Can We Understand How Large Language Models Reason?

How about we first determine _if_ they reason