A Nonlocal, Baryon-Anchored Analysis of Galactic Acceleration Excess

This preprint presents a fully nested, galaxy-separated test of whether the positive galactic acceleration excess in SPARC rotation curves can be reconstructed from a nonlocal, scale-dependent, baryon-anchored model family. The analysis uses 175 SPARC galaxies and 3391 radial measurements. The nonlocal model reaches prediction accuracy statistically indistinguishable from a foldwise fitted radial acceleration relation (RAR). The dynamical kernel remains positive in all 4500 galaxy-wise bootstrap fits, and a leave-one-out kernel with its local self-contribution removed remains positive in all outer-fold models. A conditional out-of-fold galaxy-profile reassignment test yields p = 1/2001 in all three quality samples. An additional soft-saturation correction shows no consistent incremental predictive gain and does not determine a unique transition scale. A fixed nonlocal q = 0.5 gradient response is therefore specified only as a parsimonious candidate for future external testing. The results identify a reproducible cross-radial, baryon-anchored, scale-dependent predictive structure in the SPARC development sample. They do not establish a fundamental physical mechanism, exclude particle dark matter, or constitute external validation of the frozen candidate model.