nx_openpub_census.nx

nx_openpub_census.nx -- the HONEST MEASURED starting line for a NISHI answer to apertvs.ai (Apertus, the Swiss AI Initiative's "Fully Open Foundation Model for Sovereign AI"). Drives the GENERAL, reusable nx_capability_census engine: each axis = a DATA ROW (competitor-evidence TERM + our implementing-organ path + our exceed-gate path); the engine grades NEVER self-scored -- them(src) = how many of the sovereign-banked openpub_*.raw sources attest the term (re_has; >=1 noted) us = EXCEEDS (our exceed-gate opens on disk) > PRESENT (our organ opens) > ABSENT [disk facts] verdict = GAP (they have, we ABSENT) | PARITY (both) | AHEAD (we EXCEEDS, or PRESENT & they unattested) The thesis: apertvs.ai wins on OPENNESS (you can SEE weights/data/code). Nishi EXCEEDS on the axis openness can't touch -- SOVEREIGNTY + DETERMINISM (you can REPRODUCE it BYTE-EXACT, from silicon up, with NO third-party runtime). So we expect PARITY across the publish-everything bar (we can publish all they do) and AHEAD across the determinism/sovereignty moat (several MEASURED via a real exceed-gate on disk). GAPS are reported FIRST-CLASS, not hidden: language SCALE (Apertus=1000+ langs) and a training-data PII/memorization-prevention pipeline are genuine ABSENTs today (organ-shaped, named accurately -> honest GAP). GREEN = the MEASUREMENT is honest+grounded (liar-kill armed; graded against a real banked corpus), NOT "we win" -- the scorecard speaks for itself. Run nx_openpub_research_fetch first to bank the corpus. Sovereign nx_cc->nxasm (no gcc, no 3rd-party). expect_exit: 0 license_tier: ORIGINAL