Trust as Signed Model Fidelity: Information Dynamics and the Structural Risks of Opacity

This paper develops a reframing of trust as signed confidence in a predictive model of another agent, where the sign indicates whether predicted behavior aligns with the predictor's interests and the magnitude indicates confidence in the prediction. The framework treats individuals and groups as agents for analytic purposes, while recognizing that group agency introduces propagation and attribution problems not present in individual cases. Four claims are developed. First, trust is signed model fidelity rather than a unipolar disposition, which separates predictability from alignment and dissolves several recurrent confusions in trust theory. Second, trust models have depth, with behavioral-surface fidelity distinct from generating-layer fidelity; the gap between the two is where manipulation, betrayal, and relational collapse reside. Third, information flow between agents operates within an envelope bounded below by mutual-verification requirements and above by channel capacity for the model demand each agent is running; coordination collapses through information starvation or overload by different mechanisms but with similar visible outcomes. Fourth, under conditions of asymmetric loss, opacity drives prior-formation toward threat-modeling, which means high-opacity agents face structural risk independent of their actual disposition. The paper closes with a discussion of what the framework does not claim and what would be required to give it predictive content beyond its present interpretive utility.