Term
Context Recall
Context Recall is a RAG evaluation metric (0–1) measuring how completely the retriever supplied the facts needed for the gold answer — every missing fact lowers the score.
Context Recall — explained in detail
Context Recall checks whether every fact required for the ideal answer is actually present in the retrieved context chunks. Typical implementations (RAGAS, DeepEval): 1) Atomic facts are extracted from the gold answer. 2) For each fact, an LLM judge checks whether it can be derived from at least one of the retrieved chunks. 3) Score = supported facts / total facts. A perfect recall score means the retriever delivered everything the gold answer needs — even if the ordering or the amount of noise is poor.
Example / Practical use
Question: “What is the capital of France and how many inhabitants does it have?”. The gold answer contains two facts: capital = Paris, population ≈ 2.1 million. If the retriever only returns a chunk about Paris being the capital, the population fact is missing → recall 0.5. If both facts appear in the top-k, → recall 1.0. In practice, the metric is used to tune embedding models and top-k values — too small a k drops recall, too large a k hurts precision and latency.
Distinction from related terms
Context Recall is the completeness metric, Context Precision the ranking metric. High recall with low precision = the retriever returns everything, but in any order; the inverse = clean ordering, but incomplete. Faithfulness then measures whether the generated answer stays grounded in the context — recall makes sure the right material is in the context to begin with. In eval suites, recall is often the first metric checked: without enough recall, even a perfect generator cannot produce the correct answer.
Entdecke mehr
AI Workflows by Keyword: How We Make Recurring Routines Enforceable
A typed keyword triggers a fixed AI routine — and every single step must be committed before the next one appears. Why that's the actual trick.
GlossarRecall
Recall (hit rate, sensitivity) measures the share of all actually existing relevant cases that a system finds. Formula: relevant cases found divided by all relevant cases that truly exist. Requires a known ground truth.
LexikonMeasuring LLM Quality — Evals, Benchmarks, Judges
How to tell whether an LLM system actually works: three evaluation layers from benchmarks to CI evals, plus pitfalls like Goodhart and judge bias.