nx_colorsci_gate.nx

nx_colorsci_gate.nx -- THE GATE FOR THE COLOUR-SCIENCE RULER (nx_colorsci_lib + nx_fixq30_lib), 2026-08-24. SUBJECT: the libs in-process (they have no main; the ruler IS the code under test). Teeth, in order: T1 the DERIVED constants against their published digits -- pi (Machin) and e to 1e-7, and the identities sqrt(2)^2 = 2, atan2(0,-1) = pi, sin(pi/2) = 1, cos(pi/3) = 1/2, cbrt(27) = 3, ln(exp(3)) = 3. A CORDIC whose table was mistyped cannot pass these; a series that stopped early cannot either. T2 CIEDE2000 against Sharma-Wu-Dalal 2005 Table I: ALL 34 pairs (gv_subjects binds the denominator), both directions (the formula is symmetric by construction, and pairs 7/8 exist to prove it), and the table's h_bar_prime column (the arctangent branch handling pairs 9-16 exist to trap). T3 NEG-CONTROL: pair 1 against a WRONG expectation (+0.5 dE00) must FAIL the tolerance -- a tolerance so wide that anything passes would read green on every KAT row and prove nothing. T4 ICC: the mirrored sRGB2014.icc (color.org) parses; rXYZ+gXYZ+bXYZ sum to wtpt within s15Fixed16 resolution (an ICC structural invariant, not a typed expectation); white reads L=100 a=b=0; black L=0; EVERY one of the 256 greys reads a=b=0 (the whole population, not a sample); rgb->lab->rgb round-trips within one code over all greys and the eight cube corners. T5 the gamut descriptor: derived from the profile at the conf's sampling; every sample it was built from reads INSIDE; a chroma no display reaches reads OUTSIDE; bite-proven both ways; chroma clipping brings the outside probe inside at the same L and hue. Thresholds: the KAT tolerance is the half-ulp of a four-decimal table plus Q30 headroom; everything else is a structural identity. license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0