nx_capaxes_lib.nx

nx_capaxes_lib.nx -- CAPABILITY AXIS PROFILE, pure core. No main, no I/O beyond emit helpers. WHY (operator 2026-07-30): "the capabilities census is gamed on coverage to 1000permil but with below toy quality, a case of imbalanced quality and quantity" + "capabilities graph with all the axes where there are between 1 and x and on each axis a measurement happens to see how imbalanced or how things skew in a direction, not just boolean type thinking". MEASURED DISEASE. nx_cap_census emits "torrent 377/1000 HAVE 20 / PARTIAL 6 / GAP 35 vs 0 SOTA-src" and "ECOSYSTEM (3 targets) 527/1000". (1) HAVE/PARTIAL/GAP is a PRESENCE ladder so a toy and a SOTA impl both score HAVE; (2) "vs 0 SOTA-src" = EMPTY comparator, self-referential, can only drift up; (3) 3 targets labelled ECOSYSTEM while the maturity rollup tracks 26 domains. Independently nx_sota_status reports PROVEN 0/40 and lists workermesh cov=1000 gaps=0 INSIDE CLAIM-ONLY -- perfect coverage, zero evidence. LAW L1 -- HEADLINE IS THE MINIMUM, NEVER THE MEAN. Kills coverage-gaming BY CONSTRUCTION. Averaging [coverage 1000, evidence 0] yields ~500 and reads like progress; the minimum yields 0 and reads like the truth. A capability is exactly as good as its WEAKEST axis, because that is the axis a user hits first. cax_mean exists ONLY so the gate can assert headline != mean on a gamed vector: a revert to averaging turns that tooth RED. It must never be the headline. LAW L2 -- GROUNDING ZERO REFUSES TO EMIT A NUMBER. No external reference => UNGROUNDED, never a score, because an ungrounded number is indistinguishable from a measured one once quoted. LAW L3 -- UNMEASURED IS -1, NEVER A SILENT 0, and must never become the minimum. SKEW IS THE PRODUCT, NOT THE SCORE. A balanced 600 profile and a [1000,200] profile can share a mean and are completely different objects: the second is breadth-gamed and starved names its rock. license_tier: ORIGINAL No hw writes (Rule 26).