Can We Trust the Judges? Validation of Factuality Evaluation Methods via Answer Perturbation

arXiv:2609.15561 · cs.CL · Submitted 2026-09-14 · Read on arXiv

cs.CL

Submitted: 2026-09-14

Updated: 2026-09-14

Comments: 22 pages, 7 figures. Extended version of a paper accepted at EvalLLM 2025 (CORIA-TALN 2025)

Code: https://github.com/GiovanniGatti/truthbench

License: http://creativecommons.org/licenses/by/4.0/

The gist: Evaluating the factual correctness of large language models (LLMs) is vital for many applications.

Terminology

Abstract

Evaluating the factual correctness of large language models (LLMs) is vital for many applications. But are our evaluation tools themselves trustworthy? Despite the rise of factuality-based metrics, their sensitivity and reliability remain underexplored. This paper introduces a meta-evaluation framework that systematically tests these metrics using controlled corruptions of gold standard answers. Our method generates ranked outputs with known degrees of degradation to probe how metrics capture nuanced changes in truthfulness. Our experiments reveal that pipeline-based methods, such as the RAGAS's factual correctness metric, better track degradation than LLM-as-judge approaches. We also propose a new variant of the factual correctness metric that provides a competitive and cost-efficient.

Sources

Related papers