Testing a learned model by replaying past observations to verify it correctly predicts all previous outcomes.