMPs, pure contextual products and unaffiliated specs do not score. Mark Lotame Panorama ID with interest: true wherever its row appears.",
    "deep_profiles": "The six Stage 1 selections and any evidence-driven substitutions are recorded in the brief. Verify the exact product boundary before profiling. No extra score or source grade follows from receiving a profile. A secondary, separately bought service from a profiled operator requires its own product record and its own cells."
  },
  "scale": {
    "0": "the record documents that the capability is absent or has failed the criterion",
    "1": "adjacent or incidental",
    "2": "partial or indirect, for example via a partner",
    "3": "real but thinly evidenced, narrow in scope, or beta",
    "4": "documented and specific but from vendor documentation only, or independently observed but narrow",
    "5": "the capability is documented in operational detail from an official document, filing or independent observation"
  },
  "cell_rules": {
    "scale": "Score the highest satisfied product- and cohort-specific anchor, then cap by the grade of the deciding evidence. Each 5 requires the anchor in operational detail with qualifying A/B evidence for all deciding elements. Numbered levels with 'also' or 'level 4 plus' retain their specified prerequisites. A verified source contradiction on the same route/version must be resolved before choosing an anchor.",
    "zero": "Zero needs affirmative, scoped documentation for the dimension's zero condition. Record what failed, for which output or route and as of when. A search yielding no public document, a legal proceeding, or a generic critique cannot produce zero. A zero may count on a live service; if evidence instead establishes retirement or US unavailability, revisit admission first.",
    "ne": "Write v: ne with a bounded reason and available source IDs for an applicable dimension where no anchor is supported. A blocked page, unknown funding required for an independence claim, a site search with no results or a missing denominator is never by itself proof of zero. Score a lower anchor when its positive conditions are independently satisfied.",
    "na": "The cohort's applicable list is the only source of n/a. Omit masked cells from hand-authored scores. The build assigns n/a. A denominator cannot be changed for an individual product. n/a never enters k, group means, coverage or imputation.",
    "availability": "For each numeric cell establish the capability as live or beta at the cutoff. Distinguish the service's availability from a specific feature's maturity. Announced, unconfirmed or retired features cannot yield numeric cells. An announced, retired or unconfirmed service stays outside admission.",
    "evidence_grade": "A = qualifying product-specific filing, official product record or scoped accreditor record for the exact documented fact. B = independent observation or measurement with a demonstrably disinterested party and fit design. C = product-specific technical documentation, including an interested implementer's interface or technical record. D = product marketing, unsubstantiated press or interested promotional case. A standards body's schema proves what it specifies, a platform policy can establish its own rule, and a filing establishes what it reports. None proves another product's implementation, coverage or accuracy. Vendor-originated reach, accuracy and match rates remain attributed assertions even when quoted elsewhere. Funded tests remain commissioned, with sponsor and design attached. Unknown funding does not establish independence. Unknown-funding or commissioned technical tests may support the non-independent test anchors at most as C; a promotional summary at most as D.",
    "date": "Use the most recent deciding evidence date, YYYY-MM or YYYY-MM-DD. Preserve source publication/effective dates, measurement period and actual access dates separately. An undated page with contemporaneous evidence of its state may use dated access/as-of explicitly in why. When read after the cutoff, establish that the exact conditions were public by the cutoff, using dated versions, archived copies, or other dated evidence. A current undated page alone does not backdate the feature.",
    "verification_status": "Code every numeric cell as verified, qualified (only the corrected wording supports this value) or unverified. Unsupported or inaccessible deciding evidence on recheck yields unverified if no independently reopened alternative supports any anchor. Treat unverified as n/e for admission, group means, k, IQ, coverage and bands.",
    "maturity": "Maturity applies to the actual route and version on which the anchor rests. Publication of a draft, schema, SDK, connector listing or integration announcement alone establishes neither use nor a completed transaction.",
