Reciprocal Rank Fusion throws away raw scores and merges on rank alone: each list adds weight / (k + rank). The winner is usually good on both, not best on either.
Production stacks commonly weight dense 0.8 and keyword 0.2 with k=60. Items appearing in only one list still score, but rarely beat items ranked respectably in both.
How fusion works
Dense cosine scores are bounded; BM25 scores are unbounded, so you cannot average them. Reciprocal Rank Fusion sidesteps it by operating on rank positions alone: each system contributes weight / (k + rank) for every candidate, with k usually 60. The smoothing constant stops one pipeline’s top result from dominating, so the chunks that win rank respectably in both lists.
Paste two rankings to see the merged order, and why “good on both” beats “best on one”.
How to use it
- Enter a dense ranking and a keyword ranking.
- See them merged with Reciprocal Rank Fusion (k=60).
- Read why good-on-both beats best-on-one.
Frequently asked questions
What is Reciprocal Rank Fusion?
The algorithm engines use to merge several ranked lists into one. Each document scores the sum, across every list it appears in, of one over a constant (k=60) plus its rank. It structurally rewards documents that appear across multiple lists over one that tops a single list.
Why is ranking #1 overrated under RRF?
Because with k=60, rank 1 contributes 0.01639 and rank 10 contributes 0.01429, a 13% gap. Appearing at all in a second list is worth close to 100%. Coverage across sub-queries beats depth on one.
Is the RRF Rank-Fusion Calculator free to use?
Yes. The RRF Rank-Fusion Calculator is completely free, with no sign-up, no usage limits and no watermark on the output.
Is my data private?
Yes. The RRF Rank-Fusion Calculator runs entirely in your browser. Nothing you paste or enter is uploaded, stored or sent to any server.
Related tools and reading
Keep going with the playbooks and tools behind this one.