WCD-SBAS: An Explainable Welfare-Centric Data Warehousing Architecture for Automated Scheme–Beneficiary Alignment and Policy Decision Support

Abstract

Social welfare delivery increasingly de-pends on information distributed across identity, civil-registration, income, education, health, employment, payment, grievance, and program-specific information systems. Such frag-mentation creates duplicate beneficiary records, stale or con-flicting attributes, heterogeneous schemas, limited provenance, opaque eligibility decisions, and weak cross-program analyti-cal capability. This paper proposes WCD-SBAS, an explain-able welfare-centric data warehousing architecture for auto-mated scheme–beneficiary alignment and policy decision support. The architecture integrates heterogeneous administrative sources through a governed interoperability and staging layer, performs privacy-aware beneficiary entity resolution, maintains a tempo-rally versioned beneficiary and household master, and organizes welfare processes using conformed dimensional models. Statutory eligibility is represented through versioned three-valued rules that distinguish satisfied, failed, and unknown conditions, allow-ing incomplete or stale evidence to be routed for verification instead of being interpreted as automatic ineligibility. Evidence-linked decision traces retain rule versions, source systems, times-tamps, and transformation provenance, while the OLAP layer supports coverage analysis, verification monitoring, grievance analysis, payment monitoring, and non-binding policy what-if evaluation. The proposed workflow is evaluated using a controlled synthetic validation environment designed to examine entity-resolution robustness, incomplete eligibility evidence, analytical-query behavior, and scalability. Under a stringent 40% field-corruption condition, the WCD-SBAS hybrid entity resolver achieves an F1 score of 0.817±0.002, compared with 0.810±0.003 for deterministic multi-field matching, 0.808 ± 0.002 for a weighted probabilistic matcher, and 0.674 ± 0.002 for exact-ID matching. The evidence-aware rule engine automatically resolves 93.3% of evaluated beneficiary–scheme pairs and routes 6.7% for verification when mandatory evidence is unavailable or stale. In the modeled analytical workload, dimensional queries obtain median speedups of approximately 7.75–8.48× over fragmented operational-query execution. Additional stress experiments show progressive degradation as linkage evidence becomes corrupted and increasing verification rather than automatic denial as eligibility evidence becomes incomplete. These findings demon-strate the functional behavior and robustness of the proposed architecture within the controlled validation environment; evalu- ation on authorized administrative data and production database platforms is required before operational performance or policy-effectiveness claims can be made.