    "product": "Join the exact product or unique stable alias to the register and the cohort's purchased service. Pin geography, route, unit, version if supplied, and effective period in the cell rationale. Only join evidence for the same service, route and compatible time period. Do not combine an input at one reseller with a downstream test at another as a single trace.",
    "same_standard": "Apply the same positive and adverse anchor conditions across rows of a cohort. Keep a dated R1, R2, … register of interpretations that arise in Stage 5; rerun each rule across every eligible row. A rich document library and the decision to profile a vendor confer no extra credit.",
    "grade_caps": "C caps at 4. D caps at 3. Assign the grade that supports the decisive anchor as a whole: use the best qualifying source per necessary component, then the weakest grade across those components. Do not use an A/B source for a background detail to raise a vendor-only component. An unrelated official spec, paid corroboration or repeated release is insufficient.",
    "second_pass": "Stage 5 reopens each deciding source after Stage 3 verification and checks product, scope, value, date, availability, grade, caps, n/e and the mask. Include check.verdict (confirmed, adjusted or unsupported) and check.note. An adjustment gives the proposed and final value, deciding evidence and reason. An unsupported numeric becomes unverified and counts as n/e. Retain all checked cells, including genuine zeros and n/e rationales. Masked dimensions are verified against the mask, without a hand-written n/a cell.",
    "conflicts": "Give priority to the record that directly tests the same feature, version, route, geography and effective period, including a dated superseding fix where supported. Preserve verified adverse and favorable observations. If scope cannot reconcile a genuine conflict over an essential component, fall back only to a lower uncontested anchor; otherwise use n/e and explain.",
    "representative_route": "For route-specific dimensions freeze the main sold service before scoring. Use a provider-designated primary US route. If there is no designated route, use the first alphabetically by stable technical route name among documented US implementations of that buying task. Assign this route before reading its outcomes. For a rate or coverage figure on that route, use the most recent eligible period ending by the cutoff; break ties by widest published in-scope base, then stable source ID. Record conflicting periods, customer cases and alternate routes separately. Do not transfer a case-level percentage to the product's full population.",
    "legal_and_affiliations": "Keep litigation, regulatory actions, allegations, settlement, consent orders, legal compliance, affiliations, standards membership, author interests and generic certification outside IQ. Only separately sourced technical behavior can score a control. A certifier's product-specific technical observation may establish only the behavior it actually tested; general certification never satisfies a product anchor by itself.",
    "freeze": "Registration precedes research. Stage 5 records the actual freeze date, confirms register and masks, then opens cells. Log any later change to a definition, anchor, mask, inclusion rule, threshold or product boundary in the corrections ledger with date, rationale, prior and new treatment and affected rows. Apply a changed rule to all affected rows and disclose any change relative to registration."
  },
  "grade_caps": {
    "C": 4,
    "D": 3
  },
  "inclusion_min_scored": 4,
  "dimensions": [
    {
      "key": "assignment_method",
      "label": "How the output is assigned",
      "short": "Assignment",
      "definition": "Score the method that produces the named service's output. For A, distinguish issuance, encoding and linkage. For B and D, distinguish a person, device and household and state the link direction. Deterministic and inferred methods can earn the same level. Cohort C has a separate ladder for group or inventory assignment.",
      "anchors": {
        "0": "The operator explicitly says that it does not disclose which unit its service assigns or how.",
        "1": "The record identifies the output unit, such as a token, person or household.",
        "2": "It also names an input and the rule family that issues a token or accepts a link.",
        "3": "It also specifies an observable acceptance condition for a match or issuance, such as equal normalized inputs, a stated confidence boundary or an explicit qualifying event.",
        "4": "It also describes an outcome for ambiguous or ineligible inputs and supplies a worked input-to-output example for the same method.",
        "5": "Level 4 plus an independent, product-specific observation or qualifying official product record confirming the acceptance and exception behavior for that version."
