Pro 20x usage lockout followed by reset-date mismatch: Sep 10 vs Sep 15

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

26.903.61454, verified from installed app metadata on September 11. The version at incident time was not independently verified.

What subscription do you have?

ChatGPT Pro 20x

What platform is your computer?

macOS / Apple Silicon. uname -mprs: Darwin 25.6.0 arm64 arm

What issue are you seeing?

On September 9, my available Codex usage suddenly fell from about 70% to zero while I was running one GPT-6 Astra Medium agent, with no swarm or sub-agents. The app blocked further use. Allowance later returned, but the reset shown in Usage was September 15 instead of the September 10 date advertised in the lockout banner. The expected September 10 reset did not occur.

Please investigate both the initial lockout and the reset discrepancy. The remaining issue is whether my account missed an entitlement or the app showed dates for different limits without identifying them.

Timeline and evidence

All times are Europe/Paris, UTC+2, in 2026.

Time	Observation
Sep 9, 19:07	Initial report. The screenshot says I am out of Codex and Work usage and gives the next reset as Sep 10, 18:57.
Sep 9, 19:44	Apparent recovery. The screenshot shows 76% left, with a weekly reset of Sep 15, 10:40.
Sep 9, 22:59	Before/after screenshots. I corrected my recovery report after noticing the changed date.
Sep 10, 19:02	Follow-up. The previously advertised reset time had passed without the expected reset.
Sep 11	A fresh account-limit read still shows 32% remaining, resetting Sep 15, 10:40:49.

The exact displayed dates differ by about 4 days 16 hours. My X posts described this as a one-week delay. The earlier date came from the lockout banner; the later date came from the weekly Usage panel. That distinction needs investigation before concluding that the same weekly window was moved.

Current account-limit response

limitId=codex, usedPercent=68, windowDurationMins=10080, resetsAt=1789461649. The reset timestamp is September 15 at 08:40:49 UTC. Two banked resets remain available; none was redeemed during this investigation.

The screenshots establish a temporary loss of access and conflicting reset dates. They do not establish a permanent erroneous debit or the backend cause.

What steps can reproduce the bug?

This is an observed incident, not a deterministic reproduction:

1. Use one GPT-6 Astra Medium agent on Pro 20x with available allowance.
2. Receive an unexpected out-of-usage banner with a September 10 reset date.
3. Later see allowance return, with September 15 shown as the weekly reset.
4. Check after the September 10 time and find that the expected reset has not occurred.

No deliberate quota-consuming reproduction was attempted.

What is the expected behavior?

The allowance display, enforced access, and reset schedule should agree. If another limit blocks use, the banner should name that limit and show its reset time. Please reconcile the account history and restore any allowance or reset entitlement that the investigation finds was lost.

Additional information

OpenAI marked the September 9 unexpected usage limit resets incident resolved. Please check whether this account was affected and whether its reset state was fully reconciled.

Related reports: #44254 describes a macOS lockout followed by restored access and conflicting reset dates; #44210 describes a Pro 20x usage jump and a changed reset timestamp. This report adds the before/after screenshots and the subsequent missed September 10 reset. Maintainers can determine whether these belong to the same incident.

A reply to my thread reports changed reset dates on two Pro 20x accounts without banked resets. Those accounts have not been independently verified.

I am pursuing account correction and compensation through OpenAI Support for both the initial interruption and delayed reset. Private account and billing details are omitted here.

No repository file, test, or code entry point is named. Start by comparing the current account-limit response with the linked screenshots and the resolved September 9 incident, then review related issues #44254 and #44210; done means the conflicting lockout, access, and reset states are reconciled or the account-specific investigation is routed to the appropriate service.