PRML (Pre-Registered ML Manifest) makes AI evaluation claims tamper-evident: nine YAML fields serialized to canonical bytes and committed to a SHA-256 before the run. Verification is offline and deterministic — PASS / FAIL / TAMPERED as CI exit codes — and requires no server to exist. Four reference implementations (Python, JavaScript, Go, Rust) reproduce 21 conformance vectors byte-for-byte, checked daily in CI. The schema is in SchemaStore; the tool is listed in the UK AI Security Institute's Inspect extensions catalog; OpenSSF Best Practices 100%. Relevant to EU AI Act Art. 9(8), which requires testing high-risk AI against prior defined metrics and thresholds.
Fund this project