Skip to content

Latest commit

 

History

History

pr-2279

README.md

PR 2279 acceptance evidence

PR 2279 routes eligible local iOS Simulator snapshots through the AX bridge and the existing TypeScript presenter, with typed XCTest fallback.

Final head: ae26f7afc064c75c65c71e677d32d77cdf2f9709. Matched base: 7a2d48d160aaacb582c6f4cde98b6af7531bf7af.

Matched warm corpus

120 measured snapshots per revision: 20 on each of the six existing #2189 screen definitions. Both corpora completed with no stops, failed samples, wrong-screen anchors, unreadable trees, or process-generation changes observed by admission. Fixture setup interactions also completed. This is bounded observed evidence, not a guarantee for every app or transition.

Screen Base CLI median / p95 ms Head CLI median / p95 ms Daemon median ms, base → head Failed samples, base / head
quiet 122.9 / 143.4 99.4 / 124.1 46.0 → 24.0 0 / 0
list 291.9 / 308.6 206.3 / 225.7 211.0 → 127.0 0 / 0
nested-scroll 164.5 / 197.1 109.2 / 166.0 83.0 → 32.0 0 / 0
alert 220.3 / 318.2 137.4 / 163.5 124.0 → 54.0 0 / 0
system-surface 188.0 / 279.5 133.7 / 154.3 106.0 → 51.0 0 / 0
xctest-stress 221.7 / 293.4 137.1 / 152.3 124.0 → 57.0 0 / 0

The driver reuses #2189's screen setup, sample admission, PID checks, deep-button controls, measurement extraction, and v1 result schema. It times the same fresh-process batch snapshot command on both revisions, sequentially on one dedicated iPhone 17 Pro / iOS 26.2 Simulator. State checks run outside the timed command. Host load was not controlled or randomized; retain all outliers and do not extrapolate these timings into a general speed guarantee. Runner-demand/open latency and proxy acceptance remain #2198 scope.

Every head screen has an initial debug trace proving bridge acquisition followed by TypeScript presentation. Measured responses retain the documented unavailable-hittability warning and no fallback warning. The base uses the prior XCTest route. Public snapshotDiagnostics backend counters are not used as producer proof; the request phase traces identify the bridge. Raw CLI responses are in each observations.json; statistics and all raw samples are in samples.json. Run python3 verify-corpus.py to recheck the published corpus. Process checks are performed by the recorded driver during admission; JSON verification does not recreate those live checks.

Foreground correctness and regressions

live-foreground-2931cbc944 repeats normal bridge acquisition and a same-process microphone permission dialog: typed foreground-owner-unverified fallback publishes exactly five system-dialog nodes. The production runtime is identical to the final head; provenance.json lists the intervening test-only paths. live-foreground-7cf6e7fd38 additionally retains screenshot and request proof of dismissal keeping the generation circuit disabled, followed by relaunch to a new PID and restored bridge acquisition.

validation/native-capture-guard-proof.md supersedes the earlier predicate-only record as guard-deletion proof. The real native capture entry point passes five scenarios. Removing PRE makes three cases fail; removing POST makes one case fail. Restoring both passes all five. Final-head iOS CI executed all five cases successfully; its excerpt is retained. Other planted-red records cover process identity and test selection. Both matched results retain the deep-button control: invalid shallow observation fails; full observation detects the changed descendant.

Gates, review and size

Final-head pnpm check:affected --run passed: 746 related test files / 5,691 tests, the additional selected suite, and all runnable repository gates. All final-head GitHub checks passed, including iOS, Android, macOS, Linux, Coverage, Integration Tests and Size. The fixture-build matrix was skipped; the runnable device checks passed. See validation/ci.json, validation/affected-ae26f7afc0.json, and iOS run.

The authoritative size run reports +35,845 bytes npm unpacked and +10,359 bytes tarball. Full reports are retained. The independent review accepted the owning route/foreground policy and identified the two now-resolved native-test gaps. The maintainer approved keeping required correctness and regression coverage together over the 1,000-line budget; final scope is 34 files / 1,353 gross lines. This exception does not waive validation. The implementation keeps routing/circuit policy in the Apple owner and reuses the existing presenter and XCTest modal resolution rather than adding parallel policy. Review context is retained in validation/review-context.json.

Historical diagnostics and provenance

historical/ retains earlier Settings-only diagnostic slices showing the repeated-discovery regression and its correction on the old base. These are not the final matched comparison and must not be combined with it. provenance.json records exact source revisions, fixture fingerprint, trusted artifact producer and hashes. No fixture binary is republished. SHA256SUMS covers the evidence files.