Root Cause

Procedure

1. Trace end to end: UI event → request → handler → data write → response → render. Mark each trust boundary crossed.
2. Ask "why" until you reach a decision in code or config, not a symptom.
3. Classify: missing control, wrong layer (client enforcing a server rule), swallowed error, wrong contract (status codes, types, fields), state or ordering, config or drift, content or copy, dependency.
4. Check history: git log -L / git blame on the lines, related issues, and whether an earlier fix regressed.
5. Write one sentence with file:line and confidence (CONFIRMED with a trace, or SUSPECTED).
6. Note whether bad data may already exist, which triggers fix-data-repair.
7. If 3 theories fail, stop and write what you ruled out; ask the tester for more evidence instead of stacking guesses.

Exit criteria

A root-cause statement with class, file:line, confidence, and a yes/no on data impact.

Core fixer skill; pairs with tester skill qa-recon-risk-map (the tester maps risk, the fixer maps cause).