What kinds of use cases cause this kind of situation, where the query and indexed vectors are from different distributions?