Scopus and Web of Science PRISMA / trAIce Review-Process Dataset for Construction-Site Safety Sensing, Perception, and Recognition

This dataset documents study selection and categorical coding for a systematic review of sensing, perception, and safety recognition on construction sites. Scopus and Web of Science Core Collection searches identified 617 records. Removing 11 records without a DOI and 187 duplicates left 419 publications for GPT-4o-assisted screening and human review. Of 194 reports sought for retrieval, 9 could not be retrieved and 185 were assessed, resulting in 136 included primary studies and 49 full-text exclusions. The 419-record inventory provides identifying bibliographic fields, database coverage, preliminary GPT-4o recommendations, original human decisions, current screening routes, final review statuses, and full-text exclusion reasons. Version 2026-09-29 adds seven fields for all included studies: sensing family, sensing modality, perception Target × Level pairings, primary recognition task, validation setting (S / L / O), strongest evaluation evidence (Cross / Same / Feas), and study region. Supporting files include the data dictionary, included and excluded subsets, a 24-record additional reviewer check, PRISMA counts, screening configuration, and an audit workbook. The dictionary maps Yes / Maybe / No to include / maybe / exclude. The package excludes raw database and Notion exports, abstracts, author keywords, citation counts, database-specific identifiers and deep links, article full texts, reviewer identities, and private analytical notes. CC BY 4.0 covers the review team's annotations, decisions, categorical coding, documentation, organization, and audit summaries; it does not transfer rights in underlying publications or database products.