Functionalizing the Correspondence: A Computational Framework for Truth

This paper proposes a computational framework grounded in the correspondence theory, reconceiving truth-value neither as an agent-independent property of the world nor as a purely subjective agent-relative judgment, but as the discrete output of an executable physical process. Drawing on predictive processing, naturalized semantics, and computational epistemology, the framework characterizes truth-value as the output of a truth-process. The framework further distinguishes truth-value—the binary output of a single execution—from truth: an emergent meta-status grounded in fact-proxy quality, truth-process resilience, and applicative success. This process is rooted in the evolutionary imperative of teleological agents to maintain structurally aligned internal models of their environment through truth-process execution, minimizing surprisal, enabling successful goal-directed action. The truth-process operates through three components: an internal imagination function mapping a statement to an internal conception; a fact-determining function generating a fact-proxy from evidence within an independently specifiable domain of concern; and a correspondence function verifying their structural alignment. Where the truth-process cannot execute truth-value is undefined rather than false, necessitating a three-status logic (true, false, undefined) analogous to Kleene's strong three-valued system. The threat of relativism is addressed through a meta-agent architecture and cross-model convergence across structurally independent nodes. The framework diagnoses large language model (LLM) hallucination as a dual architectural absence: the inability to register fact-proxy generation failure and to initiate domain-directed evidence-seeking, and it reframes epistemic hygiene as the optimization of the criteria, mechanisms, and epistemic pathways by which agents construct fact-proxies.