Provenance-Documented Reference-Gated Diagnostics of Pretrained Machine-Learned Potentials for Deep Eutectic Solvent Dimers

This record contains the source data, portable source geometries, fragment definitions, provenance records, bounded reanalysis code and outputs, selection-history records, provenance-gate audits, fault-injection tests, and replay-readiness documentation supporting the manuscript “Provenance- Documented Reference-Gated Diagnostics of Pretrained Machine-Learned Potentials for Deep-Eutectic- Solvent Dimers.” The archive documents a bounded existing-data analysis of three pretrained machine- learned interatomic potentials—CHGNet, MACE-OFF23, and MACE-MP-0—on deep-eutectic-solvent-relevant dimers. The release includes machine-readable source tables for Supporting Information Tables S13–S22, portable dimer geometries and fragment maps, Level-0/Level-1/Level-2 reanalysis scripts and outputs, provenance manifests, executable provenance-gate audits, fault-injection tests, selection history, and a replay-readiness assessment. No new DFT or electronic-structure calculations, MLIP inference, retraining, fine-tuning, calibration, or target-based model selection was performed for this revision. The archive supports verification of the stored Level-1 and full-case artifacts, provenance-gate behavior, source-geometry and fragment records, and the reported reanalysis outputs. It does not claim a clean one-command replay of every original production calculation. Six breadth-production scripts, exact checkpoint-to-array cryptographic bindings, a complete production environment lock, and raw reference-input bundles are unavailable; these limitations and their reason codes are documented in the replay-readiness files. The manuscript, Supporting Information PDF, Figure 1, TOC graphic, and LaTeX submission source are submitted separately through the journal system and are intentionally not duplicated in this data/ code record. This is version 1.0 of the bounded data and reanalysis release.