PR #33954: real Linux development verification, 2026-11-10 UTC
baseline github/main: 17a6a01d84e3adf9f29df2639d6b6079a434e1b2
PR exact head: 4a67ea29de284449a36c63f350d8b5498bcf07f4

$ git worktree add ../main github/main
$ vp i
$ setsid nohup vp run dev > /tmp/dev-main.log 2>&1 &
[dev-runner] serverPort=16733 webPort=8693 baseDir=/home/user/main/.t3

Playwright 1.60.0 from repository dependencies; /usr/bin/google-chrome.
Added local project through UI, using a real shallow clone of
dazbot-ooh/t3code-contributions. origin points to pingdotgg/t3code;
main tracks contributor/main. No mocks, fixtures, or replay provider.
Codex gpt-6.1-sol turn ran git status --short and git branch --show-current:
completed in 13s, response "current branch: main".

Opening thread details > Git action options dispatches:
{"_tag":"Request","tag":"vcs.refreshStatus","payload":{"cwd":"/home/user/scenario-repo"}}
Measured from the actual browser WebSocket request to matching successful Exit.
This refresh invalidates the PR lookup epoch; these are uncached refreshes.
main: 5234 ms, 5670 ms, 4766 ms; median 5234 ms.
All three responses: refName=main, isDefaultRef=true, pr.number=31793,
pr.state=open, headRef=main, baseRef=main, hasWorkingTreeChanges=false.

Stopped main by its captured process-group PID 84375 before starting PR.
Copied only isolated, app-created dev state to preserve project/thread data.
$ git switch verify-pr-33954
$ git rev-parse HEAD
4a67ea29de284449a36c63f350d8b5498bcf07f4
$ vp i
$ setsid nohup vp run dev --home-dir /home/user/t3code/.t3 > /tmp/dev.log 2>&1 &
[dev-runner] serverPort=13773 webPort=5733 baseDir=/home/user/t3code/.t3

Same UI flow, repository checkout, completed provider turn, and payload:
PR: 1259 ms, 1363 ms, 2051 ms; median 1363 ms.
All three responses identify the same open fork PR #31793.
Median round trip was 3.84x faster in these three sequential samples.
Includes the full client/server refresh, not only GitHub HTTP time.
Recordings are real-time; captions report these observed WebSocket timings.

Second real Codex gpt-6.1-sol turn completed on the PR server in 8.9s:
"branch: main. checkout is clean."

Neighbor checks via actual UI branch picker and Git action options:
1. perf/github-head-owner-lookup tracking yuruzuu/perf/github-head-owner-lookup:
   successful refresh 6610 ms; isDefaultRef=false; open PR #33954;
   correct title/headRef/baseRef and clean checkout. PR card visible in UI.
2. main tracking origin/main (pingdotgg): successful refresh 1223 ms;
   isDefaultRef=true; pr=null; clean checkout. Old PR card cleared in UI.

$ blacksmith-check run --id <task-worker> "vp test run apps/server/src/git/GitManager.test.ts packages/source-control-github/src/server/GitHubSourceControlProvider.test.ts"
2 test files passed; 168 tests passed.
$ blacksmith-check run --id <task-worker> "vp run check:changed 20c3dd3689782bfcb1a04332086d08017028c718"
Passed format, lint, both knip checks, both GraphQL codegen checks,
typecheck of 19 affected packages, and 176 tests across 4 files.
Worker history was fetched to depth 2 so comparison to the PR parent worked.
GitHub checks on the exact head were green; platform-inapplicable jobs skipped.
No source changes, commits, or pushes were needed.

Not exercised: native desktop/mobile shells, GitHub Enterprise, deleted forks,
closed/merged PRs through the live app, and simultaneous multi-project contention.
Closed/merged decoding and ordinary/default-branch routing have existing test
coverage, but that is not live-client evidence for those unexercised scenarios.