1.  [](/)
2.  [Diagrams](/diagrams)
3.  One Ranking, Four Metrics

[Back to Diagrams](/diagrams)Diagram Series

# One Ranking, Four Metrics

One Ranking, Four Metrics

![One ranking, four metrics: with relevant results at positions 1, 3, 7, and 10, Precision@5 = 0.40, Recall@5 = 0.50, MRR = 1.0, NDCG@10 = 0.83. The right metric depends on how users read your results.](/diagrams/eval-metrics.webp)Full screen

One Ranking, Four Metrics

Four metrics, four different questions about ranking quality. Pick the wrong one and you optimize for the wrong behavior.

Precision@K asks how many of the top K results are relevant. It matters when every result on screen carries equal weight, like a recommendation grid or a product carousel.

Recall@K asks how many of all the relevant items made it into the top K. It matters when completeness is the whole point: legal discovery, medical search, compliance audits.

MRR asks where the first relevant result shows up. It matters when people only look at the top hit. Voice assistants, autocomplete, "I'm feeling lucky."

NDCG rewards relevant results near the top and discounts them as they sink down the page. It matters when users scan a full page of results. Web search, document retrieval, any ranked feed.

The part most people miss is that these metrics disagree. A system with perfect MRR can have terrible recall. A system with great recall can score low on NDCG if all the relevant items sit at the bottom.

Pick the metric that matches how your users actually behave. Anything else is just measuring to fill a dashboard.

## Where this diagram appears

*   [Guide: evaluating multimodal retrieval →](/guides/evaluating-multimodal-retrieval)

Run this on your own data

Mixpeek turns video, images, audio, and documents in your object storage into searchable, timestamped results through one API.

[Search your own data](https://studio.mixpeek.com/namespaces/create?mode=managed&utm_source=homepage&utm_medium=diagram_detail&utm_campaign=mvs_onramp)

## Related diagrams

[

![Choosing a retrieval eval metric is a product decision, not a math one: use Mean Reciprocal Rank (MRR) when users trust the first result (support bots, agents, voice assistants), NDCG@10 when users scan a page of results (product and media search, e-commerce), and recall@k when users need every match (legal discovery, brand safety, compliance). Optimizing the wrong metric makes the dashboard improve while the product degrades, and unmeasured retrieval is not working: it is just not failing loudly yet.](/diagrams/foundations/eval.webp)

Pick the metric that matches user behavior



](/diagrams/eval)[

![Documents fill empty buckets with a mean; queries sum and leave empty buckets zero: the asymmetry is what makes the FDE dot product approximate Chamfer / MaxSim.](/diagrams/muvera-query-doc-asymmetry.webp)

MUVERA: Why Query and Document Are Encoded Asymmetrically



](/diagrams/muvera-query-doc-asymmetry)[

![Tier 1 FDE-encodes the query and does fast single-vector ANN over all N to get a top-K; tier 2 reranks only those K with exact Chamfer over the original token vectors.](/diagrams/muvera-two-tier-retrieval.webp)

MUVERA: Two-Tier Retrieval (Cheap FDE Recall, Exact Rerank)



](/diagrams/muvera-two-tier-retrieval)