The Ranker conducts multiple parallel evaluation passes where the candidate draft is inserted into the competitive set, prompting the model to sort all items by query relevance and output both a ranked list of identifiers and comparative rationales.
Referenced by
