reviewing-planslisted

# Reviewing plans <!-- Methodology distilled from OMC's critic + analyst agents (MIT), 2026-07-30. Provenance: Context-Engineering repo, docs/adrs/ADR-002-omc-salvage.md. --> A false approval costs 10-100x a false rejection — but manufactured outrage destroys the reviewer's signal. Both discipline and calibration. ## Protocol 1. **Pre-commit.** Before reading in detail, predict the 3-5 most likely problem areas for this kind of plan. Investigate those specifically — deliberate search beats passive reading. 2. **Verify everything.** Extract every file reference, function name, and technical claim; check each against the actual codebase (Read/Grep/git). An assertion you didn't verify is not evidence. 3. **Rate the assumptions.** List explicit AND implicit assumptions; rate each VERIFIED (evidence exists) / REASONABLE (plausible, untested) / FRAGILE (could easily be wrong). Fragile ones are the priority targets. 4. **Pre-mortem.** "This plan was executed exactly as written and failed." Generate 5-7 concrete failure scenarios; every scenario the plan doesn't address is a finding. Include rollback: if step N dies midway, what's the documented recovery? 5. **Scan for ambiguity.** Per step: "could two competent developers interpret this differently?" If yes, quote it, state both readings and the cost of the wrong one. 6. **Hunt what's MISSING.** The biggest differentiator. Load [references/gap-taxonomy.md](references/gap-taxonomy.md) and