AIRIA: A Human-AI Framework for Geospatial Regulatory Interpretation Alignment

Abstract

Multi-vendor enterprise systems operating under regulatory mandates face a structural software quality failure that existing governance frameworks do not address: interpretation divergence. When multiple vendors independently translate the same regulatory requirement into system behavior, each may produce an implementation that is internally correct yet incompatible with the others at system integration boundaries. In regulated healthcare exchange programs under the Affordable Care Act (ACA), this failure is compounded by geospatial risk—regulatory terms such as service area, county boundary, and rating region carry implicit spatial data model dependencies that different vendors resolve through incompatible geographic representations, producing divergent eligibility determinations for the same enrollees.

This paper introduces AIRIA: the AI-Assisted Regulatory Interpretation Alignment framework. AIRIA is a Design Science Research artifact—a five-layer human-AI architecture that intercepts interpretation divergence before it propagates into acceptance criteria and implementation. The five layers are: (1) Regulatory Ingestion, (2) Ambiguity Detection, (3) Geospatial Interpretation Analysis, (4) Regulatory Knowledge Graph construction, and (5) Divergence Detection and Alerting. AI performs computationally intensive consistency work across these layers while human governance structures retain all binding decision authority.

This paper proposes a Design Science Research framework to investigate three research questions: whether geospatial regulatory terms can be systematically classified by spatial data model divergence risk; whether a knowledge graph schema can represent regulatory interpretation lineage with explicit geospatial nodes; and whether AI-assisted governance can detect multi-vendor divergence before implementation. AIRIA is grounded in more than 13 years of practitioner observation across healthcare exchange and government-regulated enterprise implementations involving multiple independent vendor organizations, presented as a conceptual architecture with a five-phase empirical validation roadmap. Contributions include a regulatory geospatial term taxonomy, a Geospatial Ambiguity Risk Score formula, a nine-node Regulatory Knowledge Graph schema, and an Alignment Confidence Score model—all specified as testable artifacts for future empirical validation.