The wave-6 review gate (docs/plans/implementation.md §Review gates,
the fleet gate) read the landed implementation against the pinned
ADR text line by line: the exact-count fleet sweep outcomes vs live
postgres, the joint entry+pin tx atomicity, the sweep lease vs
ADR-013 §2, the composite fleet-validity predicate, kv-only put-time
rejection, the TTL/3 renewal lifecycle, the fleet liveness verb, the
one-window-host rule, and the non-fleet postures — all conformant,
the task-hash log matched, the wave-2..5 suites green untouched.
Fixed inline:
- the A1 termination discipline's two uncovered paths — the kv
driver's pin-row expiry mint propagated between begin_joint_tx and
pin_commit without terminating the held tx (the PG-2 hazard), and
fleet_joint_publish could abandon an already-begun tx on its
host/row paths; the drivers now resolve the row inside their
caught drives and hand it precomputed to the publish tail
- a wave-6 pg-pure test-build clippy regression: the fleet test
instruments' wildcard match arms went unreachable under
--no-default-features (helpers now ride Store::fleet_host())
- the stale/missing allow(dead_code) scope notes: store.rs's landed-
state paragraph (the fleet sweep "not yet driven" → landed),
gc.rs's sharpened remainder, local/sqlite/backend notes corrected,
the pg_lo/postgres allow-scope sentences added; a measure-then-
note check across six feature shapes (no allow removed)
- a PGLO-1-class near-miss: the new notes de-linked the private
items they name; cargo doc --no-deps zero warnings, the gated
build's pre-existing count 32 → 31 (nothing added)
Verification (tests live against pglo-poc :15432): cargo test (73/
19/3 default; 131/52/3 postgres; 167/73/3 postgres+pg-lo; 226/125/4
all-features), the pg-pure builds, clippy --all-targets -D warnings
on all seven feature variants, fmt --check, doc clean, shared-db
schema residue 0. Wave 7 (the ops surface) decomposition unblocked.
glm-5.3-flash
e84a874715