Weekly Codex quota reset/reconciliation appears broken — quota exhausted with no local activity

What issue are you seeing?

I appear to have hit a Codex weekly quota accounting/reset bug on ChatGPT Plus. I want to upgrade to Pro but these inconsistencies are blocking me.

Timeline (America/Chicago):

• Aug. 30: I legitimately used Codex.
• Last locally recorded activity was Aug. 30 19:37 CDT.
• At that point Codex's server-provided rate-limit data showed:
  • 94% 5-hour usage
  • 65% weekly usage
  • weekly reset: Sep. 5 16:24 CDT
• Aug. 31: the ChatGPT usage UI instead showed approximately:
  • 0% of 5-hour allowance remaining
  • 37% of weekly allowance remaining / ~63% used
  • weekly reset changed to Sep. 7 00:13 CDT
• I performed no intentional Codex work Aug. 31-Sep. 2 and paused my existing tasks.
• Sep. 1 account activity nevertheless showed approximately 62.3M tokens.
• By Sep. 2, the weekly allowance reached 0% remaining.

I inspected the local Codex state on the machine running Codex CLI 0.151.0:

Threads created or updated Aug. 31-Sep. 2: 0

2026-09-01: logs_2.sqlite rows=0
2026-09-02: logs_2.sqlite rows=0

2026-09-01: rollout_files_with_activity=0, token_count_events=0
2026-09-02: rollout_files_with_activity=0, token_count_events=0

Last local thread update:
2026-08-30 19:37:29 CDT

Last server-provided rate-limit snapshot:
weekly_used=65%
weekly_reset=2026-09-05 16:24 CDT

The suspicious part is that the weekly reset anchor changed from Sep. 5 to Sep. 7 while usage apparently remained around 63–65%, rather than resetting, and then the remaining quota depleted despite no corresponding local Codex session activity.

This could be delayed reconciliation across a reset boundary, server-side activity not represented locally, or incorrect account-side usage attribution.

I have attached a sanitized diagnostic report generated from state_5.sqlite, logs_2.sqlite, and rollout metadata. It contains no prompts, responses, paths, UUIDs, repository names, or tool arguments.

Could someone check the server-side usage ledger around Aug. 30-Sep. 2 and determine what consumed the weekly allowance after the reset/window change?

What steps can reproduce the bug?

Doing nothing. My weekly quota got exhausted while I still have 100% 5hr limit with a full reset on September 7

What is the expected behavior?

My quota should not be draining when I'm doing nothing.

Additional information

quota_report.txt

Start with the attached quota_report.txt and the referenced state_5.sqlite, logs_2.sqlite, and rollout metadata, comparing the Aug. 30–Sep. 2 activity with the server-provided rate-limit snapshots. Done means identifying what consumed the weekly allowance or confirming that the discrepancy requires server-side ledger investigation.