Scenario Based Government Training: Operationalizing Epistemic Accountability through Auditable Institutional Reasoning

This paper proposes a conceptual framework for scenario based government employee training that uses AI to support, not replace, the judgment of public servants. The framework is organized around four core principles: scenario based learning grounded in case based reasoning, adaptive learning tied to measured competence, hierarchical knowledge organization tied to legal authority, and source based accountability tied to validated source items. A fifth design commitment, linguistic parity across the working languages of the jurisdiction, is treated as architectural rather than downstream. At the center of the framework is the Concept Domain Claim (CDC) trace, a six field per output accountability artifact that documents the substantive claim, layer of authority, source provision, version timestamp, confidence indicator, and adaptation context for every AI generated training output. The CDC trace makes the reasoning behind each output inspectable, auditable, and reviewable by supervisors, learners, and institutional stakeholders. The framework also specifies scenario design rules for epistemic accountability, a competency progression from procedural cases to institutional judgment, an assessment mechanism that measures judgment quality rather than recall, engineered friction against automation bias, and a trunk and branch validation protocol that respects the capacity constraints of real public sector institutions. An A/B testing research design is proposed to evaluate whether scenario based training produces more auditable decisions than conventional methods. The framework is integrated for the Commonwealth of Puerto Rico, including its layered legal hierarchy of federal, Commonwealth, and municipal authority, its bilingual Spanish and English operating conditions, and the seventy eight municipalities operating under the Municipal Code (Law 107-2020) and the human resources framework of Law 8-2017. May 2026.