Factual correctness

Factual correctness is a RAGAS metric that compares the claims in an answer with the claims in a reference and reports their precision, recall or F1, with no embedding step.

Last updated: 29 Sep, 2026 · RAGAS 0.4.3

Answer correctness blends facts with wording. When you want the facts alone, and want to choose whether extra claims or missing claims matter more, factual correctness gives you the claim comparison on its own.

The FactualCorrectness API

python

from ragas.metrics.collections import FactualCorrectness

FactualCorrectness(llm=judge, mode="f1")         # both extra and missing claims count
FactualCorrectness(llm=judge, mode="precision")  # only extra claims lower the score
FactualCorrectness(llm=judge, mode="recall")     # only missing claims lower the score

It reads only response and reference: no question, no chunks.

The reference from golden g002

python

reference = "The ProBook X1 has 16GB DDR5 RAM and a 512GB NVMe SSD."

A complete answer and a partial one

python

answers = ["The ProBook X1 has 16GB DDR5 RAM and a 512GB NVMe SSD.",
           "The ProBook X1 has 16GB DDR5 RAM."]

Project files used on this pageThis lesson builds on a project from earlier lessons. The code below imports this file. Click a file to see its code, or follow the link to the lesson that wrote it. To run the code yourself, keep it in the same folder.

written in LLM as a judge

View the code here

judge.py

import os

from google import genai
from openai import AsyncOpenAI
from ragas.embeddings import GoogleEmbeddings
from ragas.llms import llm_factory

groq = AsyncOpenAI(
    api_key=os.environ.get("JUDGE_GROQ", os.environ["GROQ_API_KEY"]),
    base_url="https://api.groq.com/openai/v1",
)
judge = llm_factory("openai/gpt-oss-20b", provider="openai", client=groq)


class OneTextPerCall(GoogleEmbeddings):
    """gemini-embedding-2 turns a list into one embedding, so embed each text on its own."""

    def embed_texts(self, texts, **kwargs):
        return [self.embed_text(text) for text in texts]

    async def aembed_texts(self, texts, **kwargs):
        return [await self.aembed_text(text) for text in texts]


embeddings = OneTextPerCall(client=genai.Client(), model="gemini-embedding-2")

Three modes on two answers

ExampleAPI key

from judge import judge
from ragas.metrics.collections import FactualCorrectness

reference = "The ProBook X1 has 16GB DDR5 RAM and a 512GB NVMe SSD."
answers = ["The ProBook X1 has 16GB DDR5 RAM and a 512GB NVMe SSD.",
           "The ProBook X1 has 16GB DDR5 RAM."]

for mode in ["precision", "recall", "f1"]:
    metric = FactualCorrectness(llm=judge, mode=mode)
    scores = [metric.score(response=text, reference=reference).value for text in answers]
    print(f"{mode:<9}  complete={scores[0]:.2f}  partial={scores[1]:.2f}")

Output

precision  complete=1.00  partial=1.00
recall     complete=1.00  partial=0.50
f1         complete=1.00  partial=0.67

How the modes treat a missing fact

precision asks whether what the answer says is in the reference. The partial answer says nothing false, so it keeps a high precision.
recall asks whether the reference's claims are in the answer. The partial answer leaves out the SSD, so its recall drops.
f1 balances the two, so the partial answer lands between them.