PR 30737 Linux dev verification
Main: 8b44f42b14c8e273156474a924ccf29680dd7857 (github/main)
PR: f410444d06f2b673df736a75324a742fb61464ae (exact PR head)
No PR source edits or commits made. No mocks or seeded app data.

Developer-tool behavior (real commands, dedicated Blacksmith workers):
$ vp run knip:check  # main; sample actual knip processes every 200ms
OBSERVED 0.21 files
OBSERVED 4.41 exports
EXIT 0 SECONDS 14.61

$ vp run knip:check  # PR; same process observation
OBSERVED 0.21 files,exports
OBSERVED 4.81 exports
EXIT 0 SECONDS 10.15

Both passes overlap on the PR; main runs them sequentially. These are individual runs on separate 4-vCPU workers, not a controlled speedup benchmark.
$ npm run knip:check  # PR
vp run: 3/3 successful.
$ vp test run --project root scripts/check-changed.test.ts  # PR
15 passed.
$ vp run check:changed e68e19d605485db6ef31b2033f67a9d07634fc2b  # PR
pass unused patches
pass format
pass lint
pass knip 1/2
pass knip 2/2
pass typecheck (40 packages)
pass test (2 files; 16 tests)

App integration (Playwright from repo, /usr/bin/google-chrome):
Main: vp run dev --home-dir /home/user/main/.t3 --no-seed
serverPort=16733 webPort=8693
Added /home/user/verification-repo through Add project > Local folder.
Selected Codex gpt-6.1-sol in composer. Sent a shell-read request through web UI.
Completed: heading verification workspace; branch master.
Stopped captured main process group before starting PR server.
Copied stopped app-created state into isolated PR .t3.
PR: vp run dev --home-dir /home/user/t3code/.t3 --no-seed
serverPort=13773 webPort=5733
Reopened existing thread. Codex gpt-6.1-sol follow-up completed in 14s.
Provider appended a requested README line using shell; git diff confirmed +1.
Clicked Open diff: checkpoint diff showed exact added line. Split view also worked.
Reloaded client: conversation and checkpoint diff remained available.
Both dev servers were background processes. Only one ran at a time.

Limits:
No product behavior changes in this PR: app flow is a regression check; CLI process overlap proves the changed behavior.
No native desktop/mobile, macOS/Windows, remote/tunnel client, controlled speedup benchmark, actual commit hook, or injected-error Knip cases exercised.
Initial check:changed attempts lacked a valid base (one mistyped ref, then shallow snapshot); corrected ref and explicit remote history setup allowed the real check to pass.