[PerchAnton]

Post-merge verification (30 October 2026): merged into main as 8f119df5a745b9971f486eb2c054f3c2c3e220ea. The merge tree matches the reviewed head. Main CI 41104556559 passed: Linux 169/169; macOS ARM three complete rounds, 507/507. ARM passing-window maxima were 26.0 / 46.8 / 31.0 ms (100 ms limit, zero retries). Both platforms passed build/types/unit/guard checks and actual vanilla/React tarball consumers. Raw reports, all four CI rounds, 640 unique journey IDs, source hashes and archive checksums were independently verified. No additional local series or final 100-round acceptance was started.

---

The post-merge macOS ARM run blocked a healthy manual-tabs window with a real 126.7 ms observation gap (run 41091449659). Its snapshot took 126.3 ms, with only 3.9 ms in the separate final evaluation. The delay before that evaluation was not isolated to a particular runner or browser cause.

Read state synchronously in the final fresh locator registration and send immutable contract metadata once per journey. This removes the extra host request after resolution. Preserve fresh Playwright queries, coherent browser-clock snapshots, context checks, complete cleanup, the 100 ms limit and zero retries. Add per-locator timing diagnostics. macOS ARM CI now requires three full rounds, stops at the first failure and saves every raw report; Linux retains its complete suite. Both platforms test actual tarball consumers.

Validation at 7e48291:

• CI 41097726279: macOS ARM 3 × 169 = 507, Linux 169; all stages and both clean package consumers passed. ARM passing-window maxima: 44.0 / 69.1 / 48.6 ms; Linux: 29.4 ms.
• Local: four complete rounds (676 tests) plus 169 four-worker checks. Five raw reports, source hashes, semantic signatures and 800 unique journey IDs independently verified; no retries or skips.
• Build/types, 39 unit tests, 52 guard subprocess cases, 19 focused regressions and six semantic benchmark pilot cases passed. Real injected 150 ms gaps still block activation. The native reference independently performs the same transport work; pilot results are not release performance evidence.

The final commit 88e92fb only updates documentation/evidence and uses [skip ci]; runtime, tests, verifier and workflow are identical to the successful CI commit. See source-bound evidence and retained failures.

This short sample supports the change but cannot guarantee scheduling latency on shared hosts. Historical proofs retain their original hashes. Final 100-round acceptance, full performance measurements and publication remain deferred.