A test with two users opens two browsers. Diagnosis used to read only the first, so the other user's AI steps and console were missing. Server restarts and slow servers were often diagnosed as product bugs or test timing, because the evidence had no WebSocket details.
- Every page a test opens is logged: popups, new tabs, and a second user's browser
- WebSocket lines show how long the handshake took, and the close code and reason
- Evidence from all of a test's browsers is merged in time order
- Noisy lines such as aborted requests rank last, so the errors that explain a failure stay in view
- Element lookups log each selector tried and how many elements it matched