... Contributions: Architecture C (Dual-Memory Reconstruction):  A completely  training-free  inference engine that combines a Layer 0 Gate with a Layer 12 Cache. It achieves  71.55% zero-shot top-1 accuracy at 84.0% token sparsity  on SigLIP, retaining 92.4% of dense accuracy without modifying any weights. Architecture B (Extreme Wall-Clock Speedup):  Physically eliminates stationary tokens before the encoder. With sparse