Proceedings of the Annual Meeting of the Cognitive Science Society

Humans routinely treat propositions as settled for the purposes of reasoning and action despite remaining uncertainty. Standard probabilistic models capture graded belief, but they leave the formation and epistemic role of such commitments underspecified. We propose a minimal Bayesian network model in which commitment is represented as an endogenous endorsement event regulated by metacognitive confidence. Within this framework, endorsement functions as a confidence-gated internal signal: conditioning on endorsement can increase subjective probability once, without new external evidence, while its dependence structure prevents epistemic bootstrapping. Extending the model to include explicit evidence yields a distinctive empirical prediction: the effect of evidence on commitment is systematically moderated by confidence. This prediction supports experimental designs in which evidence strength and confidence are independently manipulated and both commitment and probability judgments are measured. The model thus provides a tractable and empirically testable account of commitment under uncertainty.