Every hour the scenario collects token usage and cost across all AI runs and files receipts into a table — so you see what each one costs.