Root-Cause Debugging
- Reproduce the failure with the smallest credible command or scenario.
- Capture the observed error and distinguish it from pre-existing noise.
- Trace the failing path to the owning boundary.
- Form explicit hypotheses and test the cheapest discriminating hypothesis first.
- Change code only after evidence supports a cause.
- Add or strengthen a regression test when practical.
- Re-run the reproducer, targeted tests, and relevant broader gates.
Do not shotgun-edit multiple suspected causes. Do not claim a root cause that was not demonstrated.