Note Protocol for signed, per-decision evidence in AI runtime governance.

Note Distribution mirror of the experimental AIREP Embedded Evaluation Profile v0.1 — not a benchmark or training dataset. An evidence contract for evaluator identity, target/access/run provenance, explicit NOT_RUN/NOT_MEASURED states, evidence digests, disclosures and verification references. Carrier: AIREP 0.2. Profile validation is not an assurance result and does not certify safety or evaluator independence.

Note Runnable laboratory for inspecting what signed runtime evidence does—and does not—establish.

Note Reusable AIREP evidence, verifier and negative test cases for runtime-governance record verification.

Note Interactive LightEval results → AIREP Embedded Evaluation Profile exporter (current parser: LightEval results_*.json; Inspect/OpenEvals are future targets, not parsed). Produces a profile payload and content-addressed evidence manifest; declared context is recorded, not verified; establishes only digests and schema validation — not evaluation truth, safety or independence.

Note Deterministic AI runtime architecture with structured state and pre-response governance.

Note Runnable lab for separating measurement validity, non-measurement and NOT_MEASURED outcomes.

Note Source-pinned cases and reference runs that keep expected and observed measurement outcomes separate.