Evolutionary Origins and Functional Architecture of Metacognitive Control: A Non-Normative Framework for Model Reliability

This paper proposes an evolutionary-functional framework for understanding metacognitive control mechanisms. Departing from normative accounts of cognition as an "ideal observer" or "truth-seeker," we posit that cognitive architectures are optimized for risk minimization and uncertainty management rather than veridical representation. We introduce "Architectural Irreducibility" as a fundamental safeguard against the formation of chronically maladaptive global models. Metacognition is framed not as an auxiliary high-level faculty, but as a structural necessity emerging from hierarchical complexity and the partial autonomy of regulatory loops. The framework delineates specific architectural layers — from reflexive processing to metacognitive monitoring — and their functional roles in preventing what we term "false global confidence."