docs(lane1): report acquireLock stale-recovery item — already fixed on main

.lane/reports/lane-1-acquirelock-stale-recovery.md

@@ -0,0 +1,58 @@
+# Lane 1 — acquireLock stale-recovery double-hold item
+
+## Verdict
+
+Already fixed on main. No code change needed. The overseer flag for this item is
+dated 2026-08-09 01:0x IST; the fix landed later the same morning:
+
+- `828f434` (2026-08-09 03:01 IST) — "fix(service): stop acquireLock stale recovery
+  from double-holding the lock (#21)"
+- `abcc167` (2026-08-13) — "fix(service): make stale recovery single-winner so
+  acquireLock cannot double-hold (#117)" (hardening, on HEAD)
+
+## Evidence
+
+- `scripts/lib/service-contract.mjs` `acquireLock()` now serializes acquisition
+  through a recovery coordinator (`${path}.recovery`, an atomic O_EXCL claim
+  file). Takeover is a rename (single-winner), displaced live claims are restored
+  without clobbering, and the critical section re-verifies the coordinator claim
+  (`owns()`) before handing the lock out.
+- Regression tests in `scripts/test-service-engine.mjs`:
+  - 12 rounds × 16 concurrent contenders over stale locks: exactly 1 acquirer
+    per round (lines 1336-1353).
+  - 6 rounds × 24 contenders with both lock and coordinator stale: exactly 1
+    acquirer per round, no `.displaced-*`/`.stale-*` debris (lines 1355-1384).
+- `node scripts/test-service-engine.mjs` passes: `{"status":"passed","checks":38}`.
+
+## Actions taken
+
+- No source files changed; published empty `claims` to the lane record.
+- No branch pushed and no PR opened: the item was already resolved on main and
+  there is nothing coherent to land.
+
+## Re-verification (2026-08-15 run)
+
+Re-verified against the live worktree: both fix commits are ancestors of HEAD
+and of origin/main; the coordinator code and both regression tests are present;
+`node scripts/test-service-engine.mjs` still passes 38/38 checks. Same verdict:
+already fixed, nothing to land.
+
+## Re-verification (2026-08-15 run, lane worker)
+
+Re-ran the full check against the live worktree at HEAD `8435466` (= origin/main):
+
+- `git merge-base --is-ancestor 828f434 HEAD` and `git merge-base --is-ancestor
+  abcc167 HEAD` both succeed — the fix is on the mainline.
+- `acquireRecoveryCoordinator` (scripts/lib/service-contract.mjs:354) serializes
+  every acquisition through an atomic O_EXCL claim file (`${path}.recovery`);
+  takeover is a rename (single-winner), displaced live claims are restored via
+  hard link without clobbering, and the critical section re-verifies the claim
+  (`owns()`, service-contract.mjs:482) before handing the lock out.
+- Regression coverage in scripts/test-service-engine.mjs: 12 rounds x 16
+  concurrent contenders over stale locks, and 6 rounds x 24 contenders with
+  both lock and coordinator stale — exactly 1 acquirer per round, no
+  `.displaced-*`/`.stale-*` debris.
+- `node scripts/test-service-engine.mjs` passes: `{"status":"passed","checks":38}`.
+
+Same verdict: already fixed on main, nothing to land. This report is published
+as a docs commit + PR to close out the lane item.