Reciprocal Rank Fusion in one paragraph

Merge ranked lists by summing 1/(k + rank). No score calibration, no training, and it beats most clever alternatives.

planted Sep 2, 2026 1 min read

You have a BM25 list and a vector list, and their scores live on different scales. Don't normalise them. For every document, sum 1 / (k + rank) across the lists it appears in, with k ≈ 60, and sort by that. A document that is decent in both lists beats one that is great in one and missing from the other, which is exactly what you want for retrieval.

score(d) = Σ  1 / (60 + rank_i(d))

Where it shows up: the hybrid search in Advanced RAG Architecture: Graph, Vector, Redis, and PostgreSQL and the judgment search in CaseCloud: multi-tenant legal SaaS with RAG over 80M+ judgments. Related paper: Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks.

Links from this note

Jun 10, 2026Advanced RAG Architecture: Graph, Vector, Redis, and PostgreSQLCombining Graph databases, Vector databases with Reciprocal Rank Fusion (RRF), Redis caching, and PostgreSQL to build a state-of-the-art RAG system.Read Dec 15, 2025CaseCloud: multi-tenant legal SaaS with RAG over 80M+ judgmentsRead Jun 2, 2025Retrieval-Augmented Generation for Knowledge-Intensive NLP TasksThe origin of the acronym. Useful mostly as a baseline to measure how far production RAG has moved: hybrid retrieval, reranking, citations and evaluation sets.Read