nx_capaxes_derive.nx

nx_capaxes_derive.nx -- DERIVES capability axis vectors from a LIVE source and ranks the rocks. WHY (operator 2026-07-30): the census is "gamed on coverage to 1000permil but with below toy quality" and partial capabilities are "treated as completed capabilities not as things that need a full living iteration loop to determine biggest rocks". nx_capaxes_lib made the SHAPE measurable; this closes the loop by feeding it real numbers and emitting the rock order. SOURCE (composition, not reimplementation -- the nx_debt_view pattern): fork nx_sota_status and read its per-domain rows. Every domain it lists carries local .matrix evidence, so GROUNDING is real for these (unlike nx_cap_census, which reports "vs 0 SOTA-src" for every row = an EMPTY comparator, debt 1785465147). AXES SET HERE, and what is HONESTLY NOT YET WIRED: coverage <- the row's own cov= figure MEASURED evidence <- 0, because every row nx_sota_status prints with MEASURED a cov= figure sits in its CLAIM-ONLY block (PROVEN is 0/40, so no row scores above this) grounding <- CAX_FULL: the domain HAS a .matrix comparator MEASURED quality <- UNMEASURED (-1), needs the ecomat maturity grade scale <- UNMEASURED (-1), needs the gate envelope N adoption <- UNMEASURED (-1), needs the nx_wirecensus join -1 is NOT a silent 0 and the lib proves it can never become the minimum (nx_capaxes_gate T3). Three axes measured is a FLOOR, declared in the envelope -- never presented as the whole picture. THE POINT: with evidence=0 the headline is 0 no matter how high coverage climbs, so LEVERAGE is what orders the work -- and leverage IS the coverage claim. A domain asserting cov=1000 with zero executable evidence is the BIGGEST rock, not the most finished one. That inverts the gamed ranking exactly. nx_capaxes_derive [topN] exit 0 ok | 4 capture-fail license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0