Training a model using detailed reasoning traces as supervision to teach it to follow correct reasoning steps.