A QA spending limit should control the actual launch of paid work
Writing a budget in an issue does little if an overnight run keeps starting purchases after a failure. In your own tooling around SimAggregator, connect permission to start paid actions with the budget of that specific test. This is a team-side control, not a promised platform feature.
Before starting, name an owner, a spending ceiling, and a stop rule. Separate confirmed costs from operations whose outcome is still unknown. A missing reply does not prove that no charge occurred; pending reconciliation should affect whether work can continue.
Test the stop mechanism with synthetic cases. Once the threshold is reached, new paid actions stop, while information about existing work remains available. Concurrent jobs need shared accounting so they do not each spend the same apparent headroom.
The final report should show the test objective, useful findings, known spending, and unresolved reconciliation. Raising the budget should be a deliberate owner decision, not a side effect of restarting a failed job. The team can then decide which parts of the next run are better covered by local simulation.