      },
      "anchors_by_cohort": {
        "C": {
          "0": "The operator explicitly says it does not disclose the group or inventory unit or its assignment method.",
          "1": "The record identifies the sold group, segment, request or inventory unit.",
          "2": "It also names an input and the rule family used to assign that unit.",
          "3": "It also gives an observable membership, eligibility or model threshold for that unit.",
          "4": "It also explains how overlaps, low confidence or unassigned units are treated and supplies a worked example.",
          "5": "Level 4 plus an independent, product-specific observation or qualifying official product record confirming assignment and at least one exception in that version."
        }
      }
    },
    {
      "key": "input_provenance",
      "label": "Where input signals come from",
      "short": "Inputs",
      "definition": "Trace a production input to the upstream collection event, the supplying party or type of party, and its role in the sold output. A signal is mandatory when the provider identifies it as required for the chosen route. Optional enrichments are listed separately. A method using one input can earn the highest level.",
      "anchors": {
        "0": "The operator explicitly states that the origin of every input to this product is undisclosed or untraceable.",
        "1": "One signal class is identified for this product.",
        "2": "For at least one input the record also identifies the type of collection event or supplier.",
        "3": "For at least one input the record gives both its collection event and supplier type and identifies its role in the output.",
        "4": "The operator identifies the full set of required inputs for the route. Each has a collection event, supplier type and role; mandatory and optional inputs are distinguished.",
        "5": "Level 4 plus a qualifying official product record or independent trace corroborating the lineage of the required inputs for that route."
      }
    },
    {
      "key": "refresh_expiry",
      "label": "Refresh and expiry of the sold output",
      "short": "Freshness",
      "definition": "Score the output's lifecycle. For A, separate token lifetime, renewal and underlying link age. For B/D, treat edge or household membership expiry separately from a file's delivery cadence. For C, use segment membership or inventory eligibility lifetime and reassignment. A versioned per-request computation may have zero persistence; explicit re-evaluation and discard on every request qualify without a stored identity.",
      "anchors": {
        "0": "A product-specific record explicitly establishes indefinite reuse with neither a refresh/re-evaluation rule nor an expiry or invalidation trigger.",
        "1": "The record says that the sold unit is updated, expires or is re-evaluated.",
        "2": "It identifies the trigger or frequency for at least one of those actions.",
        "3": "It gives a numerical interval with units or an unambiguous event boundary and specifies which sold unit the rule changes.",
        "4": "It states both when the unit is refreshed or recomputed and when it ceases to be eligible, including what happens to a stale or unrefreshed output.",
        "5": "Level 4 plus independent observation or a qualifying official product record confirming the stale-output or per-request discard behavior."
      }
    },
    {
      "key": "permission_handling",
      "label": "How an allowed use reaches the recipient",
      "short": "Use controls",
      "definition": "Trace the technical signal or product-specific rule from input to recipient for one documented route and jurisdiction. Examples of states are permitted, refused or missing. Authentication, hashing or a populated consent string alone does not establish permission. A signal-free route qualifies where affirmative evidence defines when the output is usable and what the recipient does with a restricted or ineligible request. Legal claims and opinions never score.",
      "anchors": {
        "0": "The operator or a reproduced observation explicitly establishes that the named handoff ignores the applicable use restriction for the scored route.",
        "1": "The record identifies a product-specific use control or permission signal on this route.",
        "2": "It states which party supplies or evaluates that state and the condition that starts evaluation.",
        "3": "It gives the action taken on at least two distinct input states or explicitly documented eligibility outcomes at the sending service.",
        "4": "It also traces the state or the resulting permitted output to a named receiving implementation and specifies recipient handling of a missing, denied or ineligible case.",
        "5": "Level 4 plus independent or qualifying official product-specific verification of a permitted and a restricted case at the receiving boundary, including the respective output."
