nx_microscan_gate.nx

nx_microscan_gate.nx -- THE INVERSE MICRO-ANALYSER, GATED BY ROUND TRIP. Subject: nx_microscan_lib. WHY A ROUND TRIP IS THE RIGHT REFEREE HERE, AND WHY IT NEEDS NO EXTERNAL GROUND TRUTH. The estate already owns the FORWARD law: nx_relief_lib bakes a jittered-grid pore layer whose pitch, orifice diameter and coverage are declared constants with a provenance ledger. So the honest test of an INVERSE analyser is to hand it a field the shipped generator produced and ask it to recover the number the generator was told -- WITHOUT the analyser importing any of it. nx_microscan_lib imports nx_syscalls and nx_vecmath and nothing else; this gate imports the generator, synthesises the field, and checks the recovered pitch against RLF_PORE_PITCH_UM. A wrong analyser cannot pass by knowing the answer, because it has no way to see the answer. AND IT IS CHECKED AT TWO LATTICE SCALES ON PURPOSE. A single scale can be satisfied by a constant. T5 asserts the recovered period TRACKS the generator -- double the cell, double the answer -- which is the property a hardcoded 800 provably cannot have. THE TOOTH THE OPERATOR ACTUALLY ASKED FOR IS T8/T9: "does the hair come out of the pore properly". That is a REGISTRATION question between two layers, and it is decidable today. T8 places follicles through the generator's own centre function and must read BOUND. T9 places follicles by AREA SAMPLING -- which is what nx_nxa_groom actually does, it samples scalp triangles by area and never consults a pore field -- and must read DECOUPLED. Both directions, one bar, and the null is the field's OWN measured coverage rather than an assumed rate. THE TWO POINT SETS ARE DELIBERATELY DIFFERENT SIZES and that is the finding of the first run, not an oversight. BOUND is a presence claim and carries on 256 points; DECOUPLED is an ABSENCE claim and needs enough EXPECTED chance hits to have detected binding had it existed, which 256 points at a 12-permil chance rate does not provide. ms_reg_verdict abstains there, and the last neg-control proves it abstains by handing it exactly that under-powered subset. DECLARED RESOLUTION BAR, so no reading of this gate can be inflated into a clinical claim. The fields here are synthetic and sampled at a texel derived from the generator's declared pitch. The baked per-region asset carries a 50 um texel. Dermoscopic assessment of a pigmented lesion works at roughly 10-20 um per pixel and reads COLOUR and PIGMENT NETWORK, not height. This organ measures a height/thickness field for periodicity, layer registration and departure-from-self. It emits candidates with scores. IT IS NOT A DIAGNOSIS OF ANYTHING and the melanoma bar is a statement of the resolution we do not yet carry, not a capability claimed. The dermoscopy figure is SOURCE-UNPINNED -- no reference was mirrored for this lane -- and that is stated in the verdict note too. license_tier: ORIGINAL expect_exit: 0 No hw writes (Rule 26).