Evidence receipts

Trace what ran, what did not, and why

Every review keeps a receipt for the exact commit it examined. It shows what ran, what did not, the evidence behind published findings, and any limitation that changes how the result should be read.

Receipts distinguish:

• confirmed behavior from model-only hypotheses;
• successful execution from skipped or failed execution;
• analyzer absence from a clean analyzer result;
• a policy rejection from an infrastructure failure;
• current findings from baseline state;
• complete coverage from a partial review.

Customer-facing evidence is prose or concrete execution output. Internal model votes, provider routing, cost reservations, and raw panel scores are never shown as customer evidence. Secrets and access tokens are redacted.