Incident replay
Real fleets have burned hundreds to thousands of dollars overnight when an agent fell into a loop. Pick a documented runaway pattern and run it against a Stub budget, then watch the cap hold while the overspend is refused, transaction by transaction.
Context: A real startup watched this pattern bill about $72,000 overnight, because the cloud budget was an alert, not an enforced limit. Stub caps 30 runaway calls against a $1.50 budget on the live Aurora DSQL cluster: only what fits commits, the rest collide on OCC and lose with 40001.
Source: The Register ↗Each replay runs in throwaway accounts on the real cluster and self-cleans, so your ledger is never touched. The blocked dollars are what Stub refused in this run; the headline figures are documented real-world incidents, with a source linked on each scenario.