- The OQ-21 branch-mode POC ran (commit 9f520a9 on poc/w10-oq21-lock,
dropped): the real lfs.locks API composes over the substrate stack
against a real git-lfs 3.7.0 client with lfs.locksverify true.
- The mapping decided with a corrected claim: plain co-tenant rows
carry the exclusion, not alkstore's named-TTL machinery — the
machinery's silent TTL lapse is the two-writer data-loss window,
its custody is process-local, and its trait surface carries no
enumeration for the API's list/verify verbs (all measured, alkstore
0.1.0); positions stated on TTL lapse (N/A — no TTL) and renew (N/A
— no client verb; measured that published renew DOES refresh the
exclusion, a new full TTL window from now).
- The verify-then-transfer interplay confirmed: a locked-by-another
push refused before the batch (no transfer, no residue); the unlock-
then-push riding the pre-keyed custody unchanged.
- The sqlite contact point measured: lock-traffic hammer vs receive-
pack on the co-tenant file — push median ~1.3-1.6x under concurrent
lock traffic, zero storage failures (closes the DQ-7 gap).
- Records: lfs.md §Lock server note; OQ-21 entry (Investigation
resolved, build deferral stands unchanged); OQ-03's substrate
publish blocker dropped; README's OQ summary updated; the POC task
+ tracker completed (taskgraph valid).
Verification: cargo test (default + --all-features, live pg green),
clippy --all-targets/-D warnings (both feature sets), fmt --check,
cargo doc --no-deps, the fuzz corpus gate (13 passed).
glm-5.3-flash
a54a75abd9