nx_view_transform_gate.nx

nx_view_transform_gate.nx -- IS THE VIEW TRANSFORM DERIVED, OR JUST TYPED? license_tier: ORIGINAL No hw writes (Rule 26). The claim under test is not "the curve looks nice". It is four mechanical properties, each of which the shipped page FAILED on 2026-08-28 and each of which nx_view_transform is supposed to make true BY CONSTRUCTION: 1. the exponent is COMPUTED from declared measurements, not hand-typed (the page carried 1.45, whose own comment claimed "half-way" while it moved the median 45.7%); 2. the anchor is an exact FIXED POINT, so the white point cannot drift; 3. the application PRESERVES HSV saturation exactly (the page raised S mean .313 -> .411, out of the .27-.32 band the measurement said saturation already sat inside); 4. the LUT size is SEARCHED against a declared tolerance, not chosen. THE NEG-CONTROL IS THE LOAD-BEARING TOOTH. T5 applies the OLD per-channel form to the same triple and requires that it RAISES saturation. Without it, T4 could pass on a transform that does nothing at all: a no-op preserves chroma perfectly. T5 is what proves this gate can tell the fix from the defect rather than merely tell quiet from loud.