Rights Check

• Approve

  Every check passed. The answer lists the duties that come with the use.

• Deny

  A binding rule fails, such as a territory outside the grant or a revoked consent.

• Abstain

  A person needs to decide: an undefined term, a consent that may not fit the use, something the request leaves out, or a condition not yet met.

Evidence a person confirmed

Fourteen checks run every time, and each one that relies on a contract cites the clauses with their verbatim text. Clauses and records that your agent, an import or the API wrote count only once a person confirms them; until then the check is Unknown.

Prerequisites before use

A brand approval, a payment due before use or a pending review gives Abstain with "condition outstanding", never an Approve. Once someone records who met it, when and how, a re-run moves past it.

Duties

An approval lists what must happen at use, such as credits and AI labels, and after use, such as royalty reports. A jurisdiction's labelling duty is attached only when it applies to your role and the output.

Legal review

A separate flag for advertising with a real performer's AI likeness, altered real performances, sensitive contexts, active court orders and US post-mortem rights, and for an Approve that rests on precedence or a Deny that rests only on judgement.

Rulings

A reviewer settles a judgement call, such as which countries "South Asia" means in one licence, with a ruling that records who decided, why and which clauses they relied on. Later checks with the same scope reuse it.

Revocation

Withdrawing a consent, ending a grant, correcting a clause or withdrawing a ruling flags every decision that used it. Re-runs keep the earlier decision, and known outputs are listed as draft, published or sublicensed.

POST https://raq.com/api/v1/rights-check/packs/<pack ID>/checks
Authorization: Bearer <token>
Content-Type: application/json

{
  "account_id": "<your account ID>",
  "party": "Okafor Drinks",
  "purpose": "advertising",
  "characters": ["Rosa Illingworth"],
  "performers": [{"name": "Tom Ashworth", "attributes": ["voice"]}],
  "territories": ["GB", "South Asia"],
  "geo_restricted": true,
  "media": ["advertising", "social"],
  "start_date": "2027-10-01",
  "end_date": "2027-11-30",
  "operations": ["ai_augment", "adapt"],
  "model": {
    "provider": "Google",
    "model": "Veo",
    "version": "3.1",
    "trains_on_inputs": false
  },
  "output_types": ["video"],
  "contexts": ["soft_drinks"],
  "replica_type": "employment_based",
  "alters_real_performance": false
}

{
  "decision": {
    "reference": "RC-0004",
    "outcome": "abstain",
    "reasons": [
      "Territory: \"South Asia\" is not a country or a territory group this pack defines.",
      "Consent: Whether advertising use fits C4's description (\"Narration for the feature film Brickyard Walk and its trailers\") is a judgement for review (it states no media or actions either); C4 does not state when it starts or ends; C4 does not state where it applies.",
      "Duties: Condition outstanding: Tamar Pictures brand team's approval (CH1.approval) must be met before use and is not recorded as fulfilled."
    ],
    "checks": [
      {
        "code": "CONSENT",
        "result": "unknown",
        "clauses": [
          {
            "citation": "D4 cl. 1",
            "text": "Tom Ashworth will record the narration for the feature film Brickyard Walk and its trailers, and consents to the use of his recorded voice for those purposes.",
            "verified": true
          }
        ]
      }
    ],
    "prerequisites_outstanding": [
      {
        "key": "CH1.approval",
        "holder": "Tamar Pictures brand team",
        "timing": "before_use"
      }
    ],
    "needs_legal_review": true,
    "legal_review_reasons": [
      "Advertising with an AI likeness of a real performer (an advertising deepfake)"
    ],
    "notice": "Decision support, not legal advice."
  }
}

The full decision carries all fourteen checks with every clause they cited, the record and jurisdiction pack versions it used, and every duty with its timing. A person can rule on what "South Asia" means in this licence and whether the narration consent reaches advertising, and later checks with the same scope reuse those rulings. Bad input gets a 400 that names every field at fault, never a quiet success.

Wrong approvals

0

Matched the expected answer

232 of 232

Abstained where a person should decide

71 of 71

Wrong denials

0

Cited the right clauses

100%

Legal review flag matched

100%

Revoking one consent flagged 1,000 decisions in 294 ms and re-ran them in 30.2 seconds on Apple M1 Max, 64 GB memory; local development machine with local PostgreSQL 18.6, PHP 8.5. Evaluator rc-4, measured 11 November 2026.

ODRL 2.2

Grants, prohibitions and duties are ODRL rules under a published profile that adds the AI model, the AI operation hierarchy and how conflicts are settled. A pack exports as ODRL JSON-LD. https://raq.com/ns/rights-check/v1

C2PA and CAWG

An approved use exports provenance assertions to attach to the output: c2pa.created with its digital source type, parentOf ingredients, and cawg.training-mining set to notAllowed where the asset forbids AI training.

IPTC and PLUS

The same bundle carries the IPTC digital source type, AI system used and its version, the PLUS data-mining value and a TDMRep reservation.

EIDR

Assets keep their EIDR title and edit IDs, with ISAN and ISRC where you hold them, so a decision names exactly which cut it covers.

The provenance bundle is unsigned. Signing a C2PA manifest is out of scope, so your own signing step adds the signature when it writes the output.

Decision support, not legal advice

It applies the clauses and consents you record. Judgement calls come back as Abstain for a person, and the legal review flag says when to ask a lawyer.

No automatic takedowns

A revocation flags decisions and lists outputs by state. What happens to published work is for the contract and the people involved.

Not a rights registry

Your registry stays the system of record. Rights Check reads what a decision needs from it and never writes back.

No AI inside a decision

The evaluator is ordinary code: the same records and request always give the same answer.