[PerchAnton]

The post-merge ARM run at 02ee0953 blocked manual Home/End on a 109.8 ms observation gap. Its slow snapshot took 98.514 ms; the preceding fast snapshot was followed by a deliberate host pause of up to 10 ms.

Start the next asynchronous browser read immediately after checking the previous snapshot. Reads remain sequential and locators resolve afresh. Preserve the 100 ms bound, actual browser timestamps, scenario budgets, component-failure priority, cleanup and zero retries. Positive waits retain their bounded pauses. This uses more observations during each short window and cannot guarantee a deadline on a stalled host.

The new timer-order regression passes with this change and fails when the former 10 ms pause is restored. Both injected 150 ms transport delays still block and stop activation. Match the native benchmark scheduler; preserve prior performance numbers as historical evidence until remeasured.

CI at 7a60938 passed:

Platform	Browser tests	Maximum passing-window gap
Linux	170/170	32.2 ms
macOS ARM, round 1	170/170	35.8 ms
macOS ARM, round 2	170/170	45.2 ms
macOS ARM, round 3	170/170	31.0 ms

Independently checked four raw reports, 680 executions, 644 distinct journey IDs, zero retries/skips, semantic signatures, source hashes, 52 guard subprocess exit codes per platform and both actual tarball consumers. The prior failed main run is retained in the evidence.

Local validation: build/types, 39 unit tests, 17 release-tool tests, six focused browser regressions and a six-case native-reference pilot passed. The pilot validates equivalent behavior and negative controls, not performance. The updated documentation tarball also passed clean vanilla/React consumers, ESM/CJS types/imports and the expected negative React exit.

The follow-up evidence-only commit uses [skip ci]; all runtime/test/verifier input hashes still match the successful CI source. No further full suite was repeated for documentation. At PR preparation, final 100 rounds, a new local four-worker comparison, merge and publication had not been performed. Historical benchmark and final reliability records cannot certify the changed source.

Post-merge verification — 2026-10-31:

Merged KeyTrail #13 at 1b42666fddf8c6653ae3898cbe99a61c5b671980 and gitwork #10 at 01de4a2fcd9d10592eebcd88f1a3ac5f749ee088. Both merge trees match the reviewed PR heads.

KeyTrail main CI passed on the exact merge commit:

Platform / round	Browser tests	Maximum passing-window gap
Linux / 1	170/170	26.3 ms
macOS ARM / 1	170/170	47.4 ms
macOS ARM / 2	170/170	38.4 ms
macOS ARM / 3	170/170	39.5 ms

All 680 executions passed without retries or skips. Independently checked four raw reports, 644 unique journey IDs, semantic signatures, actual environments, current source hashes and the unchanged 100 ms bound. Both platforms passed build/contracts, all 52 guard subprocess cases and the real vanilla/React tarball consumers, including the expected negative React process.

The tested archives each contain 103 files and are 360802 bytes. Linux SHA-256: cc7994bc7027d4c2f9e64affece7362cbf2a365eb06e899b377cfd59191ffc85; ARM SHA-256: fc02202f2cbe89921c28c600ae56f15658f08cff1ea74a36f7aa13c4730f4d31. The raw reports, archives and independent validation are retained locally.

Gitwork has no configured workflows; its merge commit has no workflow or check runs. No final 100-round series, new benchmark run, publication or additional commit was performed during this merge verification. Previous performance and final acceptance evidence remain historical for their recorded sources.