- The Oracle should respond to the user using Facts and Feelings; right now even its responses are just Conversations.
- Other LLM providers are not sophisticated enough to report truth claims wrapped in with a DoT baseed on output, so maybe we can map to that format using their citations as facts.
- Until we do so, we cannot end up training on the results of other LLMs (which is probably a good thing anyway).