Healing and results

A failed Playwright test may show a test problem, not a product problem. Proof checks the customer outcome again before it decides.

When a saved test fails

Proof saves the failed test, logs, screenshots, and path. It then checks the original Flow in a fresh browser session. It does not ask AI to repair selectors directly.

Check the result again

The new attempt uses the original customer goal and success evidence. Proof then classifies the failure as:

stale automation The business outcome still works.
confirmed regression The outcome cannot be accomplished or verified.
inconclusive The evidence cannot support either claim.

Safe healing

A stale-test result can create a candidate test. Proof rejects removed or weaker checks. The candidate must pass three repeatable validations. The active test stays unchanged until the candidate passes policy.

A test update cannot change success

Proof may find a new way to reach the same outcome. It may not remove a criterion, weaken protected evidence, or turn a failed customer outcome into a passing test.

Confirmed regressions

A confirmed regression means both the old saved test and a new attempt failed the customer outcome. Proof opens a Finding with the affected area, reason, screenshots, video, and other evidence. It sends the Finding to the related pull request or Slack thread.

When a human must review

Tests that affect money, data deletion, permissions, or security need human approval. Their candidate enters review_required. An owner or administrator can compare the old and new behavior and evidence before activation.