CSR Validation Rules and Evaluation Artifacts

This archive contains the validation rules and evaluation artifacts used to assess reconstruction fidelity in the Controlled Scientific Reconstruction (CSR) framework described in: Flah, M. & Nehdi, M. L. Controlled scientific reconstruction of concrete durability datasets from literature using schema-constrained large language model workflows The materials included in this record implement the validation layer applied to reconstructed datasets in order to verify structural integrity, schema consistency, and numerical agreement with reference datasets. The archive contains: • Validation rules implementing structural and domain-specific constraints applied to reconstructed tables• Validation prompts used to translate natural-language validation specifications into executable rule logic• Validation reports summarizing rule outcomes, detected inconsistencies, and reconstruction diagnostics• Evaluation artifacts produced during the case-study assessments, including intermediate outputs and validation summaries These artifacts correspond to the validation stage of the CSR pipeline and support the evaluation of reconstruction fidelity using the metrics reported in the manuscript, including row-count match rate, field balance index (FBI), absolute error (AE), and symmetric relative error (sRE). The CSR software implementation is archived separately as the CSR Reproducibility Kernel, and the prompt and runtime configuration package is provided in a companion Zenodo record. Together, these artifacts form a versioned reproducibility bundle supporting transparent inspection of the reconstruction workflow and its validation procedures. This archive is released as a versioned and citable research artifact to facilitate transparency, auditability, and independent evaluation of the CSR validation framework.