Matter-Conditioned Metric Response with Coherence-Weighted Irreversibility: A Galaxy-Held-Out CREST Test on SPARC

This release archives the CREST coherence-weighted metric-response SPARC benchmark package. The accompanying paper, “Matter-Conditioned Metric Response with Coherence-Weighted Irreversibility: A Galaxy-Held-Out CREST Test on SPARC,” rewrites CREST (Covariant Relativistic Extended Scalar Theory) as a scalar-tensor metric-response framework with an explicit coherence-weighted irreversibility law. The response sector is defined through a residual curvature tensor and constrained response law, separating integrated irreversible load F from instantaneous irreversible response flux Φirr. The empirical scope is deliberately narrow: galaxy-held-out prediction on public SPARC rotation-curve mass models. The benchmark reports 3384 pointwise measurements across 175 galaxies after finite-value checks. The declared CREST v5b response projector improves median galaxy RMSE by 7.33% over a fitted RAR/MOND-like backbone and by 3.54% over a radius-plus-baryonic-acceleration nuisance control. It also outperforms no-retuning NFW/Burkert halo-prior predictors, while flexible per-galaxy halo fits remain stronger raw curve-fitting ceilings. This release includes the paper, benchmark scripts, generated CSV summaries, validation reports, and source/results archive for the v5b framework-shrinkage projector, v7 halo-prior comparison, v9 post-fit null-collapse test, and v10 lightweight full-refit null confirmation. Claim boundary: this is a no-retuning prediction-efficiency result on SPARC residual structure. It does not claim that CREST disproves dark matter, closes lensing/cosmology, or beats fully tuned per-galaxy halo fits.