nx_worldpipe_render_gate.nx

nx_worldpipe_render_gate.nx -- GATE for the rung "erosion reaches the PUBLISHED picture". THE SUBJECT IS NOT "DOES EROSION WORK". nx_worldpipe_erode_gate already proves that, 21/21, over the full 102,400-cell population. The subject HERE is the single behavioural claim nx_worldpipe_render makes and that a green erosion gate cannot make for it: that the bake changes the heights AT THE EXACT POINTS THE RENDERER TURNS INTO PIXELS, and that a consumer which does not bake is untouched. A capability can be proven and still never reach a surface; this gate measures the reaching. EVERY ASSERTION IS BOUND TO A DENOMINATOR (gv_subjects), and the denominators are the two real populations: the whole hydrology grid, and the whole hero sample lattice the render loop walks. No sampling. ANTI-VACUITY. "The picture changed" is passed by a uniform offset, by added noise, and by any bug that perturbs every height equally -- so a tooth that only counts changed points is not a tooth. This gate additionally requires the changes to be NON-UNIFORM (at least two distinct delta values among the changed points), which a constant offset cannot satisfy. NEGATIVE CONTROLS, NAMED SO A CENSUS CAN FIND THEM: neg-control-wp_erode-INERT-... proves the opt-in is real: before any bake every one of the 102,400 cells reads exactly zero, so an importer that never bakes is byte-identical. This is the "nobody else moves" claim. neg-control-zero-tick-budget-... proves the instrument does not manufacture erosion: asked for a zero budget it must return every sampled height to its pre-bake value and report zero material moved. license_tier: ORIGINAL expect_exit: 0 No hw writes (Rule 26).