nx_uefi_ovmf_oracle.nx

nx_uefi_ovmf_oracle.nx -- BENCH-ONLY ORACLE (operator law 2026-07-20). THE THIRD REFERENCE FOR THE UEFI LANE. nx_boot_uefi AUTHORS the EFI image and nx_emu_uefi EXECUTES it -- but both are OURS, so a GREEN from that pair proves the pair is self-consistent, never that real firmware agrees. ("A TWO-WAY COMPARISON ATTRIBUTES THE INCUMBENT'S ERROR TO YOU -- BUILD THE THIRD REFERENCE ONCE AND REUSE IT AT EVERY SCALE.") The third reference is EDK2/OVMF: an independent UEFI implementation nobody here wrote. This organ does NOT run qemu (qemu is a foreign binary and lives only on the laptop; the NAS has neither qemu nor OVMF -- measured 2026-08-08). It is the RULER: it takes the framebuffer real firmware actually produced (a qemu `screendump` P6 PPM) and the framebuffer the sovereign emulator modelled (`nx_emu_uefi --shot` 24-bit BMP) and compares them PIXEL FOR PIXEL, plus re-derives nx_emu_uefi's own fb_measure statistics over the FOREIGN buffer so both engines are judged by ONE ruler on ONE scale. Orientation/channel order are NOT assumed: PPM is top-down RGB, BMP is bottom-up BGR, and the transform is applied explicitly. Geometry comes FROM each file's own header, never from a constant -- a comparator that trusts the writer's constants is one implementation, not two. Usage: nx_uefi_ovmf_oracle <ovmf.ppm> <sov.bmp> nx_uefi_ovmf_oracle selftest (teeth, incl. a planted-difference negative control) Exit: 0 GREEN (dims agree AND zero differing pixels) | 1 RED | 3 UNPROVEN (could not read) -- the third state exists so "I could not look" never reports as "it is broken". Log -> knowledge/status/nishi_os.log, canonical verdict= on the LAST field (positional anchor). Sovereign: syscalls only, no gcc/.sh. license_tier: ORIGINAL