Insider Knowledge: How Much Can RAG Systems Gain from Evaluation Secrets?

TL;DR

This paper examines the risks of circularity in evaluating RAG systems with LLM judges, demonstrating how leaked evaluation components can lead to misleadingly high scores.

Contribution

It highlights the potential for evaluation bias in nugget-based RAG systems and emphasizes the need for blind evaluation methods to ensure genuine progress.

Findings

Leaked evaluation elements can produce near-perfect scores.

Modified systems can optimize outputs for LLM judges.

Blind evaluation settings are crucial to prevent metric overfitting.

Abstract

RAG systems are increasingly evaluated and optimized using LLM judges, an approach that is rapidly becoming the dominant paradigm for system assessment. Nugget-based approaches in particular are now embedded not only in evaluation frameworks but also in the architectures of RAG systems themselves. While this integration can lead to genuine improvements, it also creates a risk of faulty measurements due to circularity. In this paper, we investigate this risk through comparative experiments with nugget-based RAG systems, including Ginger and Crucible, against strong baselines such as GPT-Researcher. By deliberately modifying Crucible to generate outputs optimized for an LLM judge, we show that near-perfect evaluation scores can be achieved when elements of the evaluation - such as prompt templates or gold nuggets - are leaked or can be predicted. Our results highlight the importance of blind evaluation settings and methodological diversity to guard against mistaking metric overfitting for genuine system progress.