Collection of reward models that judge how good an Ash "action" (message) is given the context of a conversation.