Plan Review — the sweep

Review the pre-implementation brief (plan or simple-cycle todo line) before any test is written.

The six-category sweep

Sweep all 6 categories in order:

Always perform two mechanical checks:

• Case coverage: Map every briefed Behavior case ID to its step(s). A case with no step is always a blocker.
• Red-first viability: State the failing assertion for each step. If none exists, grade at least risk.

Verify claims before reporting (grep for assumed helpers; state both readings of an ambiguity). Drop or mark confidence: low if unverified.

Severity (closed)

Grade up on doubt (blocker over risk, explaining why).

Report format

Number findings P1, P2, … most severe first:

**Subject:** plan | todo — <entity slug(s)>, cases <B1, B2>
**Checked:** <one line per category swept, and what you read>

## P1 — blocker — spec-fidelity — <one-line claim>

**Where:** plan step 4 | todo line | `path/to/file.py:42`
**What's wrong:** <2–3 sentences, mechanism not vibes>
**What it costs:** <what the cycle produces if this is not fixed first>
**Suggested resolution:** <the smallest change to the brief — one sentence, never a patch>
**Confidence:** high | medium | low

**Notes:** <tally of `note` items, one clause each>

blocker and risk get full sections; note items go in the Notes: tally. If nothing exceeds note, say so with the Checked: lines. On re-dispatch after a re-plan, check only the revised brief without reopening round-one lines of attack.