Photochemical [2+2] Cycloaddition Dataset with Traced Provenance and a Study-Grouped Benchmark

Curated dataset of 249 photochemical [2+2] cycloaddition records (220 from the review of Poplata et al., Chem. Rev. 2016, 116, 9748, and 29 from Hörmann et al., Angew. Chem. Int. Ed. 2020, 59, 9659), with reactant, catalyst and product SMILES, reaction conditions, yields and repaired stereoselectivity ratios. 248 records are traced to 272 primary-paper DOIs in 139 independent studies. The archive includes the provenance table, a log of all corrections of spreadsheet artifacts, the exact study-grouped cross-validation folds, the analysis code that reproduces every result in the accompanying paper, the complete descriptor-model outputs, and a curation template for extending the dataset.