Coded evidence-map dataset and reproducibility package for "From Carbon Prediction to Integrated Carbon Decision Support: A Systematic Evidence Map of AI in the Built Environment"

Coded dataset, screening decisions, reliability instruments and regeneration scripts behind the systematic evidence map "From Carbon Prediction to Integrated Carbon Decision Support: A Systematic Evidence Map of AI in the Built Environment" (January 2018 – May 2026). Contents: a 457-record audited database (335 primary studies, 102 full-text core) with six coded facets, record roles and machine-readable exclusion reasons; the screening worklists and per-record verdicts; the codebook; the blinded reliability instruments (screening re-screens, second coding, full-text provenance) with the author adjudications; the seven-dimension quality appraisal; and the scripts that regenerate the dataset, every reported count, the reliability statistics and all figures. SHA256SUMS.txt verifies the deposit, and regenerating the figures reproduces the archived bytes. See README.md for what regenerates what. The raw Scopus retrieval is not redistributed under Elsevier API terms, and no full-text PDFs are included. The CC BY 4.0 licence covers the authors' own coding, decisions, documentation and scripts; third-party bibliographic metadata and abstracts reproduced for audit purposes remain the property of their rights holders.