I use ChatGPT Work schedules to automatically organize email. A scheduled run can wake up but fail to complete because of usage limits. From my experience, these incomplete attempts also seem to consume or affect the remaining five-hour and weekly allowances. I do not have per-run usage records to verify this, so I would appreciate clarification on how scheduled wake-ups, quota checks, and runs stopped by usage limits are counted.
I would like to request the following scheduling controls:
- Check available usage before starting Work. If there is insufficient allowance, do not launch the Work task. The check and the skipped run should not consume Work/Codex quota.
- Let users set minimum remaining usage thresholds. For example, start a particular schedule only when at least 80% of the five-hour allowance AND 20% of the weekly allowance remain. Both thresholds should be adjustable.
- Defer the task and check again every hour. If either threshold is not met, keep the task pending and recheck every 1 hour. Waiting and checking should not launch a full Work session or consume its quota. Run the original task once both conditions are met.
- Allow regular ChatGPT scheduled tasks to read usage and control Work schedules. With user authorization, they should be able to read the remaining Work/Codex five-hour and weekly allowances and enable or pause a specific Work schedule.
- Make the status visible. Show why a task was deferred, when it will be checked again, and how usage is counted, so users can distinguish a postponed task from a completed one.
For email organization and similar recurring workflows, this would help avoid repeated wake-ups that produce no completed work, while allowing users to reserve capacity for other tasks.
This is a feature request. The apparent quota consumption is an observation that needs confirmation, not a claim of a verified accounting bug or an individual refund request. If others have encountered similar behavior, I would welcome observations that do not include personal account information.