Costly Honesty: A Stability Condition for Epistemic Systems

Human information systems are increasingly destabilized by a structural asymmetry: the marginal cost of producing claims is often lower than the cost of verifying them. Under these conditions, false or weakly supported claims can propagate faster than they can be evaluated, not because agents are irrational, but because verification exceeds cognitive and economic constraints. This article proposes a general stability condition for epistemic systems under adversarial pressure: reliable information environments require that the expected cost of producing claims is at least as great as the expected cost of verifying them. When this condition is violated, deception scales; when it is satisfied, deception becomes economically disfavoured. This principle—termed costly honesty—is formalized as a system-level constraint rather than a property of individual cognition. It distinguishes between semantic truth (correspondence with the world) and structural truth (integrity of state transitions), and shows how constraint-based architectures can stabilize the latter even when the former remains uncertain. Proof-of-work consensus is analyzed as a concrete instantiation in which irreversible expenditure anchors credibility. However, such systems introduce a fundamental trade off: the same costs that deter deception also increase the cost of legitimate correction. This trade off is formalized through an epistemic health function H(C), yielding an optimal cost level C* that varies systematically across domains according to their relative weighting of deception versus correction failure. The framework yields falsifiable predictions about misinformation dynamics, cognitive load, trust allocation, and AI-era miscalibration, and provides a general foundation for designing epistemic systems that remain reliable under adversarial pressure.