Mobile task history is ~1 day behind desktop, then fails to reconnect; messages remain queued

What version of the Codex App are you using (From “About Codex” dialog)?

Desktop: 26.901.51231 (build 8109), verified from the installed ChatGPT.app bundle. Mobile app version not yet recorded.

What subscription do you have?

ChatGPT Pro

What platform is your computer?

Darwin 25.3.0 arm64 arm (uname -mprs on the Mac). Affected remote client: iPhone; exact iOS/mobile-app version not yet recorded.

What issue are you seeing?

The same long-running Codex task shows different conversation history and execution state on desktop and mobile. The Mac UI is up to date and can continue local work, but the iPhone repeatedly shows turns from approximately a day earlier. This has persisted across multiple observations on 8–9 September 2026.

By 9 September, approximately 07:26–07:31 UK time (06:26–06:31 UTC), the mobile UI displayed these exact messages:

• "Codex server returned an error."
• "This task couldn't reconnect"

A mobile message, "What's happening now?", remained marked "Queued". The stale conversation showed "Confirming user pause persistence" and an old explanation that the migration was paused, while the desktop was showing later work.

Impact: I cannot reliably monitor progress, determine whether the task is running or paused, or know whether my messages and approvals have reached the active task from my phone. I need to return to the Mac to coordinate a long-running migration. This is not just a cosmetic delay.

What steps can reproduce the bug?

Feedback ID: 01a068fc-e70c-7a92-b84e-a3bd6d01e644

What is the expected behavior?

Desktop and mobile should show the same latest turns and task state. A mobile message should be delivered to the current task, or clearly fail rather than remain indefinitely queued. Reconnection should resume from current history instead of repeatedly presenting an old pause state. If the connection is unavailable, the UI should distinguish cached history from current execution state.

Additional information

Screenshots documenting the stale turns, queued message and both errors are available. No private migration files, credentials or full session transcript are included in this public draft.

Additional diagnostic observation from 9 September: the app's read_thread operation also returned stale recent-turn data for this task. Its task metadata had an update time of 2026-09-09T08:16:38Z, but the newest returned turn was marked inProgress with a start time of 2026-09-08T10:10:51Z and contained the old "Confirming user pause persistence" message. The current desktop conversation had progressed beyond that turn. This suggests the problem may extend beyond the phone's display cache, but the cause is unconfirmed.

Please investigate task-history synchronization/reconnection using the feedback/session ID above and advise whether there is a safe recovery that preserves the existing task and its approval history. I have not deleted or recreated the task as a workaround.

Start with feedback/session ID 01a068fc-e70c-7a92-b84e-a3bd6d01e644 and compare the desktop task with the mobile read_thread results, including the stale turn and task metadata timestamps. Reproduce the reconnect and queued-message behavior if possible; done means mobile and desktop show current history and state, messages do not remain indefinitely queued, and unavailable connections are clearly reported.