# Corrected evidence matrix — proof audit pass 2

Target: [#18537](https://github.com/pingdotgg/t3code/pull/18537), [#18537](https://git.cafe/pingdotgg/t3code/pulls/18537). Audited local branch `takeover/pr-18537`, application source only in `/home/ubuntu/wt/pr-18537`.

Before: replacement merge-base [7fd2e09e38](https://github.com/pingdotgg/t3code/commit/7fd2e09e381a61f427da09dd7a8f2c9c1558fe59).
After: clean committed [cefc1ace94](https://github.com/pingdotgg/t3code/commit/cefc1ace94bfe0d9f0bd97655a899fc4eccdffe3).
This replaces the stale provenance and pending statuses in the [pass-1 matrix](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/50aef7d7336f6c39/matrix.md). No uncommitted implementation is part of the proof. The before commit is confirmed as this final after commit's merge-base with the requested baseline.

Server-only override: runtime transcripts replace pixels, viewports, themes, screenshots and video. Environment: Linux fs-01 x86_64 kernel 6.1.102, Node v26.12.1, Git 2.43.0, actual dev runner server origin `http://127.0.0.1:15913`, web port 7873; browser/emulation/DPR/font/viewport/theme N/A. Same fixture topology and file contents, real upload-pack, 37s transport pause; fresh per-session homes and fixture timestamps/commit IDs intentionally differ. Each client starts with local and cached origin at A, remote main at distinct B.

| Row | Fixture | Affected | Before at 7fd2e09e38 | After at cefc1ace94 | Audit status |
|---|---|---|---|---|---|
| R1 | Healthy fetch, 37s transport delay, stale A / fresh B | Yes | Fails 30.032391s after transport start; clear timeout, no new worktree | Done 37.278359s after transport start; HEAD=B; independent replay also done with fresh B | PASS: before + after real launch/setup transcripts and independent replay |
| R2 | Upload-pack starts then blocks indefinitely without refs | Yes | Redundant with R1-before's 30s timeout | Failed 300.051576s after transport start; pending through 270s; clear timeout, no worktree | PASS: real production timeout; wrapper has no deadline |
| T1 | Deferred/TestClock healthy unscoped fetch | Yes | New test fails with timeout instead of success | Pass after virtual 31s | PASS: baseline failure log, final full-file success, independent focused pass |
| T2 | Deferred/TestClock hung unscoped fetch | Yes | New test fails because already settled by 299s | Pending at 299s, fails at 300s with `Git command timed out.` | PASS: baseline failure log, final full-file success, independent focused pass |
| S1 | Scoped worktree-base fetch for tracking local base | Yes | 30s default | Shared 300s options | PASS: R1/R2 through launchThread -> fetchWorktreeBaseCommit -> fetchRemote; source confirms explicit remote base uses same helper |
| S2 | Unscoped fetch, missing-scoped-ref all-refs fallback, ref-lock retry | Yes | Existing fallback/retry behavior | Same branching with 300s per invocation | PASS: shared options source; full 165-test run includes real missing-ref fallback and ref-lock tests; focused retry/errors also rerun |
| U1 | No fetch: startFromOrigin:false, root/existing-worktree, no fetchable remote | No | Existing skips | Same skips | UNCHANGED: source gating inspected; no new capture needed |
| U2 | Immediate errors, missing/invalid refs/path, branch collision | No new branching | Existing behavior | Same control flow/error mapping; waits use new duration | UNCHANGED: source diff and adjacent existing tests; no exhaustive new capture claim |
| U3 | Direct vcs.createWorktree, checkout/removal, submodules/setup | No | Existing behavior | Same code | UNCHANGED: direct RPC would bypass regression; not used as proof |
| U4 | fetchRemoteBranch, fetchRemoteTrackingBranch, fetchRefCommit, push, status refresh | Separate paths | Existing deadlines | Same deadlines | UNCHANGED: source inspected; no claim that all Git timeouts increased |
| U5 | Clients, providers, relay/tunnel, layouts/themes/widths | No changes | Existing clients/contracts/transports | Same code | N/A: shared server behavior proven once; no UI/platform claims |

Evidence: [index](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a817abd4f5dc98ff/index.md), [R1-before](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/4430a91c7983d78a/R1-before.md), [R1-after](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/3b6612c08952966f/R1-after.md), [R2-after](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/e6f1976ade78292d/R2-after.md), [scoped tests](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b5b5f4c08c9fc2c1/scoped-tests.md), [reproduction recipe](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/9f1e34b6f47db346/REPRODUCE.md). Independent replay transcript and timing comparison accompany the pass-2 audit report.

What else changed: the complete specified before/after source diff is two files, 46 insertions: fixed timeout constant/comment + shared options entry; TestClock import + two parameterized cases in the existing test file. Increasing timeout above DEFAULT_TIMEOUT_MS also selects execute's existing path outside the short-command semaphore. Retry/fallback branches, diagnostics, other helpers, checkout/removal and client code are unchanged. No dependency or configuration knob change. This is a fixed 300,000ms limit PER FETCH ATTEMPT, not a five-minute total workflow deadline; ref-lock retry and public missing-ref fallback can each start another bounded attempt. The genuinely hung fetch fails rather than returning a retryable exit code.

FRICTION: original matrix was captured during implementation and named stale SHAs; corrected here. Remote original PR remains at the contributor SHA because publishing is forbidden. The independent auditor's first replay was interrupted by a dev watch restart while focused tests ran concurrently; that attempt is excluded and cleaned up, then rerun alone with a fresh home. No live userdata, browsers, application source edits, commits, pushes, watches or forge comments were used in this audit.

VERDICT: SUFFICIENT