_gpu_dxg_r1_recon.nx

_gpu_dxg_r1_recon.nx -- SCOPE/RECON probe for SOVEREIGN-GPU rung R1: adapter DESCRIPTION + positive RTX 5080 ID via LX_DXQUERYADAPTERINFO(KMTQAITYPE_ADAPTERREGISTRYINFO). RAW syscalls ONLY to /dev/dxg (no libvulkan/libcuda/libdxcore/libd3d12/libc). Reuses the R0 enum mechanism. What this recon PINS LIVE (the kernel rejects a wrong code with -ENOTTY=-25 and a wrong size/type with -EINVAL=-22, so we probe before claiming): 1. The exact LX_DXQUERYADAPTERINFO ioctl code. Encoding (same _IOWR(0x47,...) family as R0): code = (3<<30) | (24<<16) | (0x47<<8) | nr for a 24-byte struct. We sweep nr over a small range and report which nr the kernel ACCEPTS (ret==0) vs -ENOTTY. 2. struct d3dkmt_queryadapterinfo (24B): adapter_handle u32@0 / type u32@4 / private_data u64@8 / private_data_size u32@16 / reserved u32@20. 3. KMTQAITYPE_ADAPTERREGISTRYINFO type value (expected 8) -- we sweep a few type values and report which returns a real UTF-16 string. 4. Whether QueryAdapterInfo works on the ENUM handle directly, or needs LX_DXOPENADAPTERFROMLUID first. We try BOTH paths and report which path produced the real string. 5. The ACTUAL UTF-16LE adapter_string each adapter returned (decoded to ASCII for printing) so we can positively distinguish the RTX 5080 from the WSL/Microsoft software render adapter. NO-WAVE: reach + identify only. No throughput/exceed. AUTHOR=ORGAN / NO-FALSE-GREEN: every printed string is the REAL bytes the kernel wrote into private_data, never a literal. A wrong handle/type goes to a real errno, reported as-is. license_tier: ORIGINAL