Unexpected weekly usage reset overwrites active Codex window and moves reset date — third occurrence

What version of Codex CLI is running?

codex-cli 0.147.0

What subscription do you have?

ChatGPT Plus

Which model were you using?

gpt-5.6-sol with reasoning set to High

What platform is your computer?

Ubuntu Linux

What issue are you seeing?

On 2026-08-13 at approximately 10:30 CEST, my Codex weekly usage unexpectedly returned to 100% remaining even though I did not intentionally apply or redeem any usage reset.

Immediately after the event, /status showed:

OpenAI Codex (v0.147.0)
Model: gpt-5.6-sol (reasoning high, summaries auto)
Account: ChatGPT Plus
Session: 019ffa3e-6ddf-7af0-8873-48136545ac09
Weekly limit: 100% left (resets 10:30 on 20 Aug)

This is the third time I have observed an unsolicited full usage reset.

The important problem is not that extra usage was granted. The problem is that the reset appears to replace the currently active weekly window with a new seven-day window starting at the time of the unsolicited reset. That changes the previously displayed reset schedule and makes it impossible to plan usage reliably.

I had been deliberately conserving the remaining weekly allowance based on the reset date shown by Codex. If an unsolicited reset starts a new seven-day window immediately, then conserving usage can be counterproductive: unused capacity from the active window is effectively superseded rather than preserved or added separately.

I did not intentionally trigger a banked reset before this happened.

What steps can reproduce the bug?

I do not know the server-side trigger, so I cannot reproduce it deterministically.

Observed sequence:

1. Use Codex on a normal ChatGPT Plus weekly usage window.
2. Have some amount of usage remaining and a future reset timestamp shown in /status / the usage UI.
3. Do not manually apply a usage reset.
4. At an apparently unsolicited time, the weekly allowance returns to 100%.
5. The reset timestamp moves to exactly about seven days after the unsolicited reset event.

On this occurrence, the new state was:

100% left
resets 10:30 on 20 Aug

observed at approximately 10:30 on 13 Aug 2026.

What is the expected behavior?

Unsolicited promotional, service-recovery, milestone, or account-level reset grants should not overwrite an active weekly usage window without an explicit user choice.

Preferably:

• additional resets should be banked and shown as available to redeem later;
• the active weekly reset timestamp should remain unchanged until the user chooses to apply a reset;
• if a true hard reset must occur, the UI should clearly explain what caused it and whether unused allowance from the prior window was preserved;
• the usage interface should expose a reset-event history showing the source, timestamp, previous balance/reset time, and resulting balance/reset time.

A user who deliberately conserves usage should not lose the value of that planning because an unexpected reset silently replaces the active quota period.

Additional information

This appears related to the broader reset-window predictability problems discussed in:

• #9508 — Make Weekly Limit Reset Deterministic
• #28246 — Quota windows anchor to first post-reset use, causing lost subscription time
• #30816 — Weekly usage reset date changed unexpectedly after subscribing to ChatGPT Plus
• #36471 — reports unexpected weekly usage returning to 100% as an account-state anomaly

I am happy to provide screenshots or additional local telemetry privately if useful. I have intentionally omitted my account email from this public report.

No repository file or test is named, and the reporter cannot reproduce the server-side trigger. Start by comparing the /status behavior with related issues #9508, #28246, #30816, and #36471, then determine whether the reset-window behavior is implemented in this repository. Done means the active window and reset cause are reliably identified and the expected handling is defined.