Governed Sovereign RAG for Public Administration: Tripartite Epistemic Routing, Facet-Guided Clarification, and Deterministic Numeric Verification in Citizen Services

A research deployment on the official government portal of Tlaxcala, Mexico, demonstrates a sovereign Retrieval-Augmented Generation (RAG) system tailored for public administration citizen services. The system spans 1,143 administrative procedures and 924 decoupled offices, resolving operational tensions among data sovereignty restrictions, underspecified citizen inquiries, and zero-tolerance regulatory standards against invented fees. The architecture employs tripartite epistemic routing, facet-guided clarification, and deterministic verification to maintain jurisdictional and procedural precision.

This initiative matters because it establishes a repeatable blueprint for public sector entities to adopt conversational AI without violating strict sovereign data protections or exposing citizens to costly generative hallucinations. By proving that deterministic guardrails and epistemic routing can function at state scale, it decouples public automation from unregulated commercial cloud models. Consequently, public agencies can preserve legal certainty while significantly reducing administrative friction at digital service desks.

Public administration digital assistants face structural friction between data sovereignty mandates prohibiting commercial cloud solutions and strict zero-tolerance thresholds for factual errors such as invented administrative fees.

Forward consideration, not a verified fact.