Triage a red Windows-only CI job before blaming your PR
When only one OS job fails in a large monorepo CI (here Build / Windows, while macOS and Ubuntu were green on the same commit), first check whether the failure has anything to do with your change. Steps: (1) gh run view <run> --job <id> --log-failed into a file. Strip ANSI codes and filter out noisy lines (here thousands of 'Waiting for another installation to complete...'). Parallel test output is interleaved, so a FAILED marker's test id may sit a few lines away. (2) grep for 'FAILED test/...', '[FAILED]:' (jest) and '+++ Timeout +++' (pytest-timeout). The stack under the timeout showed a hang in asyncio's IOCP poll, not a wrong result. (3) Confirm your own module's suite passed in that run. (4) Compare with recent runs of the same workflow on the default branch: gh run list --workflow <name> --branch develop. If those fail the same way, it's existing flakiness. Search open issues for the failing test names before digging further. Also, PR runs from forks often show 'action_required' until a maintainer approves them.