bank top-up 2026-08-14: +24 gold items, reaches n=30 measurement floor (dated delta)
059687f verified | {"id": "det-001", "criterion": "effectiveness", "predicate": "detects_own_marking", "question": "Given an asset marked by provider P, does P's own detector return DETECTED?", "labels": ["DETECTED", "NOT_DETECTED", "ERROR"], "why": "Baseline. A detector that cannot read its own marking fails before interoperability is in question."} | |
| {"id": "det-002", "criterion": "interoperability", "predicate": "detects_foreign_marking", "question": "Given an asset marked by provider Q, does P's detector return DETECTED?", "labels": ["DETECTED", "NOT_DETECTED", "ERROR"], "why": "This is the Commitment 2 obligation. It is the one nobody currently measures."} | |
| {"id": "det-003", "criterion": "interoperability", "predicate": "agrees_on_negative", "question": "Given an unmarked asset, do P and Q both return NOT_DETECTED?", "labels": ["AGREE", "DISAGREE", "ERROR"], "why": "False positives are the failure mode that destroys trust in a detector network."} | |
| {"id": "det-004", "criterion": "robustness", "predicate": "survives_transform", "question": "After a declared transform (recompress, crop, resize, screenshot, re-encode), does the detector still return DETECTED?", "labels": ["SURVIVES", "DESTROYED", "ERROR"], "why": "Shares its transform ladder with GSPC provenance (ProvBench), so the two axes are comparable."} | |
| {"id": "det-005", "criterion": "reliability", "predicate": "stable_across_runs", "question": "Across k repeated runs on the identical asset, is the verdict identical every time?", "labels": ["STABLE", "UNSTABLE", "ERROR"], "why": "A non-deterministic detector cannot support a compliance claim."} | |
| {"id": "det-006", "criterion": "effectiveness", "predicate": "declares_absence_honestly", "question": "Where no marking is present, does the detector say so rather than guessing?", "labels": ["DECLARES_ABSENT", "GUESSES", "ERROR"], "why": "The UNMEASURED discipline, applied to detectors."} | |
| {"id": "det-007", "criterion": "robustness", "predicate": "survives_recompression", "question": "Given a marked asset recompressed at platform-default quality, does P's detector still return DETECTED?", "labels": ["DETECTED", "NOT_DETECTED", "ERROR"], "why": "Every major platform recompresses uploads; a marking that dies on recompression is operationally absent."} | |
| {"id": "det-008", "criterion": "robustness", "predicate": "survives_screenshot", "question": "Given a screenshot of a marked image, does the detector return the correct state and does the manifest survive or degrade gracefully?", "labels": ["DETECTED", "NOT_DETECTED", "ERROR"], "why": "Screenshotting is the most common casual laundering step."} | |
| {"id": "det-009", "criterion": "robustness", "predicate": "survives_format_conversion", "question": "Given conversion across declared formats (png→jpeg, wav→mp3), does detection persist?", "labels": ["DETECTED", "NOT_DETECTED", "ERROR"], "why": "Format conversion is routine in every distribution pipeline."} | |
| {"id": "det-010", "criterion": "robustness", "predicate": "survives_crop_resize", "question": "Given declared crop/resize transformations within stated tolerance, is the marking still read?", "labels": ["DETECTED", "NOT_DETECTED", "ERROR"], "why": "Cropping is both benign (thumbnails) and adversarial (stripping edges)."} | |
| {"id": "det-011", "criterion": "security", "predicate": "resists_stripping_attack", "question": "Given a declared adversarial stripping attack on the soft binding, does the system detect tampering or refuse rather than return a false CLEAN?", "labels": ["DETECTED", "TAMPER_FLAGGED", "FALSE_CLEAN", "ERROR"], "why": "A false CLEAN under attack is worse than no detector — it launders content."} | |
| {"id": "det-012", "criterion": "security", "predicate": "rejects_forged_manifest", "question": "Given a manifest with an invalid signature or wrong trust-chain root, is it rejected rather than displayed as valid?", "labels": ["REJECTED", "ACCEPTED", "ERROR"], "why": "Forged provenance is the core attack on any manifest scheme."} | |
| {"id": "det-013", "criterion": "security", "predicate": "replay_protection", "question": "Can a valid manifest from asset A be attached to asset B without detection?", "labels": ["BLOCKED", "REPLAYABLE", "ERROR"], "why": "Without binding, provenance of a real asset can be rented out to fakes."} | |
| {"id": "det-014", "criterion": "correctness", "predicate": "false_positive_rate", "question": "On a declared corpus of unmarked authentic assets, what fraction are falsely flagged SYNTHETIC?", "labels": ["BELOW_THRESHOLD", "ABOVE_THRESHOLD", "UNMEASURED"], "why": "False accusations of synthesis burn the public square; the FP rate must be measured and published."} | |
| {"id": "det-015", "criterion": "correctness", "predicate": "false_negative_rate", "question": "On a declared corpus of marked synthetic assets, what fraction are missed?", "labels": ["BELOW_THRESHOLD", "ABOVE_THRESHOLD", "UNMEASURED"], "why": "The FN rate is the laundering yield; it must be quantified per transformation."} | |
