Verification strategy

Before you rely on Testing as a merge gate: local/CI parity, Keys resolution behaviour, flake policy, and artefact usefulness under failure.

  • Local run matches CI for the same commit and env variables (modulo base URLs).
  • Forced failure produces a triage path within the MVP target (~30 minutes for a competent engineer).
  • indeterminate verdicts are rare, logged, and understood—not ignored.
  • Keys misconfiguration or outage fails loudly; tests do not silently run against the wrong model.

← Phase 5 — CI · Walkthrough index · Secrets and CI setup →

Agent prompts

Optional: use these with a coding agent to implement this phase in your repo in a safe, gated sequence. Expand only when you need them.

  1. Run the checklist above on a feature branch; file gaps as issues with severity.
  2. Define team policy for when indeterminate blocks merge vs needs human override.
  3. Schedule a monthly review of top flaky goals and artefact noise.