Triage a test suite that fails at random

My CI is red often enough that people rerun jobs instead of reading them. Failure data: {{paste test names, failure counts, and recent failure output}}. Stack and test runner: {{e.g. pytest + Postgres, Jest + jsdom}}. How tests run in CI: {{parallel workers, shared DB, retries on/off}}.

Do this:
1. Classify each failing test by likely cause: shared mutable state, test order dependence, real concurrency race, time or timezone, network or external service, resource limits, or a genuine intermittent product bug. Quote the evidence in the output that puts it in that bucket.
2. Separate the tests that are flaky from the tests that are correctly catching a real race in the product. Say which is which and why - this is the important distinction.
3. Rank fixes by (failure frequency x how cheap the fix is). Give the concrete change for the top three: the fixture, the isolation boundary, the wait condition.
4. Name what to stop doing - blanket retries, sleeps, mocking away the race - and what it hides.
5. A measurement plan: what to record for two weeks to prove the suite got better.

Do not suggest deleting or skipping a test without saying what coverage is lost.