nx_swcompare_evidence_gate.nx

nx_swcompare_evidence_gate.nx -- THE REFEREE'S GROUNDING RULER, AND THE ONE QUESTION IT MUST NOT GET WRONG: CAN A COMMENT GROUND A PUBLISHED CLAIM? WHY IT EXISTS (measured 2026-09-01). nx_swcompare_evidence decided grounding with ev_find_line -- a WORD-BOUNDARY SUBSTRING over the raw organ source, comments and string literals and call sites all included. gt_drape_mesh returns matches=1 over 23,412 files (corpus_complete=1) and that ONE hit is a COMMENT in nx_garment_twin.nx line 11. The referee reported phototwin's GARMENT row [LANDED] and fired LANDED-UNFLIPPED -- the flip worklist is where a seat looks for the cheapest closure on a board, so the instrument was actively instructing someone to publish a capability that does not exist. A SCANNER THAT DOES NOT SKIP COMMENTS MEASURES THE DOCUMENTATION, NOT THE CODE. THE SUBJECT OF THIS GATE IS THE RULER, and the fix was an ADOPTION, not an invention: the referee now composes nx_symdecl_lib's sd_present_buf, the estate's ONE SYMBOL RULER, which the GENERATOR (nx_swcompare_matrix), the REGEN, the RANKER and the WATCH-STATE lib were ALL already calling. The referee was the fifth wheel. So every tooth here is IN-PROCESS against that lib plus the DEMOTED ruler kept beside it as a named control -- a fixture on which the two rulers AGREE proves nothing about either, so each discrimination tooth asserts BOTH answers. AND THE CENSUS IS THE POINT, NOT AN EXTRA. A ruler change rewrites every published cell, so it may not ship on a sampled number: `census` walks EVERY .matrix row in BOTH compare trees and reports how many currently-grounded axes are carried ONLY by a non-declaration occurrence. That count is the number of published capabilities that do not exist. It is REPORTED, never voted on -- an uncalibrated classifier must report numbers and never verdicts, and a gate that went RED on other boards' data would be permanently red and therefore ignored. license_tier: ORIGINAL No hw writes (Rule 26).