You run BM25 and dense retrieval in parallel and merge the results by adding each raw BM25 score to the cosine similarity. Lexical hits dominate every query, even purely conceptual ones. What is the most robust fusion method?

1. A Reciprocal Rank Fusion, which combines results by rank position rather than raw score, so the two scorers' different scales stop mattering.
2. B Multiply cosine similarity by 10 to bring it into BM25's typical range, then add the two scores as before to produce the final ranking.
3. C Route conceptual queries, detected by a classifier, to dense-only retrieval, and keep raw score addition for all the remaining queries.
4. D Normalize BM25 by dividing by the maximum possible BM25 score across the corpus, then sum with cosine similarity for the final ranking.