# Evidence audit PASS 2

AUDIT PASS — no corrections required.

Target: [#24271](https://git.cafe/pingdotgg/t3code/pulls/24271), [GitHub counterpart](https://github.com/pingdotgg/t3code/pull/24271).
Audited final HEAD: [1bee62f103](https://github.com/pingdotgg/t3code/commit/1bee62f10398d27e8bd669b4bc917eee24d5c5e5).

## Independently confirmed

- GitCafe live PR head is exactly 1bee62f10398d27e8bd669b4bc917eee24d5c5e5, source branch t3/mobile-audit-toolbar-20261023, target main, state open. Read returned no warnings.
- Live description exactly equals /tmp/ev-24271/pr-body.md. It starts with the complete original description from original-pr.json, byte-for-byte as decoded text. The original seven-test statement is preserved historical text; Takeover explicitly explains 7 → 4 → 6 and names the final total. No blocking contradiction under the requested original-body preservation constraint.
- Final commit merges reviewed head 3526fde95203012921d319d9cfe50af35305be01 and main 27e0c684c585f955252d65f06206dd11a143bff5. Toolbar module and test have no changes between reviewed and final heads. Relative to merged main, only NativeHeaderToolbar.tsx and NativeHeaderToolbar.test.tsx differ.
- Source publishes per-instance handler storage in a layout effect and clears it on cleanup. Button/direct-menu/nested-menu registrations use dispatchers reading committed storage. The three placement fixtures retain actions through a handler-only update and assert current captured values with unchanged mocked setOptions counts. Suspension and isolation/unmount fixtures have the narrower button scope described in fixture-audit.md. Custom-header coverage uses a synthetic context, not a real navigation container or device.
- Android NativeHeaderToolbar.android.tsx implementations return null; AndroidScreenHeader and FilesHeader use separate in-flow NativeAppBar/IconButton/AndroidAnchoredMenu paths. The PR does not claim Android device proof or runtime coverage of that path.
- Independently reran from apps/mobile: vp test run src/native/NativeHeaderToolbar.test.tsx. Exit 0, one file, six tests passed, 09:28:06 UTC on 2026-10-24, duration 1.02 seconds. Checkout clean before and after inspection.

## Evidence boundaries

refreshed-checks.log records dependency refresh and passing mobile typecheck, six tests, scoped lint, scoped format, and diff check at the final head. This pass inspected that log and independently reran only the six tests. The dependency-refresh explanation is compatible with the merge and log; this pass did not reproduce the initial stale-install failure.

Earlier original-fix revert failures, deliberate mutation detection, and two reviewers’ MERGE verdicts remain attributed prior reports; this pass did not reproduce those actions or certify their original logs. The body explicitly distinguishes pass-1 independent measurements from earlier mutation reports.

iOS and Android device surfaces remain UNVERIFIED. No native bridge-registration, native performance, screenshot, or recording proof is claimed. JavaScript callback tests do not establish those properties. Platform limitations are disclosed and are not blockers under the task instructions. Historical physical-iPhone wording is preserved original text, not fresh proof from this pass.

## FRICTION

No new blocking friction. Local device tooling/runtime limitations remain as disclosed; no device tooling search or capture was repeated. GitCafe read succeeded with ~/.bun/bin on PATH. No source edits, pushes, comments, watchers, servers, browsers, devices, other machines, live userdata, credentials, permission changes, or evidence commits. Mandatory T3 PR registration was performed; no PR content was changed. This is an evidence audit, not merge authorization or CI certification.