The review-004 precedent's five code-quality dimensions (god-files,
duplication, coverage, dead code, doc drift) over the core as landed
through wave 4; the wave-4 gate's ADR verdict is not re-litigated.
- Findings volume inside the inline bar — no remediation wave.
D-1/D-2 fixed and stamped in the review session (predecessor
commit).
- R-1 booked: the session.rs split — three clean-seamed wave-4
workstreams, flat siblings per convention 14, zero public-API
change, two in-file helper dedups folded in — recommended as the
first task of wave 5 (refactor -> fuzz -> docs -> publish). Both
substrates recorded no-split-now with rationale + revisit trigger.
- R-2/R-3/R-4 duplication keeps with rationale (including the
explicit verdict that nothing in the client-mirror grammar shares
without coupling client to server internals).
- R-5 booked: the per-ingress coverage table — wave 5's fuzz-target
inventory input; coverage disposition = review-time evidence,
per-gate.
- God-file split measurement, dead-surface scans, sha256 feature
matrix exercise, allow/TODO inventories all verified sound.
The plan's wave-4 section records the review (the wave-5
decomposition reads it); the task's acceptance criteria checked.
Verification: cargo test (360), --all-features (361), both clippies
clean, fmt clean, doc zero warnings, taskgraph validate green.
glm-5.3-flash
3c0cff43c3