I investigated whether routing a transformer's forward activations through a lossy Dual E8 (E16) lattice bottleneck and injecting them back into the residual stream is viable, and where the boundary of generative stability lies. **The core finding:…