Cation-screening dataset and analysis code for solvent-induced nucleation and crystallization (SINC) of ionic hybrid metal halide glasses

This dataset contains the complete A-site cation screening library, preprocessing workflow, similarity rankings and analysis code supporting the study of solvent-induced nucleation and crystallization (SINC) in ionic hybrid metal halide glasses. Important clarification. The 49,276 entries constitute a virtual screening library evaluated with a pretrained MoLFormer chemical language model. They are not training data: the model was used as a fixed feature extractor and was neither retrained nor fine-tuned. Screening funnel. Raw database records (ChemSpider, PubChem): 49,276 → records yielding exactly one cation fragment with net charge +1: 32,434 → unique canonical cations after deduplication: 19,558 → similarity-ranked unique cations: 19,558 → molecular-weight gate, MW(PPh4+) ±10%: 3,326 → after exclusion of Se-containing, P–P and PH+-like motifs: 2,954. Files.• screening_master.csv — record-level audit of the full library (curation, deduplication, molecular weight, motif exclusion, similarity, final-pool flags)• funnel_counts.csv — stage-by-stage counts reconciling the screening schematic• selected_candidates.csv — the anchor (PPh4+) and the three experimentally evaluated cations (BzTP+, 4-MBz+, CHTP+) with cosine/Euclidean distances and ranks• descriptor_summary.csv — conventional molecular descriptors for the four highlighted cations• embedding_metadata.json — model, anchor, distance metrics and filter definitions• integrity_report.json — verification of stage counts and candidate retrieval• code/reproduce_tables.py, code/requirements.txt — script to regenerate the tables from the deposited data• README.md — full documentation Similarity values are record-level MoLFormer distances inherited by each canonical representative and are reported as deposited; they are fully reproducible from the included data and code.