Gravity - Rotation Curves Under a Global-Only Policy

We test a strictly global, finite–refresh correction to the baryonic response, encoded by a deterministically computed weight w(r) built from baryonic maps and catalogued photometric geometry only (no kinematic inputs), under a strict global–only policy (single stellar M/L, shared error model, predeclared inner–beam mask). On the SPARC Q=1 subset (NQ1 = 127), after uniform masks the effective samples are NILG = 126 and NMOND = 125. With identical data vectors and loss, a like–for–like MOND (simple ν) baseline attains median χ2/N= 2.47 and mean 4.65. The ILG benchmark yields median 2.75 and mean 4.23. The 1D proxy is not competitive (median 3.782, mean 10.602). Series note. This manuscript is part of a coordinated companion pair. Paper I defines the fixed, global phenomenology w(r) and its scope; Paper II (this work) tests that w(r) on SPARC under identical masks/error model shared with all baselines. Both are co-submitted with shared artifacts and a single cover letter.