Tenant-Aware HR Master Data Harmonization: A Logical Bronze-Silver-MDM-Gold Workflow for Multi-Tenant HCM Data

Enterprise human resources (HR) systems often apply inconsistent labels to shared employee master-data attributes. This study evaluates an ordered, tenant-aware harmonization workflow represented by Bronze, Silver, master data management (MDM), and Gold logical layers; it does not claim an implemented transactional lakehouse platform. Four synthetic HR datasets contribute 3,000 core employee records. A fixed-seed simulation creates four tenants and introduces controlled categorical corruption, missingness, and duplicate identifiers while retaining independent canonical labels. Exact, normalized, synonym, and bounded fuzzy matching run in sequence, and unresolved nonmissing values are referred for manual review. At a fuzzy cutoff of 0.82, the study records 8,842 nonmissing field values, 8,731 canonical assignments, and 8,480 correct assignments, giving 98.74% assignment coverage, 95.91% recall under the study definition, and an F1 of 0.9651. Field-level denominators reconstructed from the method-by-field matrix allow independent arithmetic checking and identify 251 erroneous assignments across all four matching tiers. In the deduplicated employee classification population, department and job-title vocabularies shrink from 228 raw labels to 37 mastered labels. The downstream comparison gives mixed point estimates: ROC-AUC rises by 0.0321 and PR-AUC by 0.0005, while F1 falls by 0.0181. Because paired confidence intervals and counterfactual harmonizer runs are not available, the downstream benefit is inconclusive and the robustness of the threshold is not established. Interpretation is limited to one simulation and one held-out split. The contribution is a documented logical integration protocol with an arithmetically auditable, single-cutoff synthetic benchmark result.