Dataset for "Compact-Object Closure Profiles in MEG": White-Dwarf and Neutron-Star Basin Mechanics under a Frozen Hydrogen–Rydberg Density Coordinate

This dataset accompanies the paper “Compact-Object Closure Profiles in MEG: White-Dwarf and Neutron-Star Basin Mechanics under a Frozen Hydrogen–Rydberg Density Coordinate.” It contains the row-level inputs, authoritative mechanism-audit results, canonical same-sample confirmation products, equation-of-state-family transfer controls, response-morphology results, figure data, and reproducibility metadata used in the study. The analysis uses the fixed logarithmic energy-density coordinate n(u)=−ln⁡(u/uH)ln⁡CRy, with uH=4.672363908777×1012 J m−3 and CRy=9.668554. Neither quantity was fitted using compact-object data. Under this coordinate, the study examines an 87,025-row ordinary white-dwarf population near n=−14, a 73,975-row speed-of-sound equation-of-state/Tolman–Oppenheimer–Volkoff neutron-star ensemble near n=−23, a 71,516-row piecewise-polytropic negative-control ensemble, and a 145,491-row non-derived CS/TOV source table. The dataset supports the paper’s test of whether compact-object density addresses are isolated numerical alignments or extended closure basins whose internal positions are recoverable from observable structure. The primary all-row neutron-star audit compares raw mass with a derived mass–radius posture representation on the same 73,975 CS/TOV configurations, increasing held-out R2 from 0.0948 to 0.4627. A stricter canonical comparison uses 47,951 analysis-distinct configurations and identical five-fold assignments, increasing R2 from 0.1336 for the mass-only model to 0.50410.5041 for the mass-plus-posture model. The posture-augmented model reduces both absolute and squared prediction error for 80.9% of the canonical configurations. A separate 7,823-row response-available subset supports a full posture-and-response composite result of R2=0.5075. The release also includes held-out response-morphology results and permutation controls, together with a no-refit CS/TOV-to-PP/TOV transfer test. The CS-trained common-feature model succeeds on native CS/TOV rows but fails when transferred without refitting to the PP/TOV family, while a PP-native model gives a strong family-specific reconstruction. The deposited package includes the primary input tables, fold assignments, row-level out-of-fold predictions, exact metric recomputations, model summaries, transfer results, response-morphology results, figure-source tables, source-to-result maps, file inventories, manifests, and SHA-256 checksums. Coordinate-defining quantities are excluded from the strict predictor matrices used to reconstruct signed basin position. The scope of the release is mechanism-level compact-object basin analysis. It does not determine the dense-matter equation of state and does not perform selection-corrected compact-remnant population inference.