Systematic debugging

Establish expected behavior, observed behavior, the candidate revision, and relevant environment. Preserve the smallest useful reproducer and its actual failure. Distinguish a product defect from a fixture, dependency, command, or environment failure.

Trace the failing value or event through the owning boundaries. State a falsifiable explanation and choose a small probe that distinguishes it from plausible alternatives. Change one causal variable at a time when possible; do not accumulate speculative fixes.

Use targeted source and Git history to understand a suspicious constraint before removing it. If repeated fixes fail, reassess the explanation and missing evidence. When reproduction is unavailable, report that gap rather than claiming an observed root cause.

Engineer implements the causal fix in its assigned worktree. QA preserves independent reproduction and tests in its own worktree and returns production defects to the owner. Retain a still-failing reproducer on the QA branch or in the finding until it is resolved.

Run the original reproducer against the fix, then the relevant regression and repository-required checks. Report the failing baseline, cause supported by evidence, change, fresh result, and remaining uncertainty. Redact sensitive data from diagnostic output.