Manual real-app comparison, not a test result. Base: 9441a7285c5f3466e743d6fa1489fbefed9043f1 Head: 6d132628840ed883d0ac7438e226f62c93ede371 Web client: Chromium, 1280x800 CSS pixels, light theme. Independent fresh T3 homes, synthetic workspace, authorized provider API credentials. No production database was seeded. Codex and Claude controls returned OK on both commits with proxies passing traffic. Codex outage: upstream CONNECT refused with 502, provider alive; periodic provider retry notifications continued beyond five attempts. Base Codex: first retry 14:29:40.488 UTC; still running at captured 12m18s, attempt 19. Head clean repeat: first retry 14:42:23.694 UTC, terminal failed 14:52:23.715 UTC, attempt 18; retry window 600.021 seconds. Claude outage: initial CONNECTs refused until retry notices appeared; later CONNECTs held open without answering. Both replacement runs reached attempt 8/10 and then 9/10. Base Claude replacement: first retry 14:36:25.305 UTC; still running in captured UI at 12m23s, attempt 9/10. Head replacement: first retry 14:36:20.849 UTC, failed 14:46:21.544 UTC, attempt 9/10; retry window 600.695 seconds. This establishes the missing Claude ten-minute deadline. It does not establish an infinite Claude hang: its native attempt limit remains relevant. Codex recovery on both commits: refused initial requests, then restored pass mode before deadline; reply OK at 14:35:13 UTC. Still completed more than ten minutes after the initial retry window. Codex user Stop on both commits: clicked visible Stop during provider retries. Both showed Run interrupted / Stopped by user. Head interrupted at 14:47:51.908 UTC; base interrupted at 14:49:03.394 UTC. The initial head Codex run was delayed by capture-related server event-loop stalls and is excluded from the clean timing verdict. The initial Claude comparison lost a proxy and is excluded. The same fault conditions were repeated with consolidated proxies and fewer browser tabs; successful replacement recordings are provided. The scripts are reproduction helpers; the observations above came from the real app UI and its persisted runtime state, not unit tests. No deadline constant or provider protocol was patched. Final UI inspection at 14:59:28 UTC: both Codex user Stop runs still showed Run interrupted / Stopped by user, beyond ten minutes from their first and last retry notices.