_gpu_dxg_gate.nx

_gpu_dxg_gate.nx -- GATE for the SOVEREIGN-GPU foundation rung (R0): proves Nishi reaches the REAL RTX 5080 through /dev/dxg (the WSL dxgkrnl last-mile) with RAW syscalls only -- the emitted elf links NOTHING (no libvulkan / libcuda / libdxcore / libd3d12 / libc). The verdict comes from a REAL ioctl to the REAL device returning REAL adapter data (count + non-zero LUID), NOT from $?. AUTHOR=ORGAN / NO-FALSE-GREEN -- the gate runs the real nx_gpu_dxg_probe mechanism three ways against the SAME binary and asserts ALL must hold: (A) REAL CALL on /dev/dxg: LX_DXENUMADAPTERS2 ret==0 AND num_adapters>=1 AND a non-zero adapter[0] LUID/handle came back from the 5080. (B) TAMPER (magic-0x99 ioctl on the SAME real /dev/dxg fd): the device rejects it with a real errno (< 0, observed -25 ENOTTY) that DIFFERS from the real ret==0. A 0 here = the fd is a no-op => RED. This proves ret==0 is a real device ACCEPT, not "any ioctl returns 0". (C) NON-DXG FD control (/dev/null): the EXACT SAME LX_DXENUMADAPTERS2 ioctl returns a real errno (< 0, ENOTTY) -- NOT 0 and NOT adapter data. This proves the green depends on the REAL dxgkrnl device behind /dev/dxg, not on merely having an open fd. GREEN iff (A) AND (B) AND (C). Any failure -> RED with the real returned values, NEVER a fake green. SCOPE (NO-WAVE): GREEN here means ONLY "Nishi reaches the real RTX 5080 sovereignly via the /dev/dxg paravirt last-mile". It claims NO throughput / speedup / exceed (that is R7, a real cuBLAS head-to-head many rungs up). This is the FOUNDATION rung the GPU ladder rests on. SELF-CONTAINED: imports nx_gpu_dxg_probe and calls dxg_enumadapters() in-process -- it does NOT fork a prebuilt _offc/*.elf, so the /tmp->_offc install landmine does not apply to this gate. Evidence -> knowledge/status/gpu_dxg.log. license_tier: ORIGINAL