      }
    },
    {
      "key": "match_rate_definition",
      "label": "How a reported match rate is defined",
      "short": "Match rate",
      "definition": "A, B and D only. Examine one observed rate on the registered product's main matching or translation task. Match direction, unit, numerator, base, exclusions and period belong to that same rate. A denominator may be specified as an eligibility rule even when counts are withheld. Distinguish all submitted records from eligible or successfully processed records. Score the disclosure, irrespective of how high the rate is. Never reframe an accuracy, reach, fill or conversion rate as a match rate.",
      "anchors": {
        "0": "The provider explicitly states that it does not generate or report a match rate for the sold task.",
        "1": "An observed numeric match rate is attributed to the product and reporting party, but the rate's eligible base is unspecified.",
        "2": "An observed numeric rate identifies its denominator population and unit.",
        "3": "It also defines the numerator, matching direction, and whose inputs were matched, with the observation period or an as-of date for a snapshot.",
        "4": "It also defines the linkage or translation method, de-duplication and treatment of excluded or unmatched units for the same period and population.",
        "5": "Level 4 plus sufficient published counts or a reproducible test record to recompute the rate, with independent or qualifying official product evidence for the deciding calculation."
      }
    },
    {
      "key": "recipient_interface",
      "label": "Recipient's delivery interface",
      "short": "Interface",
      "definition": "A recipient may accept a bidstream ID, resolve an uploaded record, ingest a labelled deal or segment, or use a household link. Evaluate one implemented route's output contract. A standard is evidence of a representation; the presence of a field in an unaffiliated schema earns no credit for product adoption. A documented working file or constrained cohort interface is as eligible as a public API.",
      "anchors": {
        "0": "The operator explicitly states that there is no machine-readable handoff to any recipient for the sold service.",
        "1": "A product-specific implemented handoff identifies the route and its file, API, message or equivalent representation.",
        "2": "It also identifies the transmitted field or key, the output unit and the recipient role.",
        "3": "It also gives a versioned schema or field definition and a concrete valid delivery example for that route.",
        "4": "It also specifies recipient behavior for a valid output and an absent, invalid or expired output.",
        "5": "Level 4 plus an observed independent end-to-end test or qualifying official product implementation record covering a valid and an exception case."
      }
    },
    {
      "key": "destination_coverage",
      "label": "Receiving systems that use the output",
      "short": "Destinations",
      "definition": "Count distinct named downstream system operators that are documented as consuming the exact service in production at the cutoff. One operator counts once across regions, brands, endpoints or middleware. A DSP, measurement provider, data platform or publisher can qualify when the service's output is actually used. Count only recipients on a documented US-supported route. For ties at a cutoff use the inclusive threshold. A logo, catalog listing without implementation or self-reported partner total is insufficient.",
      "anchors": {
        "0": "A complete product statement explicitly establishes zero external production recipients or that every external route is disabled.",
        "1": "One distinct recipient with documented production consumption.",
        "2": "Two distinct recipients with documented production consumption.",
        "3": "Three distinct recipients with documented production consumption.",
        "4": "Four distinct recipients with documented production consumption.",
        "5": "Five or more distinct recipients with documented production consumption. Corroboration adequate for grade A or B must establish at least five for an uncapped 5."
      }
    },
    {
      "key": "us_coverage_measure",
      "label": "US coverage with a stated base",
      "short": "US coverage",
      "definition": "Audit a product-specific US coverage measure for a single disclosed period or snapshot, available by the cutoff. A/B/D may use eligible unique persons, households, tokens or devices, explicitly labelled. For C use eligible groups, publishers, requests or inventory units. Supply the measurement window or snapshot, how the unit is counted, and whose eligible universe supplies the denominator. A case-level statistic stays case-level. Keep gross events and uniques, observed and modelled, sample and market, separate. Larger reported volume earns no extra credit.",
      "anchors": {
        "0": "An explicit product statement establishes that it does not measure US coverage for the sold unit.",
        "1": "A product-specific US count identifies the observation unit and period or snapshot.",
        "2": "It also identifies the observed supply, panel, client or source population.",
        "3": "It also states the eligible US base, using the same unit and period for numerator and denominator.",
        "4": "It also distinguishes observed versus modelled units and specifies how repeats, filtering and excluded units are handled.",
        "5": "Level 4 plus independent assessment or a qualifying official product record supplying the matching base and numerator counts, or a documented sample and weights sufficient to reproduce the estimate."
