Modifying raw reward signals to guide learning more effectively without changing the underlying task objective.