Ordinal Solvation Framework (OSF) dataset

Sheet-by-sheet CSV export of the frozen experimental workbook underlying the Ordinal Solvation Framework, a six-state ordinal description of polymer solvation (undissolved, softening, swelling, dispersion, colloids, dissolved) with a continuous solvation coordinate s defined as the expected rank over the ladder. The deposit contains over 36,000 screening records, of which 35,285 labelled observations trained the released models. Coverage spans 94 polymers, 60 single solvents and 805 binary solvent pairs between 4 and 75 °C. Each record carries the observed solvation state on the six-state ladder together with SMILES identifiers and a curated physicochemical descriptor block for the polymer and the solvent. Four data sheets are provided. The homopolymer and copolymer sheets hold single-solvent screens, the binarysolv sheet holds the binary solvent mixtures used to train and evaluate the model, and the unseen sheet holds external validation screens excluded from all training. Two audit sheets document every relabel and descriptor repair applied before freezing, and KNOWN_ISSUES.md catalogues corrections identified after freezing, which will be applied in a versioned v2 release together with retrained models. This is the exact dataset version on which the released model artifacts were trained (github.com/zjlucam/ordinal-solvation-framework, tag v0.2.0). The source workbook SHA-256 is stated in the README and per-file checksums are included. Binary solvent screening beyond the training set is reserved for follow-up work and is not part of this deposit. Files are under restricted access during peer review of the associated manuscript and will be made openly available on publication. Please cite the associated manuscript (reference to be added on publication).