Compression-Aware Intelligence is the framework that treats hallucinations and contradictions not as errors to eliminate, but as measurable signals of representation strain inside any cognitive system, and uses those signals to guide stability, coherence, and self-correction TLDR: when a model receives two prompts that are semantically equivalent, it should produce functionally equivalent outputs. if small wording changes cause noticeably different outputs, that means the model is storing meaning in unstable or mismatched latent representations. hallucination happens when the model tries to reconcile conflicting latent states during generation. www.compressionawareintelligence.com