Typed UNKNOWN: Preserving Resolution-Relevant Semantics in AI-Mediated Decision Processes

This paper proposes Typed UNKNOWN, a semantic representation for preserving resolution-relevant distinctions among unresolved conditions in AI-mediated decision processes. Conventional systems may represent heterogeneous unresolved conditions using a single generic UNKNOWN value. While this preserves the fact that determination has not been completed, it may collapse distinctions that remain relevant to subsequent resolution. Typed UNKNOWN addresses this representational problem through a compositional four-component state vector distinguishing evidence acquisition or verification, domain or scope clarification, input interpretation clarification, and governance-boundary review. The proposed representation is deliberately limited to semantic representation. It does not define runtime execution, authorization, institutional legitimacy, actor assignment, accountability, Human-in-the-Loop control, or transition enforcement. In particular, resolution-oriented semantic classes are descriptive categories and do not constitute executable obligations or assignments of responsibility. The paper formally defines the typed unresolved state, its compositional Boolean representation, resolution-oriented semantic mappings, and a projection from typed states to generic UNKNOWN. The formal analysis demonstrates that this projection can be non-injective and can therefore collapse resolution-relevant semantic distinctions. The paper also defines operational boundary conditions, a minimum decision-time preservation requirement, illustrative cases, and a future evaluation protocol for type distinguishability, inter-rater agreement, and distinction preservation. The contribution is intended as a semantic representation layer preceding subsequent research on provenance, temporal memory, reconstruction, governance, and AI-mediated state-transition control.