      }
    },
    {
      "key": "us_availability",
      "label": "US availability and use",
      "short": "US use",
      "definition": "Document the route through which a US buyer can obtain and use the service at the cutoff. A global site, a US office or a multinational parent is insufficient. Grade the US product and any restrictions on access. Other territories appear as attributes only.",
      "anchors": {
        "0": "The operator explicitly excludes US access. Reclassify the product as listed if it has no in-scope US route at all.",
        "1": "A product-specific record identifies a US test or limited US access route.",
        "2": "A product-specific record documents a US beta or production access only through a named reseller or partner.",
        "3": "A product-specific record documents general US availability and the access route.",
        "4": "General US availability is documented and a dated US production deployment identifies the product and buying task.",
        "5": "Level 4 plus independent observation or a qualifying official product record confirming that US deployment. A commercial partner's testimonial alone remains an interested source."
      }
    },
    {
      "key": "output_validation",
      "label": "Evidence testing the sold output",
      "short": "Validation",
      "definition": "Assess test design and public evidence, regardless of the direction or commercial attractiveness of the result. For A/B/D, measure the claimed entity relationship or output behavior against an external truth criterion that matches the claim: a token-generation test alone cannot test person or household accuracy. Record product/version, unit, comparison rule, observation period, evaluation sample, denominator, sponsor, errors and limitations. A tester must disclose independence or provide evidence establishing it. Repeated publication of one study counts once.",
      "anchors": {
        "0": "The provider explicitly states that the sold output is not evaluated against any external reference, and no qualifying product-specific evaluation supersedes that statement.",
        "1": "A product-specific performance test or accuracy claim is reported, but its reference labels and denominator are unavailable.",
        "2": "An evaluation identifies its reference source and the unit compared and reports a product-specific result.",
        "3": "It also describes how the reference labels were made, gives the evaluation period and denominator and reports a defined error or agreement metric with its direction. Vendor or commissioned tests can qualify.",
        "4": "It also uses a held-out evaluation population or independent external test set and documents its separation from development and the choice of positive and/or negative examples for the named metric.",
        "5": "Level 4 from a demonstrably independent test with public error counts or uncertainty and explicit scope limits for the scored unit. A low or adverse measured accuracy can earn the same documentation level as a high result."
      },
      "anchors_by_cohort": {
        "C": {
          "0": "The provider explicitly states that no review of the sold assignments against an external benchmark is carried out, and no qualifying product-specific evaluation supersedes that statement.",
          "1": "A product-specific result is claimed for audience or inventory assignment, but the reference and denominator are unavailable.",
          "2": "An evaluation identifies a benchmark for the sold group or inventory task, the unit compared and the result; personal linkage is unnecessary.",
          "3": "It also explains how the benchmark was made, gives the period and denominator and defines a group-assignment or inventory-eligibility error or agreement metric. Campaign lift by itself does not test assignment.",
          "4": "It also evaluates a held-out or independent external sample and documents its separation from development and the sampling of eligible or ineligible units appropriate to the metric.",
          "5": "Level 4 from a demonstrably independent test with public error counts or uncertainty and explicit scope limits. Person-level tracking earns no additional credit."
