Implementation:Explodinggradients Ragas FactualCorrectness Metric

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

Overview

FactualCorrectness evaluates the factual accuracy of responses by decomposing text into atomic claims and verifying them against a reference using Natural Language Inference (NLI).

Description

The FactualCorrectness class decomposes a response into atomic claims using a claim decomposition prompt, then verifies each claim against the reference text using an NLI prompt. It supports three evaluation modes: precision (how many response claims are supported by the reference), recall (how many reference claims appear in the response), and F1 (harmonic mean of both). The decomposition granularity is controlled by atomicity and coverage parameters. It inherits from MetricWithLLM and SingleTurnMetric.

Key attributes:

• mode -- Evaluation mode: "precision", "recall", or "f1" (default "f1").
• beta -- Beta value for F-beta score calculation (default 1.0).
• atomicity -- Level of claim decomposition: "low" or "high" (default "low").
• coverage -- Level of claim coverage: "low" or "high" (default "low").
• claim_decomposition_prompt -- Prompt for breaking text into claims.
• nli_prompt -- Prompt for NLI verification of claims.

Usage

The metric requires response and reference columns. An LLM must be configured.

Code Reference

Property	Value
Source Location	src/ragas/metrics/_factual_correctness.py L165-307
Class Signature	class FactualCorrectness(MetricWithLLM, SingleTurnMetric)
Import	from ragas.metrics import FactualCorrectness

I/O Contract

Inputs

Parameter	Type	Required	Description
response	str	Yes	The generated response to evaluate
reference	str	Yes	The ground truth reference text

Outputs

Output	Type	Description
score	float	Factual correctness score (0.0 to 1.0), computed based on the selected mode

Usage Examples

from ragas.metrics import FactualCorrectness
from ragas.dataset_schema import SingleTurnSample

metric = FactualCorrectness(mode="f1", atomicity="low", coverage="low")
# metric.llm = ...  # Set your LLM

sample = SingleTurnSample(
    response="Albert Einstein was a German theoretical physicist who developed the theory of relativity.",
    reference="Albert Einstein was a German-born theoretical physicist. He developed the theory of relativity and contributed to quantum mechanics."
)
# score = await metric.single_turn_ascore(sample)

Related Pages

• Explodinggradients_Ragas_AnswerCorrectness_Metric -- Alternative correctness metric combining factuality with semantic similarity
• Explodinggradients_Ragas_Faithfulness_Metric -- Shares the NLI prompt for statement verification
• Explodinggradients_Ragas_NoiseSensitivity_Metric -- Measures the impact of noise on response accuracy