is there any way to store user feedback or create an eval on a request that is neither binary nor a score, but rather qualitative i.e. the "expected" model output?