[PerchAnton]

Independent target and role reads previously waited for each other. Resolve them concurrently while awaiting every read before observing DOM state or disposing handles. Preserve deterministic error ordering, fresh locator resolution and the existing keyboard/time limits. Add delayed-sibling cleanup and five-value canary regressions for every journey type.

Add a source-bound repeatability verifier and a fixture-specific native Playwright benchmark. The recorded checkpoint contains 45 complete sequential rounds of 153 tests (6,885 executions), without retries or semantic differences, plus a matching final four-worker run. The maintainer accepted this reduced scope; the interrupted next round is excluded and the original 100-round criterion remains explicitly unmet. A version-bound policy and package gate enforce the recorded scope and source hashes.

The benchmark uses 5 warmups and 30 alternating measured pairs per workload, semantic equivalence checks and three negative controls. Paired median overhead is 8.7% for dialogs, 0.8% for manual tabs and 4.8% for forms; each also meets the 20% target by ratio of medians. README graphics link to raw pairs, median/p95, environment and limitations. These synthetic fixtures do not establish universal performance or accessibility claims.

Validation: strict types; 30 unit tests; 48 actual subprocess exit-code cases; 6,885 sequential fixture executions and 153 final isolation tests with 6,900 unique journey UUIDs; clean offline tarball installation, ESM/CJS imports and declarations, six browser examples and three guarded journeys. Final Linux/macOS CI passed on 94dccde: https://github.com/PerchAnton/keytrail/actions/runs/40747597014.

Source candidate 0.1.0-alpha.5; no package or release is published. DEV10 remains separate.