Verifying vNM-rationality requires an ontology — LessWrong