Use this once a week — solo or with agents — to keep Part X's system honest. It catches drift before it hardens into rework. No new concepts; it only re-reads what Classes 41–44 already built.

The ritual (timeboxed)

1. History (5 min). Run git log --oneline -15 and skim the week's diffs. Ask: is every commit one coherent change with an imperative message? Flag any "AI changes" dump for smaller tasks next week. 2. Docs truth (5 min). Open ROADMAP.md, DECISIONS.md, and any INTERFACE-NOTE.md touched this week. Ask: does the code still agree with each file? If not, run the Class 42.3 plan-change protocol now — update the doc, record the decision, re-scope the task. 3. Board + risk (5 min). Open TASK-BOARD.md. Ask: is every task's owner, dependency, and risk level still correct? Is the integration owner named? Move anything blocked to "Blocked" with an owner and a question — never leave a guess in progress. 4. Next slices (5 min). Confirm the next 1–3 task cards each have: one outcome, one boundary, two acceptance checks, a risk level, and a stop/escalate line. If a card fails that test, shrink it before any agent starts.

Finish line

A dated REVIEW-NOTE.md (5–10 lines): commits reviewed, doc contradictions fixed, board risks re-scored, and the next 1–3 task IDs with owners. Link it from TASK-BOARD.md or DECISIONS.md when it records a decision.

Why weekly

Daily loops (Lesson 44.4) keep one task honest. The weekly review keeps the *system* honest: history readable, docs canonical, board truthful, next work ready. Twenty minutes here saves the two-day contradiction cleanup Class 42.3 warns about.