What should a budget receipt include when an AI coding run is paused?

[Keesan12]

Teams keep talking about limits, quotas, and token burn, but the missing thing in practice is usually the receipt a human sees when a run is paused or blocked.

A useful budget stop should probably say more than just over budget. My current mental model is that a good receipt should include:

• current spend in the active window
• attempted call estimate
• budget window and reset boundary
• whether the decision was block, downgrade, queue, or pause
• the decision or order id if multiple near-cap calls raced
• enough context that an operator can tell whether the run should be resumed later or redesigned

The reason I care is that vague budget failures get treated like flaky model behavior, and then people debug the wrong layer.

If you have actually run coding agents in production or even just let them work unattended for a while, what fields would you want in that receipt so the stop reason is explainable and actionable?