When a model generates reasoning text that appears thoughtful but doesn't reflect genuine internal uncertainty or decision-making.