Measuring how well a human can read a reasoning trace is not the same as ensuring the model is being honest. Safety researchers often treat Chain-of-Thought (CoT) transparency as a structural given. The assumption is that if we can see the work, we can catch the error.