Hybrid lexical dense

Lexical and dense retrieval fail differently, and fusing them helps. In BEIR (18 datasets, 10 systems), BM25 was a robust zero-shot baseline, re-ranking and late-interaction models achieved the best average zero-shot results at high computational cost, and dense and sparse retrievers often underperformed. Reciprocal rank fusion (Cormack et al., SIGIR 2009) consistently beat every individual system it combined. Anthropic's Contextual Retrieval write-up describes the common RAG stack as BM25 for exact matches plus embeddings for semantic similarity, merged by rank fusion, and reports that adding contextual BM25 to contextual embeddings cut top-20 retrieval failures from 5.7% to 2.9%, and adding reranking to 1.9%.