        }
      }
    },
    {
      "key": "withdrawal_controls",
      "label": "Stopping or removing an output",
      "short": "Withdrawal",
      "definition": "Score the documented operational control once a use is withdrawn, an input is removed, or an assignment must be corrected. Record who can trigger it, what is suppressed or removed and where, and any reappearance limits. State whose request is handled. Level 4 requires a stated time or event boundary. This is separate from the initial transfer decision and never determines legal compliance. Cohort C may operate with no persistent person or device key.",
      "anchors": {
        "0": "The operator explicitly states it cannot suppress or remove the relevant output from any future use after a valid withdrawal.",
        "1": "A named product-specific route accepts a withdrawal, removal or suppression request.",
        "2": "The record also identifies the requester, the affected key or assignment, and the input or authentication the service accepts.",
        "3": "It also explains the effect on the next computation or release and identifies the party that enforces it.",
        "4": "It also gives a time or event boundary for enforcement and explains downstream treatment of previously released outputs or the technical limit on their recall.",
        "5": "Level 4 plus independently observed or qualifying official product-specific evidence of suppression or withdrawal across that boundary, including the stated downstream effect or limit."
      },
      "anchors_by_cohort": {
        "C": {
          "0": "The operator explicitly states that a withdrawn input or ineligible inventory unit cannot be excluded from subsequent assignments.",
          "1": "A product-specific route for withdrawing inputs, disabling a segment or excluding an inventory unit is documented.",
          "2": "It also identifies the participant who initiates the action and the input, segment or unit affected. A product that assigns per request can document the corresponding exclusion trigger.",
          "3": "It also specifies how the next computation, request or segment release enforces that action.",
          "4": "It also gives the time or event boundary and states how prior assignments expire or cease to be usable. A documented in-request exclusion with no retained output meets the same criterion.",
          "5": "Level 4 plus independent observation or a qualifying official product-specific record that verifies subsequent suppression or per-request exclusion at the claimed boundary."
        }
      }
    },
    {
      "key": "change_traceability",
      "label": "Version and correction history",
      "short": "Changes",
      "definition": "Trace a buyer-visible change to a product's rule, data definition or delivered output. Record effective time, scope and the effect on recipients. A change of logo, legal owner or corporate name is outside this dimension.",
      "anchors": {
        "0": "The provider explicitly states that versions or effective dates are unavailable for all changes to the sold output.",
        "1": "A dated product-specific technical release or correction notice identifies the changed behavior.",
        "2": "It also identifies the version or effective date and which output, field or rule changed.",
        "3": "It also identifies affected recipients or records and explains how buyers can distinguish old and new behavior.",
        "4": "It also states the migration, reprocessing or compatibility treatment and provides a public before/after example or reproducible test for the same change.",
        "5": "Level 4 plus independent observation or a qualifying official product record confirming the dated change and its effect on the sold output."
      }
    }
  ],
  "peer_groups": [
    {
      "key": "A",
      "label": "Interoperable IDs",
      "definition": "A buyer or publisher obtains a specific identifier whose documented buying task is transport and recognition across named parties in an advertising handoff."
    },
    {
      "key": "B",
      "label": "Identity graphs and resolution",
      "definition": "A buyer obtains a distinct identity resolution, onboarding or cross-entity link service. The purchased output is a match, resolved record or graph."
    },
    {
      "key": "C",
      "label": "Cookieless targeting alternatives",
      "definition": "A buyer obtains group, modelled audience or seller-defined targeting explicitly sold for addressability in place of a conventional identity route. Ordinary contextual targeting and generic deal curation stay outside."
    },
    {
      "key": "D",
      "label": "CTV and household identity",
      "definition": "A buyer obtains a specific household link, assignment or graph for TV or CTV addressability or measurement. The purchased TV household service must be distinguishable from general resolution."
