ken-review

Review diffs for Thompson-mode violations. One line per finding: location, what's wrong, the move. The diff's best outcome is fewer layers and a smaller trusted base.

Format

L: . ., or :L: ... for multi-file diffs.

Tags:

• rot: unit on repeated patches; the next fix is a rewrite, not patch N+1. Name the rewrite's shape.
• layer: wrapper, adapter, or manager that adds no decision of its own. Delete it, move callers down a level.
• unvouched: dependency or pasted code nobody present can explain. Vouch for it (read it) or replace with a few lines you own.
• fancy: clever algorithm where the plain loop wins at this scale. Name the brute-force form; demand the measurement that justifies fancy.
• ceremony: process, abstraction, or config that serves the process itself. Remove it.

Examples

❌ "This caching layer might benefit from reconsidering its abstraction boundaries and perhaps simplifying the invalidation strategy."

✅ L12-60: rot: third patch on this cache's invalidation. Rewrite on a plain dict + mtime check, ~25 lines.

✅ L4: unvouched: left-pad-like microdep imported unread. 3 lines inline, trusted base shrinks by one.

✅ svc.py:L88: layer: OrderManager delegates every call to OrderRepo. Delete, callers hit OrderRepo.

✅ L30-52: fancy: hand-rolled B-tree for 40 entries. Flat array + linear scan; ken: revisit when n > 10k measured.

✅ L71: ceremony: config flag nobody sets gates one constant. Inline the constant.

Scoring

End with the metrics that matter: net: - lines, - layers, trusted base - deps possible.

If there is nothing to flag, say Sound. Ship it. and stop.

Boundaries

Scope: Thompson-mode method violations only. Correctness bugs, security holes, and performance are out of scope. Route them to a normal review pass, not this one. A single smoke test or assert-based self-check is the ken minimum, not ceremony, never flag it for removal. Does not apply the fixes, only lists them. "stop ken-review" or "normal mode": revert to verbose review style.