A flaky result.
A testable next step.

1. Read the result without guessing

Imagine a cart test that times out waiting for the total on its first attempt, then passes on retry. Playwright calls this flaky. The retry is evidence that the outcome changed; it does not establish a slow server, a broken locator or a fixed application.

The free analyzer lists the result, attempts and accumulated duration. It cannot inspect the application, trace or test code. Use Playwright's retry documentation to understand the classification and the JSON reporter to produce the input.

2. Make the missing evidence explicit

In this fictional case, we do not know whether the cart request completed, whether the assertion targets the current cart, or whether another test changed the same account. A useful investigation asks for a sanitized failing assertion, relevant setup and the difference between the failed and successful runs. It does not ask for production credentials.

3. Choose an experiment that can change your mind

Hypothesis: another test modifies shared cart state. Confidence: low until supported by setup or trace evidence. Experiment: compare the affected test with isolated data against its original setup, keeping the commit and environment fixed. Record repeat outcomes in both conditions. If failures persist with isolated data, shared state becomes a weaker explanation.

Do not change timeouts, selectors and test data at once. A passing modified run alone cannot tell you which change mattered. Keep a record of what would disprove your explanation.

4. Define the handoff

A reviewed diagnostic should separate observations, unresolved questions, plausible explanations and the exact next check. If the evidence does not support a correction, the honest deliverable is an investigation plan and its limits.