Benchmark Dataset: AI-Assisted versus Expert Curation of a Battery Materials Database (BAI–AWB Benchmark)

This dataset supports the article "Scientific Database Construction in the Age of AI: Benchmarking AI-Assisted and Expert Curation of a Battery Materials Database" (submitted). It contains the complete data underlying the record-level benchmark between BAI, a battery-materials database constructed by an AI-assisted workflow (Claude Opus 4.8, model identifier claude-opus-4-8), and AtomWork-Battery (AWB), an expert-curated database developed at the National Institute for Materials Science (NIMS). All results reported in the article can be reproduced from the tables in this dataset. Contents: (1) The complete BAI database (bai_database.xlsx) constructed from 81 publications on solid-state electrolytes and cathode materials, including materials, practical phases, properties, and bibliographic metadata with DOIs for all source publications. (2) Material pair-matching tables between BAI and AWB records from the 79 shared publications, covering the three matching levels described in the article: composition-matched pairs within each publication, pairs with identical chemical systems, and the 265 pairs additionally matched on processing temperature (within 200 K) that form the basis of all reported numbers. (3) Comparison tables for material-type assignment, main-phase composition, space group, lattice parameters, ionic conductivity, and discharge capacity, together with per-pair detail tables, failure-case lists, and root-cause analyses. All data correspond to a frozen snapshot of both databases taken on 16 July 2026. Complete AWB records beyond the compared fields are not included, as AWB is a curated database product of NIMS. The detailed curation rules and extraction instructions used by the BAI workflow contain proprietary NIMS technology and are likewise not included; their scope and structure are described in the article. See README.md in the dataset for file-level documentation. This work was supported by JST COI-NEXT 'Center for Advanced Battery Collaboration', Grant Number JPMJPF2016.