Fixing a bug

Reproduce first, from the outside

Before reading code or forming a theory, reproduce the bug end to end, as close
to how a real user hits it as possible. For this site that means the built app
in a browser, not a unit test and not the dev server:

npm run build
npm test                  # or drive the specific route with Playwright

A bug you cannot reproduce is a bug you cannot verify you fixed. A theory
formed from reading code alone tends to find a plausible cause rather than the
real one.

Then

1. Narrow the reproduction to the smallest set of steps that still shows it.
2. Fix the cause, not the symptom. If the fix is a workaround, say so plainly
   and explain what the real fix would be.
3. Add or extend a test that fails without the fix. Put it at the level the bug
   was reproduced at, so it guards the behaviour a user actually sees.
4. Re-run the full suite. A fix that breaks something else is not a fix.

Flaky tests

Treat flakiness as a defect as serious as a wrong result, and fix it when you
see it, including when it is unrelated to the current task.

Find the actual race before changing anything. Adding a sleep, a retry, or a
longer timeout hides the problem and keeps the signal untrustworthy. Ask whether
the flake is in the test or in the product, because a test that races the app
often means users can race it too.

Re-run a suspect test several times before calling it fixed.