Trust Calibration

Trust calibration is the dynamical process by which agents update their assessments of another's reliability based on accumulated evidence from interaction outcomes. It is not a Bayesian computation performed in isolation; it is a network-coupled phenomenon in which each agent's calibration is shaped by the calibration of their neighbors. When your friends distrust a source, your own distrust amplifies even without direct experience — a mechanism that can produce information cascades of misplaced trust or distrust.

The process connects epistemic vigilance — the cognitive mechanisms for evaluating testimony — to structural causation: the network topology determines which evidence reaches which agents, and therefore whose calibration is accurate and whose is systematically distorted. In tightly clustered trust networks, calibration errors can persist for long periods because dissenting signals never cross community boundaries. The study of trust calibration thus requires not merely psychology but reputational dynamics — the formal analysis of how reputation propagates, decays, and saturates in social networks.

The calibration of trust is not a rational update; it is a social process that only sometimes produces rational outcomes.

The Network Geometry of Calibration

The accuracy of trust calibration depends on the topology of the trust network in which it operates. In a random network with short path lengths and low clustering, calibration errors dissipate rapidly: a false signal of untrustworthiness propagates but collides with contradictory evidence before it can saturate the network. In a small-world network, calibration is faster but more fragile: the short paths that enable rapid correction also enable rapid contagion of misinformation. The network structure is not merely a channel through which trust flows; it is a filter that determines which evidence reaches which agents and in what order.

In highly clustered networks — the typical structure of political communities, scientific specialties, and online echo chambers — calibration errors become self-sustaining. An agent whose neighbors all distrust a source receives no contradictory signals, and their distrust becomes self-validating. This is not irrationality; it is rationality operating on a structurally incomplete evidence base. The agent updates correctly given what they know; what they know is determined by the network's community structure. The error is systemic, not individual.

The implication is that trust calibration cannot be understood as either a purely individual cognitive process or a purely social contagion. It is a coupled dynamics: individual Bayesian updating at the node level, network-mediated evidence propagation at the graph level. The rationality of the whole system depends on the rationality of its parts and the topology of their coupling. A network of rational agents can produce collectively irrational outcomes if the network topology systematically filters evidence. This is the trust-calibration analogue of the activator-inhibitor principle: local rules produce global patterns that none of the local rules explicitly encode.