Representation Governance: Producing Governed State from Unstructured Reality

Deterministic governance operates over governed state. The Multi-Dimensional Deterministic Governance Model identifies five assurance functions that must be controlled and demonstrably composed at the runtime authorization boundary: governance ontology, constraint codification, deterministic decision logic, the authorization artifact, and execution enforcement. This paper examines an antecedent control problem: how unstructured or heterogeneous source inputs become the governed state on which those functions rely. It defines representation governance as the controlled production of governed state from source inputs under declared and versioned ontology, constraint, and provenance conditions. Representation governance is not a sixth assurance function, a truth-determination mechanism, or an authorization verdict producer. It names a control obligation at the input-producing edge of the five-function model. Its output becomes a bound input to deterministic decision logic, and the authorization artifact binds the source inputs, established provenance evidence, applicable versions, authority state, governed state, evaluation evidence, and verdict required for independent reconstruction. A central observation follows: Deterministic decision logic guarantees identical outcomes for identical governed state. It does not guarantee that identical real-world phenomena produce identical governed state. Representation either produces governed state with the required provenance and version evidence or leaves representation unestablished. If required meaning, origin, authority, or evidence is missing or unresolved, the runtime authorization boundary emits ABSTAIN, and execution remains blocked pending authorized human override. Only the boundary emits ALLOW, DENY, or ABSTAIN. Provenance establishes origin, not truth. Replay is tied to a declared mode, State-Replay or Protocol-Replay, and establishes reproducibility under that mode, not substantive correctness. Ungoverned representation is treated as a failure condition at the input-producing edge of the boundary, classified by the integrity or composition property it defeats, not as a new failure family or a sixth function. The discussion is vendor-neutral and implementation-neutral. It names no product and prescribes no implementation. Within a declared workflow that holds action vocabulary, meaning, policy, authority, state, and execution topology sufficiently fixed, a bounded execution gate can support genuine pre-execution authorization; the paper addresses what must be made explicit when deterministic authorization reaches beyond that boundary. Representation governance controls how source inputs become governed state. Pre-execution authorization controls whether a proposed action may proceed on that state. Version 3.0.0 (August 2026) supersedes v2.0 (June 2026). It aligns the paper with the parent model at v2.0.0, relocates the verdict to the runtime authorization boundary, adds the provenance and replay-mode treatment, replaces the worked example, updates the reference spine, and relicenses the work under CC BY 4.0. This concept DOI resolves to the current version. Related publications: Deterministic Governance Is Multi-Dimensional (DOI: 10.5281/zenodo.18902166), A Taxonomy of AI Governance Approaches (DOI: 10.5281/zenodo.18275969), The Authorization Boundary Integrity Model (DOI: 10.5281/zenodo.20929115), The Closed-World Bargain (DOI: 10.5281/zenodo.21643658), The Five Tests Standard (5TS) (DOI: 10.5281/zenodo.21040295), The Authorization Artifact Test (DOI: 10.5281/zenodo.20013582), Versioned Meaning: How to Make Ontologies Audit-Stable (DOI: 10.5281/zenodo.18328587), and Evidence Projection (DOI: 10.5281/zenodo.20277841).