Beyond the Illusion of Consensus: From Surface Heuristics to Knowledge-Grounded Evaluation in LLM-as-a-Judge

TL;DR

This paper reveals that high agreement among LLM evaluators often masks superficial consensus based on surface heuristics, and proposes knowledge-grounded rubric generation to improve evaluation reliability.

Contribution

It introduces the Evaluation Illusion concept and the MERG framework for dynamic, knowledge-driven rubric generation to enhance LLM evaluation accuracy.

Findings

Model-level agreement is high but sample-level agreement is fragile.

Shared rubric structure significantly improves agreement.

Knowledge-grounded rubrics increase agreement in objective domains.

Abstract

The paradigm of LLM-as-a-judge relies on a critical assumption, namely that high inter-evaluator agreement indicates reliable and objective evaluation. We present two complementary findings that challenge this assumption. \textbf{First}, we demonstrate that this consensus is frequently illusory. We identify and formalize \textbf{Evaluation Illusion}, a phenomenon where LLM judges generate sophisticated critiques yet anchor scores on shared surface heuristics rather than substantive quality. Through a large-scale study of 105,600 evaluation instances (32 LLMs $\times$ 3 frontier judges $\times$ 100 tasks $\times$ 11 temperatures), we show that model-level agreement (Spearman $\rho = 0.99$) masks fragile sample-level agreement (Pearson $\bar{r} = 0.72$; absolute agreement ICC $= 0.67$), that merely sharing rubric structure restores 62\% of total agreement, and that high-quality outputs paradoxically receive the \textit{least} consistent evaluations. \textbf{Second}, we demonstrate that dynamically generating evaluation rubrics grounded in domain knowledge produces more meaningful assessment. We introduce MERG (Metacognitive Enhanced Rubric Generation), a knowledge-driven rubric generation framework whose domain-selective effects confirm this. Agreement \textit{increases} in codified domains (Education +22\%, Academic +27\%) where knowledge anchors evaluators on shared standards, while it decreases in subjective domains where genuine evaluative pluralism emerges. These findings suggest that evaluation rubrics should be dynamically enriched with expert knowledge rather than relying on generic criteria, with implications for reward modeling in RLAIF.