Assembly, 1993. Future Crew is building Second Reality—full 3D environments, texture-mapped tunnels, particle systems, soundtrack. They have 65,536 bytes. That's it. Not 65 megabytes. Sixty-four kilobytes including the music. A single uncompressed 640×480 image would blow the entire budget. So they cut assets to zero and grow everything procedurally. Textures generated from 200-byte algorithms. Meshes built from mathematical formulas. The music synthesized in real-time, no samples. Every pre-stored byte surrendered becomes headroom for the generative engine. The lake evaporates; the mountain rises. You're hitting limits. Storage, bandwidth, attention budget—something maxes out. The instinct is to optimize what exists. Compress harder, cache smarter. That's fine. But 損 asks: what if you removed the asset entirely and generated it on-demand? Trade stored weight for procedural altitude. The failure mode isn't aggressive reduction. It's random reduction. Future Crew didn't delete the renderer to fit more textures. They deleted the textures because the renderer could synthesize them. Decrease the lower (pre-baked assets) to feed the upper (generative capability). Not loss—transfer. Constraint formalizes focus. Every byte you surrender must buy you something higher. If it doesn't, you're not practicing 損; you're just starving the system.