    }
  ],
  "cohorts": [
    {
      "key": "A",
      "label": "A · Interoperable IDs",
      "definition": "A buyer or publisher obtains a specific identifier whose documented buying task is transport and recognition across named parties in an advertising handoff.",
      "applicable": [
        "assignment_method",
        "input_provenance",
        "refresh_expiry",
        "permission_handling",
        "match_rate_definition",
        "recipient_interface",
        "destination_coverage",
        "us_coverage_measure",
        "us_availability",
        "output_validation",
        "withdrawal_controls",
        "change_traceability"
      ],
      "banded": true
    },
    {
      "key": "B",
      "label": "B · Identity graphs and resolution",
      "definition": "A buyer obtains a distinct identity resolution, onboarding or cross-entity link service. The purchased output is a match, resolved record or graph.",
      "applicable": [
        "assignment_method",
        "input_provenance",
        "refresh_expiry",
        "permission_handling",
        "match_rate_definition",
        "recipient_interface",
        "destination_coverage",
        "us_coverage_measure",
        "us_availability",
        "output_validation",
        "withdrawal_controls",
        "change_traceability"
      ],
      "banded": true
    },
    {
      "key": "C",
      "label": "C · Cookieless targeting alternatives",
      "definition": "A buyer obtains group, modelled audience or seller-defined targeting explicitly sold for addressability in place of a conventional identity route. Ordinary contextual targeting and generic deal curation stay outside.",
      "applicable": [
        "assignment_method",
        "input_provenance",
        "refresh_expiry",
        "permission_handling",
        "recipient_interface",
        "destination_coverage",
        "us_coverage_measure",
        "us_availability",
        "output_validation",
        "withdrawal_controls",
        "change_traceability"
      ],
      "banded": true
    },
    {
      "key": "D",
      "label": "D · CTV and household identity",
      "definition": "A buyer obtains a specific household link, assignment or graph for TV or CTV addressability or measurement. The purchased TV household service must be distinguishable from general resolution.",
      "applicable": [
        "assignment_method",
        "input_provenance",
        "refresh_expiry",
        "permission_handling",
        "match_rate_definition",
        "recipient_interface",
        "destination_coverage",
        "us_coverage_measure",
        "us_availability",
        "output_validation",
        "withdrawal_controls",
        "change_traceability"
      ],
      "banded": true
    }
  ],
  "cohorts_note": "Peer chips mirror the four buying tasks. Only the cohort supplies the field mean. Band eligibility requires at least one counted cell in each applicable group and at least six counted cells. Cohort C excludes individual match-rate disclosure by design and can score group assignment, validation and withdrawal at every level without a person identifier.",
  "row_order": [],
  "matrix_title": "Public evidence for advertising identity products",
  "matrix_note": "Each cell shows a registered documentation criterion for one product and task. Open it for the deciding source, scope, grade and recheck. A blank applicable criterion means insufficient evidence. Bands describe the record within a buying task.",
  "matrix_foot": "Read each IQ with its cohort, coverage, missing groups and sensitivity. The method does not estimate whether a particular person's or household's identity is correct.",
  "composite_index": {
    "used": true,
    "name": "Identity Quotient (IQ)",
    "added": "Registered on 11 November 2026 before research or scoring. Stage 5 records the actual pre-scoring freeze of this rule set, the cohort masks and the resolved product register.",
    "groups": {
      "Depth": {
        "label": "Depth",
        "meaning": "How the product generates and refreshes its output, handles permission and defines a match where that criterion applies.",
        "dims": [
          "assignment_method",
          "input_provenance",
          "refresh_expiry",
          "permission_handling",
          "match_rate_definition"
        ]
      },
      "Reach": {
        "label": "Reach",
        "meaning": "How the output travels, which systems use it, how US coverage is measured and where US access is documented.",
        "dims": [
          "recipient_interface",
          "destination_coverage",
          "us_coverage_measure",
          "us_availability"
