A Replay-Verifiable Epistemic State Machine with Deterministic Governance and Instrumentation Constraints

Abstract We present a system architecture for maintaining epistemic state under adversarial conditions using a replay-verifiable, event-sourced model with deterministic governance constraints. The system separates probabilistic cognition from deterministic authority through a formally defined admission predicate and a compiled safety automaton. We introduce an intermediate instrumentation layer that produces empirical evidence without possessing authority, preventing metric capture and adversarial optimization of measurements. All system state is reconstructable via replay, providing a lineage-based notion of truth.