A retrieval-augmented generation (RAG) system can produce a faithful answer yet miss important information because retrieval did not supply it. Context recall measures whether the retrieved context contains the specific information needed for a reference answer. Unlike document recall, it focuses on answer-relevant information rather than relevant files. section 6.3 section 6.3.3
To compute context recall:
- Prepare a ground-truth answer and split it into individual sentences. section 6.3.3
- Check each sentence against the retrieved context. Mark it attributable if its information can be inferred from the context; otherwise, mark it not attributable. section 6.3.3
- Divide the number of attributable sentences by the total number of ground-truth sentences. section 6.3.3
Financial research assistant
A reference answer about Q3 growth and Q4 risks has five key claims. Retrieved reports cover revenue, year-over-year growth, and the cloud services driver, but omit rising operating costs and Q4 risks. Context recall is 3/5, or 0.60: two claims lack retrieved support, regardless of how well the model uses the available context. section 6.3.3
Context recall requires a ground-truth answer, so it is typically computed offline using curated reference answers. It measures coverage of that reference, not whether the generated answer is faithful or relevant. Evaluate retrieval first: poor recall or precision can make generation results hard to interpret. section 6.3.3 section 6.4.1
Table 6.4 Interpreting metric combinations for RAG diagnosis
|
Faithfulness
|
Answer relevance
|
Recall
|
Precision
|
Diagnosis
|
|
High
|
High
|
High
|
High
|
Working well
|
|
Low
|
High
|
High
|
High
|
Hallucination beyond retrieved context
|
|
High
|
Low
|
High
|
High
|
Off-topic generation. Misaligned.
|
|
High
|
High
|
Low
|
High
|
Retrieval missing relevant documents
|
|
High
|
High
|
Low
|
Low
|
Retrieval is fundamentally misaligned with the query
|
|
Low
|
Low
|
Low
|
Low
|
Systemic failure
|
Use the diagnostic matrix to interpret recall alongside precision, faithfulness, and answer relevance. section 6.4.1
Low recall is a signal to investigate, not a complete root-cause diagnosis. For example, a keyword query may miss relevant information because its vocabulary differs from the corpus. In an agentic system, the problem may instead be query formulation; a reasoning gap that prevents a needed search may not be surfaced by context recall at all. Evaluate individual retrieval steps as well as the final output. section 6.2.1 section 6.4.1 section 6.1.3


