Reward signals that adapt based on the current stage of a task, focusing learning on the most critical interaction moments.