Rescuing a Playwright suite nobody trusted
The Playwright suite at Dover failed roughly 30% of the time. Nobody could tell a real regression from noise, so red runs were being ignored — which is the same as having no suite at all.
I rebuilt it as a modular framework with Page Object Models, then went through historical runs and test data to find the actual causes of flakiness rather than papering over them with retries.
Failure rate down from 30% to a 2% margin, with results that hold steady across environments.

