Training signal derived from intermediate and final results of experiments, used to improve decision-making in long-horizon tasks.