Internal representations of a model's awareness about its own knowledge, confidence, or reasoning quality.