Tag
This paper proposes token-efficient retrieval methods for LLM analysis of transactional legal documents, demonstrating that structured retrieval can match full-corpus injection in answer quality while using significantly fewer tokens and lower cost.