Numerical Integration Sensitivity and Decision Screening for Fixed-Geometry Molecular Energy Comparisons

Dataset overview This dataset provides a reproducible numerical investigation of molecular orientation sensitivity and integration-grid effects in density functional theory (DFT). It contains 3,088 ORCA 6.1.1 fixed-geometry single-point calculations, including original input files, privacy-redacted computational outputs, molecular coordinates, machine-readable energy tables, convergence assessments, decision-policy benchmarks, and Python-generated scientific figures. The main challenge set comprises 24 fixed-geometry molecular pairs, including 12 previously examined molecular identities with newly selected geometries and 12 chemically distinct holdout molecules. The challenge includes 2,016 calculations sampling molecular orientations, numerical integration grids, and M06-L/PBE calculations where specified. Candidate geometries were selected using MMFF94 without consulting DFT energy outcomes. The dataset supports assessment of numerical integration errors, orientation-dependent energy variations, dense-grid reference stability, and the computational cost of alternative decision-screening protocols. An adaptive two-orientation screening procedure identifies an overconfident single-orientation decision in this test set. However, a simpler exploratory energy-margin baseline achieves comparable classifications at lower computational cost; superiority of the adaptive method is therefore not established. All reported comparisons concern fixed-geometry electronic energies, not equilibrium conformer free energies or experimental thermodynamic observables. Dense-grid calculations serve as operational numerical references within the selected approximations. Reproducibility and provenance The archive includes independently executable Python analysis scripts, CSV source tables, six reproducible scientific figures, SHA-256 integrity manifests, and documented computational settings. Original inputs are preserved, while archived ORCA outputs have been sanitized to remove machine-specific information. The historical RDKit version used to generate the original geometries could not be conclusively established; canonical molecular coordinates are preserved. Licensing Research data, molecular structures, figures, and associated documentation are distributed under Creative Commons Attribution 4.0 International (CC BY 4.0). Author-released analysis and validation source code is distributed under the MIT License. The license scope is documented in the archive. Third-party software is not redistributed under these licenses. Creator: Ahmet Güneş Affiliation: Department of Basic Sciences, Turkish Naval Academy, National Defence University, Istanbul, Türkiye. ORCID: 0000-0003-0966-4025 Correspondence: [e-mail]