Level 2 Exit Criteria
- First-pass success ≥95%
- Recovery success ≥95%
- Average recoveries/query <1
- Average files read ≤4
- Average confidence ≥0.70
- Grep fallback ≤20%
- Zero UTF-8 failures
- Human evaluation completed without identifying new planner failure modes
Level 2 closes only when every criterion is met.