← Files Web App QA & Release CopilotARCHIVED FILE
references/evidence-model.md
1.25 KB · Nov 22, 2026 · 18:39 UTC
# Evidence model and claim calibration Tag each important statement with its basis: - `observed`: personally inspected source/rendered behavior or reproduced in a real authorized environment; give tool, path, environment, time/commit where known. - `automated`: supported by logs, Playwright/CI/Lighthouse/axe/scan artifacts; specify command/config/sample. - `source-supported`: established by a supplied product specification, source code or authoritative reference, but not necessarily running behavior. - `inferred`: hypothesis awaiting decisive reproduction. - `unverified`: required evidence absent or access unavailable. Do not treat a screenshot as proof of end-to-end functionality, or one CI success as production readiness. Preserve privacy by redacting PII, credentials and tokens from traces, videos and logs. Record artifact paths only when they actually exist. ## Minimal evidence record `gate_id`, `scope`, `status(pass|fail|unknown|na)`, `basis`, `environment`, `tool/command`, `artifact`, `owner`, `notes`. `na` requires a reason and should only be allowed for optional or demonstrably irrelevant gates. Missing mandatory evidence yields `unknown`, never `pass`. When running in advice-only chat, report what a user should execute, not an invented tool result.
SHA-256: 0a3f48ad9828956f76a3922ab96df7765012d37fa9105a730ddc66b77d483874