nx_ng_world.nx

nx_ng_world.nx -- CAP-WORLD-MODEL, generation 14: THE ENDPOINT of the neural-graphics ladder. The leap past every prior rung (which rendered STATIC scenes): a learned DYNAMICS model -- state x action -> next state -- that can be ROLLED OUT (applied repeatedly) and RENDERED frame-by-frame = an interactive neural game engine (Ha&Schmidhuber world model / GameNGen / Genie / Sora-class). src ng_world_sora, ng_world_t2v, ng_world_worldmodel. Reuses the renderer (software raster into nx_image, displayed via the nx_scene_demo canvas bridge -- the only browser touch). Learns the transition rule from observed (state,action,next) transitions, then SIMULATES a NOVEL action sequence and renders the rollout as a spacetime filmstrip (rows = time, cols = space). Q16, no float. Honest gated proof of the DEFINING (interactive) property: T1 LEARN-DYNAMICS -- recover the transition rule from data (|V-V*| small); T2 INTERACTIVE ROLLOUT -- a NOVEL action sequence (not the training one) is simulated correctly (predicted trajectory == true world, err~0) = actions drive the next frame, generalizes; T3 NEG-CONTROL -- a model that IGNORES the action (V=0, static) CANNOT predict (the action is load-bearing = what makes it a *world* model, not a static scene); T4 RENDER+EMIT -- the rollout renders to a viewable filmstrip. HONEST: 1-DOF state (object x-position), linear dynamics; full world-model = high-dim latent state + learned non-linear dynamics + frame-conditioned generation = the follow-on. Sovereign: nx_image + nx_base64 + syscalls. The emitted artifact is PUBLISHED via nx_ng_publish (NOT local show-and-tell). license_tier: ORIGINAL expect_exit: 0