The previous version had a qualitative likelihood relation defined by (aggregation over) rules, and an axiomatization based on these relations being observable. The current version has the corresponding results for probabilities, with some comments on the common structure to these derivations.
This version has a variant of the model (with very similar results), and some comments on its relationship to the literature on vector addition systems.
This sharpens some arguments in the note from 2007 with the same title. In particular, it emphasizes the fact that only "psychological" free will is discussed, and highlights the relationship to formal models of reasoning in general.