AI-Mediated Judgment Cities: Governance Application Shells for Auditable Judgment Delegation

This working paper introduces the concept of the Judgment Infrastructure Gap in AI governance: the structural absence of observer-facing mechanisms that make the delegation of judgment to AI systems explicit, auditable, and reversible within institutional decision workflows. While existing AI governance frameworks primarily focus on model behavior, outputs, compliance, and ethical principles, they leave largely unaddressed how judgment authority shifts between human actors and AI systems in practice—particularly under conditions of scale, time pressure, and institutional liability. This paper argues that many governance failures attributed to AI error or misalignment instead originate from invisible judgment delegation and asymmetric responsibility allocation. To address this gap without disclosing internal system logic or model implementations, the paper proposes a governance-oriented, system-agnostic framework centered on observer-facing judgment markers and risk trajectories. These concepts are operationalized through three AI-Mediated Judgment Cities—government decision support, healthcare triage, and educational assessment—which function as standardized application shells illustrating recurring governance failure modes across domains. The contribution of this work is threefold: (1) it reframes AI governance from a model-centric problem to a judgment delegation problem; (2) it provides a deployable conceptual and design framework for making judgment flows observable at the institutional level; and (3) it establishes judgment infrastructure as a necessary complement to existing AI risk management and governance regimes. This paper is intended as a field-defining working paper suitable for academic citation, policy discussion, and further empirical or systems research, while deliberately preserving system integrity by excluding internal state-generation mechanisms.