PR 29196 real runtime verification, 2026-11-02, Linux Boat VM
main: github/main 30084f0018
PR: b7208c06beb1e4306f827fad807dae83303d89b8
Client: repo Playwright 1.60.0, /usr/bin/google-chrome
Provider: real Codex gpt-6.1-sol, local proxy
Both checkouts: vp i; background dev servers, sequential, task-owned state.
App-created project/threads; Codex created 3,221,225,472-byte untracked binary.
Actual git capture timeout; no mocks, replay or injected errors.

Measured server spans along UI -> orchestration -> Codex path:
main follow-up 1: startRootRun 30215.8ms; Codex start at +30173.0ms.
main follow-up 2: startRootRun 30209.3ms; Codex start at +30173.5ms.
PR initial baseline: 30130.0ms Failure; startRootRun 30225.2ms.
PR initial post-turn: 30123.3ms Failure.
PR next baseline: skipped after two timeouts; startRootRun 53.8ms;
  Codex start at +32.9ms, returned CAPTURE_TIMEOUT_PROBE in the web client.
PR post-turn probe still ran: 30061.8ms Failure.
PR repair turn: baseline skipped; Codex excluded binary in .git/info/exclude.
  post-turn capture 82.0ms Success.
PR fresh thread: baseline 80.6ms Success, startRootRun 200.2ms.
PR healthy edits: README diff visible, post-turn capture 75.6ms Success.
App-created isolated worktree: healthy snapshots and visible note.txt diff.
UI Edit from here > Revert files too restored note.txt from
  "isolated second snapshot" to "isolated first snapshot" and rewound chat.
Shared-checkout restore was refused by existing isolation guard; file preserved.

Blacksmith: 63 focused checkpoint/capture/run-execution tests passed.
check:changed against 47f2c8d29e98: format, lint, knip, GraphQL,
  affected typechecks, and adjacent tests passed.
Existing CI: all 31 cafe check contexts successful at tested head.
Two independent scoped read-only code reviews: no blocking findings.
Unverified: native desktop/mobile, Windows/macOS, reporter exact workspace,
  other providers, and original follow-up-steer symptom (unchanged by PR).