find-causelisted

# Debug Prove the mechanism before fixing it. The enemy is the plausible patch that hides the symptom without explaining it. The overcorrection is investigating after the evidence already names one causal line that predicts the failure. **No production edit before a reproduction.** If the symptom cannot be triggered, the deliverable is reproduction evidence, not a fix. ## Activation gate The cause is proven only when observed evidence names one causal line or boundary and a mechanism that predicts the symptom; a diagnostic's file and line count only when that source holds the invalid symbol, type or operation it describes. A proven cause skips to Step 4; anything less runs the loop. ## The loop 1. **Reproduce.** Capture a failing test, repeatable command or specific input, run as one bare command. Output over the cap goes to `<scratch>/debug-repro.log`, `<scratch>` being what `node "${CLAUDE_SKILL_DIR}/../../lib/scratch-path.mjs" debug` prints in this checkout, read with `tail -n 40`, on every run in this loop, because a path outside the checkout stops a worktree agent and a pasted log rides in every later turn. When missing infrastructure blocks the symptom, name it and reproduce at the first repository-owned boundary that feeds it, claiming no equivalence. Then apply the test-design row. 2. **Instrument.** Keep two hypotheses the reproduction leaves standing and add one observation that tells them apart at their first divergent boundary. Locate that boundary and its c