| {"id": "det-016", "criterion": "transparency", "predicate": "threshold_disclosure", "question": "Are the decision thresholds and their confidence bands disclosed so results are reproducible by third parties?", "labels": ["DISCLOSED", "UNDISCLOSED"], "why": "An undisclosed threshold is an ungradeable detector."} | |
| {"id": "det-017", "criterion": "privacy", "predicate": "no_content_leakage", "question": "Does the verification flow avoid uploading the asset or its identifying hash to a third party without disclosure?", "labels": ["LOCAL", "DISCLOSED_EGRESS", "UNDISCLOSED_EGRESS"], "why": "Verification of sensitive media must not itself become an exfiltration channel."} | |
| {"id": "det-018", "criterion": "governance", "predicate": "revocation_handling", "question": "Given a revoked signing credential, are assets signed after the revocation time distinguished from those signed before?", "labels": ["DISTINGUISHED", "CONFLATED", "ERROR"], "why": "Key compromise is a when, not an if; revocation semantics decide the blast radius."} | |
| {"id": "det-019", "criterion": "governance", "predicate": "version_negotiation", "question": "Does the detector handle manifests from older and newer spec versions with declared behaviour rather than silent failure?", "labels": ["DECLARED", "SILENT_FAILURE", "ERROR"], "why": "The ecosystem will run mixed versions for years."} | |
| {"id": "det-020", "criterion": "interoperability", "predicate": "trust_list_conformance", "question": "Does the validator honour the declared trust list and flag credentials outside it?", "labels": ["HONOURED", "IGNORED", "ERROR"], "why": "A validator that trusts everything verifies nothing."} | |
| {"id": "det-021", "criterion": "effectiveness", "predicate": "cross_modal_coverage", "question": "For each declared modality (image, audio, video, text), is detection measured separately rather than claimed as a single number?", "labels": ["PER_MODALITY", "AGGREGATE_ONLY", "UNMEASURED"], "why": "Aggregate claims hide the modality where the detector is blind."} | |
| {"id": "det-022", "criterion": "effectiveness", "predicate": "text_provenance", "question": "For text, is there any working marking/retrieval mechanism at all, or is the claim absent?", "labels": ["MEASURED", "ABSENT", "UNMEASURED"], "why": "Text is where synthetic volume is largest and provenance weakest; claiming coverage without measurement is the known gap."} | |
| {"id": "det-023", "criterion": "performance", "predicate": "latency_budget", "question": "Is verification latency within the declared budget at declared scale (e.g. feed-rate ingestion)?", "labels": ["WITHIN", "EXCEEDS", "UNMEASURED"], "why": "A detector too slow for the feed is decorative."} | |
| {"id": "det-024", "criterion": "correctness", "predicate": "ingredient_chain_integrity", "question": "For edited assets, does the ingredient/derivation chain validate end-to-end rather than only the final signature?", "labels": ["VALIDATED", "FINAL_ONLY", "ERROR"], "why": "Provenance of edits matters as much as provenance of capture."} | |
| {"id": "det-025", "criterion": "security", "predicate": "downgrade_resistance", "question": "Can an attacker strip the manifest and rely on soft-binding absence to make the asset appear merely 'unknown' rather than 'tampered'?", "labels": ["DISTINGUISHED", "CONFLATED", "UNMEASURED"], "why": "'Unknown' and 'tampered' must never be the same signal; conflation is the laundering path."} | |
| {"id": "det-026", "criterion": "transparency", "predicate": "detector_card", "question": "Is there a published detector card stating training distribution, known failure modes, and evaluation corpus hashes?", "labels": ["PUBLISHED", "MISSING"], "why": "Same doctrine as model cards: an unevaluated detector is an unaccountable one."} | |
| {"id": "det-027", "criterion": "governance", "predicate": "deprecation_policy", "question": "When an algorithm or format is deprecated, is there a declared migration path and sunset behaviour?", "labels": ["DECLARED", "UNDECLARED"], "why": "Cryptographic agility is mandatory across a decades-long compliance horizon."} | |
| {"id": "det-028", "criterion": "interoperability", "predicate": "third_party_verification", "question": "Can an independent party re-run the detection verdict from the published artefacts without vendor involvement?", "labels": ["REPRODUCIBLE", "VENDOR_ONLY"], "why": "A verdict only the vendor can check is a claim, not a measurement."} | |
| {"id": "det-029", "criterion": "effectiveness", "predicate": "generative_model_coverage", "question": "Is detection accuracy reported per generator family, with UNMEASURED for families not evaluated?", "labels": ["PER_FAMILY", "AGGREGATE_ONLY", "UNMEASURED"], "why": "Detectors overfit to the generators they were tuned on; the unknown-generator case is the live threat."} | |
| {"id": "det-030", "criterion": "correctness", "predicate": "adversarial_example_suite", "question": "Is there a maintained adversarial suite whose results are re-run and published on each detector version?", "labels": ["MAINTAINED", "STALE", "ABSENT"], "why": "A detector evaluated once against 2024 attacks is answering a question nobody is asking anymore."} | |