One query is never enough: why top RAG systems search three times

LangChain's MultiQueryRetriever and LlamaIndex's SubQuestionEngine exist for a reason. Here's the math behind multi-query fusion - and how to do it in one API call.