Re-running a fixed set of test scenarios every time code changes, to confirm existing behavior hasn’t broken. In a simulation context, that means replaying the same market conditions against a new version of an algo and comparing results to the last known-good version.