The practice of having a model explicitly output its internal thought process and problem-solving steps rather than keeping them hidden.