Reproducibility Data and Code for Rules vs Machine Learning for Pre-Execution Validation of AI-Assisted ERP Procurement Decisions

This reproducibility package accompanies the manuscript “Rules vs Machine Learning for Pre-Execution Validation of AI-Assisted ERP Procurement Decisions: Evidence from Supplier Selection.” It contains two synthetic ERP supplier-selection benchmarks, a six-rule deterministic validation library, and three machine-learning baselines: Logistic Regression, Random Forest, and XGBoost. The package includes fixed-split and repeated-holdout evaluation results, rule-knowledge coverage analysis, paired unsafe-case disagreement analysis, publication figures, a Google Colab notebook, a standalone Python implementation, dependency specifications, a reference execution environment, and automated verification utilities. The study investigates the comparative strengths and limitations of explicit rule-based procurement validation and data-driven machine-learning validation before AI-assisted ERP procurement decisions are executed. All datasets included in this repository are synthetically generated by the supplied deterministic experimental code. No production ERP records, supplier-identifying information, employee data, customer data, or other confidential organizational information are included. The repository provides both a Google Colab-compatible workflow for accessible scientific reproduction and a locked reference environment for archival deterministic reproduction.