        ]
      },
      "Standing": {
        "label": "Standing",
        "meaning": "How the sold output has been tested, how it can be withdrawn and how buyers can trace changes.",
        "dims": [
          "output_validation",
          "withdrawal_controls",
          "change_traceability"
        ]
      }
    },
    "weights": {
      "Depth": 0.45,
      "Reach": 0.25,
      "Standing": 0.3
    },
    "field_mean_method": "raw_mean",
    "prior_strength": 4,
    "shrink_alternatives": [
      0,
      2,
      8
    ],
    "missing_scenario_values": [
      1,
      4
    ],
    "missing_envelope_values": [
      0,
      5
    ],
    "bands": [
      [
        "Most documented",
        12
      ],
      [
        "Well documented",
        1
      ],
      [
        "Partly documented",
        -8
      ],
      [
        "Least documented",
        -1000
      ]
    ],
    "thin_record_min": 6,
    "cohort_min_banded": 6,
    "method": "Within each cohort, use admitted named products, including thin rows. The mean of counted values in each group is the group score. Omit groups with no counted value. Raw is the weighted mean of present groups with weights renormalised over those groups. F is the arithmetic mean of admitted rows' Raw values. For k counted numeric cells, Shrunk = (k × Raw + 4 × F) / (k + 4). IQ = 100 × Shrunk / 5 at full precision. Round displayed IQ only, halves up. An empty cohort has no IQ.",
    "reweightings": [
      {
        "name": "Equal thirds",
        "weights": {
          "Depth": 0.3333333333333333,
          "Reach": 0.3333333333333333,
          "Standing": 0.3333333333333333
        }
      },
      {
        "name": "Depth-led",
        "weights": {
          "Depth": 0.6,
          "Reach": 0.2,
          "Standing": 0.2
        }
      },
      {
        "name": "Reach-led",
        "weights": {
          "Depth": 0.3,
          "Reach": 0.45,
          "Standing": 0.25
        }
      },
      {
        "name": "Standing-led",
        "weights": {
          "Depth": 0.3,
          "Reach": 0.2,
          "Standing": 0.5
        }
      }
    ],
    "why_not_a_sum": "Group means retain the registered weights despite different masks. Shrinkage limits the effect of a small observed k. A missing low cell can still raise a result, so report coverage, deletion tests and missing-evidence scenarios.",
    "limits": "The IQ describes the public evidence for a specific product and buying task. Actual accuracy, adoption volume, investment value, legal compliance and commercial fitness require separate evidence. No rank or band comparison across cohorts. A band affected by deletion or the missing-evidence scenarios must be labelled beside its coverage. No independent second scorer.",
    "bands_rule": "Use the unrounded offset 20 × (Shrunk − F). Most documented at offset ≥ 12. Well documented at 1 ≤ offset < 12. Partly documented at −8 ≤ offset < 1. Least documented below −8. Equal offsets share the band within a cohort.",
    "thin_rule": "Require at least six counted cells and at least one in each applicable group. An admitted row below this test has a thin record; name any missing group and show coverage with no band. A cohort needs at least six eligible rows before any bands are shown.",
    "shrink_rationale": "K = 4 is an editorial prior strength. The dimensions are different capabilities. Test K = 0, 2 and 8 as alternatives and report changes per cohort. K = 0 gives the unshrunk Raw score.",
    "missing_rule": "Report k divided by the applicable mask size for each row. Delete one counted cell at a time and recompute k and eligibility, reporting quotient movement, admission loss and band eligibility loss separately. For missing-evidence scenarios, fix F and all other rows, fill the row's applicable n/e cells with 1, then with 4, using observed k and eligibility. Flag missing-evidence sensitivity when the observed band changes. Run fills of 0 and 5 separately as hypothetical full-scale scenarios; they never trigger that flag. An imputation cannot admit a product or confer a band. Report a separate deletion-sensitivity flag where deletion changes an observed band.",
    "reweighting_rule": "Recompute Raw and F across the same admitted rows within each cohort for each registered variant and report movement. Count observed k and band eligibility under the same masks."
  },
  "scorer": {
    "independent_second_scorer": false,
    "disclosure": "One AI-assisted scoring process with a separate refute pass on every cell and an audit of supporting claims. No independent second scorer or inter-rater statistic."
  }
}
