Everything so far ran locally. This is where it runs on GitHub Actions, secrets and all, then splits across two runners with real sharding.

Same identity pool, same fixtures, three browser engines instead of one. Then a real collision showed up, five workers against four identities.

Identity was solved. Test data wasn't fully verified until I checked, with AI review, and found a real gap checkout had no protection against.

Two fixtures, same file, opposite lifetimes. One built once per worker, one rebuilt every test. Here's why, and what happens if you swap them.

One shared login collided with itself the moment I added a second worker. Here's the architecture that fixed it, three classes, each with exactly one job.

Everything runs on Phase 2's framework, but you'll need your own qa-cart.com login and a few environment variables — here's exactly how to get it running before we start breaking it.

When we start learning Playwright, most of our focus naturally goes toward UI automation. We open the browser. We locate elements. We click buttons. We verify text on screen. That is absolutely the right way to begin. But not everything needs to be tested through the UI. Many validations are faster, simpler, and more reliable when we interact directly with the system via APIs. This is where API testing in Playwright becomes extremely useful. Instead of always interacting wit