nx_procgen_water_test.nx

nx_procgen_water_test.nx -- KATs for P4 causal water (rivers + rills). Exit 0 = all pass; first failing KAT's number is the exit code. KAT 1 trace physics: on a synthetic cone, a trace from the apex is strictly height-decreasing and reaches the map edge. KAT 2 trace stops at a local minimum (synthetic bowl) -- water collects. KAT 3 carve lowers exactly the channel: centre cells down by depth, a far cell untouched; mask marks the path. KAT 4 determinism: v2 twice -> byte-identical heightmaps AND masks. KAT 5 v2 differs from v1 (water actually happened) and the mask is non-empty on every preset x 2 seeds (width: rivers always form). KAT 6 water mask overrides feature kinds: with-mask WATER count > no-mask WATER count; conservation holds. KAT 7 LOCAL_VARIATION axis moves UP on v2 vs v1 (worst-of-3 mountain seeds) -- the ADD_DETAIL claim, instrument-proven. KAT 8 no grade regression: v2 grade >= v1 grade on the same (seed, preset), worst-of-3 mountain (water never makes the verdict worse). KAT 9 caprock resists: a top-band cell on a carve path is untouched (and unmasked) while a lower cell on the same path carves -- the STRUCTURAL-ESCALATION rung's law, geometry-exact. license_tier: ORIGINAL