Decoder. plain-English AI glossary

Recall at K

● Core

Also called Recall@K

Of all the right answers out there, how many showed up in your top K results.

Think of it like

Casting a net and asking: of all the fish in the pond you wanted, how many did you actually pull up?

Example

If a query has 4 relevant docs and 3 appear in the top 10, recall@10 is 0.75.

How it actually works

Recall@K measures coverage: did retrieval surface the relevant items within the first K? It matters most in RAG, where a missed passage means the model literally cannot answer. High recall usually comes at the cost of precision — a bigger K catches more truth but also more noise, which then leans on reranking to sort out.

For product teams

If recall is low, no amount of prompt tuning helps — the answer never reached the model.

For engineers

|relevant ∩ top-K| / |relevant|; the recall/precision trade-off is set by K and the retriever.

Related

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open DecoderAdd to Chrome