Reproducibility Package for "A Governance Architecture for AI-Assisted Validation of Construction Cost Knowledge: Combining Domain Rules with LLM Semantic Analysis"

Reproducibility package accompanying the manuscript "A Governance Architecture for AI-Assisted Validation of Construction Cost Knowledge: Combining Domain Rules with LLM Semantic Analysis" submitted to Engineering, Construction and Architectural Management (Emerald). The package contains the sanitised derived data, evaluation outputs, baseline-model implementation scripts, and a single-command verification script that reproduce every quantitative claim in the paper from the released artefacts. A reproducibility supplement (PDF) documents the full method, claim-to-artefact map, and step-by-step reproduction protocol. Contents:- verify_all_paper_numbers.py — single-command entrypoint reproducing all reported metrics (expert F1/Precision/Recall/Kappa, cross-region F1, McNemar test, i.i.d. and TenderId-cluster bootstrap CIs, false-negative-rate CI, 5-fold stability)- outputs/ — all generated tables and figures' source data (CSV + formatted Excel workbook)- research/data/ — sanitised derived feature matrices, expert labels, per-mode binary predictions, frozen statistical-test artefacts- research/scripts/ — leakage-safe baseline implementations (logistic regression, random forest, XGBoost, isolation forest, TF-IDF) under GroupKFold by TenderId- research/prompts/ — LLM prompt structure and output schema- reproducibility_supplement.pdf — full reproducibility documentation Raw BOQ text and original monetary amounts are restricted under active client contracts; the released artefacts include sanitised identifiers and component-percentage features sufficient to reproduce all reported metrics within rounding tolerance. Run `pip install -r requirements.txt` then `python verify_all_paper_numbers.py` to reproduce all numbers.