nx_nofloat_exceed_gate.nx

nx_nofloat_exceed_gate.nx -- the UNIQUE Nishi exceed axis: NO-FLOAT determinism. Operator's differentiator for image/video/LLM/coding models: build on INTEGER/fixed-point, not float. WHY it is S-class exceed (not raw speed -- DETERMINISM + EXACTNESS): float add is NON-ASSOCIATIVE, so the SAME dot product gives DIFFERENT bits when summed in a different order (or on different hardware / thread counts). Integer accumulation is associative + exact -> BIT-IDENTICAL across every ordering, every machine, forever. For a CODING model this is decisive: same prompt -> same code, reproducibly. sdcpp/GGML/llama.cpp accumulate in float = non-deterministic by construction; the no-float stack does not. This proves the axis on one dot: a=[2^24,1,1,1,1] . b=[1,1,1,1,1] = 16777220 exactly. No hw writes (Rule 26). expect_exit: 0 license_tier: ORIGINAL