Why Better Search Recall Did Not Mean Better RAG Answers

A paired 230-question experiment with the same BM25 implementation, two tool interfaces, and a result that made me inspect the evidence actually delivered to the model. Disclosure: I build Minimal RAG, the personal research project…