Implementation:Explodinggradients Ragas ContextRecall Metric

Field	Value
source	Repo
domains	Metrics, Evaluation
last_updated	2026-02-10

Overview

ContextRecall estimates how much of the ground truth reference can be attributed to the retrieved contexts, with multiple variants including LLM-based, non-LLM, and ID-based implementations.

Description

The context recall module provides several classes:

• LLMContextRecall -- Uses an LLM to classify each sentence in the reference as attributable (1) or not attributable (0) to the retrieved contexts, then computes the ratio of attributed statements.
• ContextRecall -- Convenience alias for LLMContextRecall.
• NonLLMContextRecall -- Uses non-LLM string similarity to compare retrieved contexts against reference contexts, applying a threshold for binary decisions.
• IDBasedContextRecall -- Directly compares retrieved context IDs with reference context IDs for recall.

The LLM-based variant uses an ensembler to aggregate multiple classification responses for robustness. The non-LLM variant supports configurable distance measures (Levenshtein, Hamming, Jaro, Jaro-Winkler).

Usage

LLM-based variants require user_input, retrieved_contexts, and reference. Non-LLM requires retrieved_contexts and reference_contexts. ID-based requires retrieved_context_ids and reference_context_ids.

Code Reference

Property	Value
Source Location	src/ragas/metrics/_context_recall.py L88-282
Class Signature	class LLMContextRecall(MetricWithLLM, SingleTurnMetric)
Import	from ragas.metrics import ContextRecall

I/O Contract

Inputs (LLMContextRecall / ContextRecall)

Parameter	Type	Required	Description
user_input	str	Yes	The user query
retrieved_contexts	List[str]	Yes	The retrieved context passages
reference	str	Yes	The ground truth reference answer

Outputs

Output	Type	Description
score	float	Recall score: ratio of attributed statements (0.0 to 1.0)

Usage Examples

from ragas.metrics import ContextRecall
from ragas.dataset_schema import SingleTurnSample

metric = ContextRecall()
# metric.llm = ...  # Set your LLM

sample = SingleTurnSample(
    user_input="What can you tell me about Albert Einstein?",
    retrieved_contexts=[
        "Albert Einstein (14 March 1879 - 18 April 1955) was a German-born theoretical physicist."
    ],
    reference="Albert Einstein was born on 14 March 1879 and was a German-born theoretical physicist."
)
# score = await metric.single_turn_ascore(sample)

A pre-configured instance is available:

from ragas.metrics._context_recall import context_recall

Related Pages

• Explodinggradients_Ragas_ContextPrecision_Metric -- Complementary precision metric for retrieval
• Explodinggradients_Ragas_ContextEntityRecall_Metric -- Entity-level context recall
• Explodinggradients_Ragas_Faithfulness_Metric -- Statement-level faithfulness against contexts