Skip to content

Rehearse a reviewed behavior check

Use the team's existing test tools. This shows one reviewed check detects one concrete regression; it does not score the suite or pick a universal framework.

  1. Inspect sources. Read [setup.steps], [checks.commands], checks.required, .mise.toml (including an empty [tools] table), test-runner configuration, manifests, CI commands and the relevant requirement or acceptance source. Record exact paths, the behavior each source establishes and unknowns. Do not infer coverage from filenames, a syntax check or successful setup.
  2. Propose and review the mapping. In the delivery slice, record requirement and check IDs, observable behavior, inspected sources and the exact team-owned command. Add prerequisites, the required/optional choice, unknowns and the review source. Use existing authorization; do not add a second approval ritual. Change shared configuration only after this review, preserving authored tests, required IDs and unrelated commands.
  3. Prepare the normal runner. Run the reviewed setup action separately through ai-dlc project setup; a check never installs its own tools or silently changes shell. mise is required for ai-dlc project check even when the tools table is empty. If runtime resolution reports it unavailable, no check ran and no passing receipt exists. Direct shell execution is diagnostic only.
  4. Preflight an external disposable fixture. Copy the reviewed fixture to an explicit temporary location outside the active checkout. Record the source revision and fixture identity, and keep the exact original bytes. Stop if the fixture is in the active checkout, isolation is uncertain, or the exercise would use production data or mutate a remote service.
  5. Observe pass, regression failure, restored pass. In the disposable copy, run ai-dlc project check --check CHECK_ID and require a pass. Introduce only the reviewed regression, rerun it and require the expected failure. Restore the saved bytes, verify the fixture matches its baseline, and require a final pass. If the changed behavior still passes, or any outcome is unobserved, the rehearsal is incomplete.
  6. Retain bounded evidence. Record requirement/check IDs, exact command, engine and runtime identity, fixture/source identity, the deliberate change, all three outcomes and limitations. This demonstrates detection only for the selected behavior. It does not establish complete adequacy, human quality, productivity, native-platform qualification or comparison results.

Never alter the active checkout to manufacture a failure, and do not change the test until it appears successful.

A focused receipt is edit feedback. It retains the full configured required list and cannot satisfy missing completion outcomes. After target-branch integration, run ai-dlc project check --required for complete required evidence.