ANCHORED QUALITY RECEIPTS

Anchored Quality Receipts

A signed, anchored receipt that an evaluation happened as recorded — which suite scored which output or version, by whom, to what result. It pins the exact eval suite by hash so anyone can re-run it, turning "we tested this" from a claim into a reproducible, tamper-evident fact. It records evaluation provenance, never a correctness guarantee.

AQR · Shipped2 wire typesApache-2.0 · CC0 spec
← All familiesVerify a receipt →

WIRE TYPES

ar.quality.v1 (quality.output)

Per-output runtime eval. Commits to the evaluated output (salted SHA-256), names the suite and its hash, the scorer (deterministic / LLM-judge / human / hybrid), and the result — score, verdict (pass/fail/warn), threshold, and cases passed of total — at a UTC time. Optionally links the originating agent run by request_id_commit.

ar.quality.v1 (quality.release)

Pre-deploy version gate. Same suite, scorer, and result fields, but bound to a plaintext software version (e.g. agent-core@0.5.0) and a gate flag recording whether the eval blocked the deploy or was advisory — the proof a version cleared its suite before shipping.

WHAT IT PROVES

  • — A named eval suite, pinned by hash, was run against a committed output or version at a specific UTC time (suite_hash + Merkle inclusion proof).
  • — The recorded score, verdict, threshold, and case counts are exactly as the issuer sealed them and have not been altered (Ed25519 signature check).
  • — The receipt was signed by a registered issuer key (Ed25519 signature check).
  • — The exact suite definition is reproducible — a third party can re-run it against suite_hash and compare (salted SHA-256 commitment).

WHAT IT DOESN'T PROVE

  • — That the evaluated output is correct, safe, or actually "good" — the receipt records a score, not ground truth.
  • — That the verdict follows from the score — the issuer asserts the pass/fail/warn call, including human overrides; the verifier never re-derives it.
  • — That the eval suite measures anything meaningful — trust in the number comes from inspecting the named suite, not from the receipt.
  • — Anything about EU AI Act conformity — that is a separate family (AER); a quality score is not a regulatory attestation.

COMPOSES WITH

AQR receipts reference other family members via body-level composition pointers — verifier-coordinated, not signature-mandated.

AActRAnchored Action Receipts

A quality.output receipt links to the agent run it evaluated by request_id_commit — the same run's AActR records what authority drove it, the AQR records how the output scored.

AERAnchored Evaluation Receipts

Deliberately distinct: AER carries EU AI Act conformity and transparency attestations; AQR carries engineering output-quality scores. A buyer wanting both holds two receipts — neither family carries the other's claims.

Verify any
AQR receipt.

verify.dekimu.com ↗

Paste any claim ID to verify a receipt, check its anchor, and inspect the issuer signature.

REFERENCES

AI Act Art. 15 — Accuracy, robustness and cybersecurity (EUR-Lex) ↗
AI Act Art. 9 — Risk management system (EUR-Lex) ↗

Anchored Quality Receipts are cryptographic provenance and privacy-lifecycle protocols. verify.dekimu.com is a reference implementation, not a qualified trust service under Regulation (EU) No 910/2014 (eIDAS) or successor.