PLAYWRIGHT CLI · CHECKED OCTOBER 15, 2026
The URL is unchanged. The response is not.
A URL-set comparison can say that both runs visited the same endpoint while hiding the response that made the flow diverge.
Our fixture is a small, owned signup-verification page. It has an email field, a verification button and a ready state. The reference run fills the field, clicks the button and observes the ready state. In a second run, the same button makes the same GET request, but the server returns 503. The ready-state wait then times out. This is a controlled reproduction, not evidence about a production service.
Read the two layers separately
| Observation | Capture A | Capture B |
|---|---|---|
| Verification request | GET /api/verify · 200 | GET /api/verify · 503 |
| Button action | End recorded | End recorded |
| Ready-state wait | End recorded | Error recorded |
A completed click means that the action reached its recorded end. It does not mean the application succeeded. Conversely, a network response does not tell us why the application handled it in a particular way. The request outcome and the later wait are separate records that support a narrower statement: this capture contains a 503 and the expected ready state was not observed within the wait.
Reproduce the comparison
- Open the comparison and choose Same URL / 503 response.
- Review the suggested pairs. Match the click to the click and the ready-state wait to its counterpart. Do not match source call IDs across recordings.
- Select the click interval. The ordered request rows show the same endpoint alias, with different HTTP outcomes.
- Open a request row for the source filename and physical line. Then open the action source to see whether its end or error was recorded.
- Keep the interval selected for export. Confirm the pairs only after checking that both actions mean the same thing.
Reference trace sample · Reference network sample · Failure trace sample · Failure network sample. These are field-reduced native records with physical line positions retained. They are not complete Trace Viewer archives.
A method change is a different observation
The GET / POST example changes the request method while preserving the endpoint and 200 response. Both ready-state waits finish. A URL set and request count can look identical here too. TracePair retains the method on every occurrence, so the method change remains visible. It does not decide which method is correct; that belongs to the application contract.
Use exact URL matching first. Ignoring an origin is useful when comparing equivalent local and staging endpoints, but it can merge unrelated services. Ignoring queries can also erase a real difference in a token, resource identifier or pagination argument. The workspace reports when a matching mode combines different raw URLs.
What to test next
Repeat the same isolated flow with one controlled change: server behavior, application build or browser state. Save the expected state and the observed response separately. A single pair of captures is enough to locate a difference; it is not enough to establish reliability, a failure rate or a root cause. The staging reproduction worksheet shows how to turn this example into a useful follow-up test.