nx_compare_sections.nx

nx_compare_sections.nx -- SECTION CENSUS of the published /compare surface: which <h2> sections every domain page carries, the union across the surface, and -- against a DECLARED required shape -- which sections each page is MISSING, named per domain. WHY THIS EXISTS (2026-09-02, operator: "all the /compare on the same emitted output, not random; if a previous compare had more capabilities bring it into the emitter"). Two generators emit the surface (nx_swcompare_matrix, nx_swcompare_sota) and their pages had drifted to two SHAPES: the sota-class pages lacked Evidence profile, Do this next, Critical path and the Capability matrix that every matrix-class page carries. The first attempt to measure that was `nx_shelltool grep "<h2"` over the docroot, which hit its 1 MiB budget on the FIRST page: emitted HTML is minified, so one section is one line of 100 KB and a line-oriented grep measures nothing. A TAG-AWARE WALKER IS THE INSTRUMENT, and it belongs in the estate, not in a shell one-liner (NishiLang for everything; shell is the wire). THE REQUIRED SHAPE IS DATA, NOT A LIST IN THIS FILE: knowledge/compare/sections.required rows section|<title prefix>|<data ext or -> A section is REQUIRED of a domain when its data class exists for that domain (<kdir>/<dom>.<ext>) or when the ext is "-" (required of every page). A page that has the data but not the section is the emitter-shape defect this organ names; a page without the data is not blamed for the section. Title matching is by PREFIX after normalisation (tags stripped, cut at the first " &mdash;"), so the generators' long explanatory headings ("Do this next &mdash; computed by the ranker...") match their short declared prefix and a reworded tail cannot silently un-match a section. nx_compare_sections <docroot> <kdir> <required.conf> docroot the live compare docroot holding <dom>/index.html (sites/nishifamily/compare from nishihost) kdir the knowledge compare tree the regen READS (buildroot/knowledge/compare) required the declared shape Prints one DOM line per page (sections carried, missing named), one MISSING row per absent required section, the union table (title, domains carrying it), then the partition WITH its sums, then the verdict LAST (gv_last_line anchors by position). Exit: 0 GREEN (no page misses a required section) . 1 RED (at least one MISSING row) . 2 usage . 3 UNOBSERVABLE (docroot or conf unreadable, or zero pages read -- an empty population is never a pass). license_tier: ORIGINAL Read-only on every path it touches. No hw writes (Rule 26).