Abstract
A core tenet of Bayesian epistemology is that rational agents update by Bayesian conditionalization. Accuracy arguments in favor of this norm are well-known. Meanwhile, in the setting of quantum probability and quantum state estimation, multiple updating rules have been proposed, all of which look prima facie like analogues of Bayesian conditionalization. These include Luders conditionalization, retrodiction, and Bayesian mean estimation (BME). In this paper, we present expected-accuracy and accuracy-dominance arguments for Luders and BME, which we show are complementary rules. Retrodiction, on the other hand, is shown to be accuracy-dominated, at least on many measures.
