Historical Codex Desktop report: recent chat history unavailable with "without an active turn id" steering error

What version of the Codex App are you using?

Historical desktop app version: unknown. The original conversation's
21 August 2026 session-start metadata identifies Codex Desktop and CLI
0.148.0-alpha.9. This does not establish the desktop version or the version
active during the incident on 24 August.

For current context only, a local package-metadata check on 7 September found
OpenAI.Codex 26.901.1978.0 (x64). This was not read from the About dialog and
is not asserted to be the incident version.

What subscription do you have?

The reporter's supplied 7 September screenshot shows Pro. The subscription
active during the 24 August incident has not been independently recovered.

What platform is your computer?

Current measured environment: Windows 10 Pro 22H2, build 19045.6466, x64.
The exact OS build during the incident is unknown.

What issue are you seeing?

This historical report was drafted by Codex from the reporter's retained records.
The observations below are user reports, not independently measured host telemetry.

On 24 August 2026, the reporter could no longer inspect recent history in an
original chat and reported a steering error in a separate recovery conversation:

• 19:39:36.372 UTC: reported the displayed error below. The conversation ID
  is redacted; the remaining wording and line break are preserved.

Error creating chat
Cannot steer conversation [REDACTED_ID] without an active turn id

• 19:39:59.073 UTC: reported that the original chat was still going.
• 19:40:12.321 UTC: reported that the displayed history for that day was
  gone and they could no longer see what the original chat was doing.

These are recovery-message timestamps, not exact failure times. The impact was
loss of visibility into the work and the need to start a recovery conversation.
Continuing backend execution and permanent deletion of persisted history are
not established.

What steps can reproduce the bug?

No controlled reproduction has been performed. The exact triggering action,
recurrence rate and incident desktop build were not recovered. The timeline
above is evidence of the report, not a reproduction recipe.

A later literal search of the original conversation transcript did not recover
either exact steering-error phrase. The error wording therefore comes from the
retained user message in the recovery conversation.

What is the expected behavior?

Recent chat history should remain accessible. An attempted interaction should
either reach the task or explain its current state and provide a clear recovery
option without making the work inaccessible.

Additional information

Related reports were checked: #38609 describes a steering error with a different
suffix and an idle macOS target; #40674 describes disappearing active Windows
chats; #24217 describes an active display despite a reportedly idle backend.
These may overlap, but a shared cause has not been established. Please link or
deduplicate this report as appropriate.

Which minimal diagnostics would help distinguish display loss, a stale active-turn
reference, and a backend task that actually stopped if this recurs? Original
identifiers and narrow provenance can be considered for a separately reviewed
follow-up. No full session, private repository, screenshot or raw log is attached.

The report names no source file, test, or entry point; start by reviewing related reports #38609, #40674, and #24217 alongside the recovered timeline. Done means establishing a controlled reproduction or diagnostics that distinguish display loss, a stale active-turn reference, and a stopped backend task.