A Governance-Aware Synthetic Benchmark for Identity Resolution under Asymmetric Risk

This record contains a deterministic, governance-aware synthetic benchmark dataset for evaluating Identity Resolution (IR) and entity matching systems under asymmetric risk and signal conflict. The dataset is designed to stress-test governance-critical failure modes, particularly high-confidence false positive merges that can cause irreversible downstream impact in enterprise Master Data Management (MDM) and CRM systems. The archive includes the synthetic pair dataset, base entity records, fixed train/validation/test split indices, configuration files, and reference experiment scripts corresponding to the results reported in the associated study. The dataset is generated using a fully deterministic pipeline with fixed random seeds and configuration parameters to ensure reproducibility. It comprises multiple conflict-oriented clusters that capture common and high-severity enterprise scenarios, including hard negatives with strong lexical, geospatial, and multi-signal contradictions. This benchmark is intended to support research on risk-bounded decision policies, hard-negative evaluation, and governance-aware threshold selection beyond average-metric optimization (e.g., F1 or AUC). It is provided as a frozen archival snapshot to enable transparent evaluation and reuse in related identity resolution and record linkage studies.