RCScore: Quantifying Response Consistency in Large Language Models

TL;DR

RCScore is a framework for measuring how instruction style affects large language model responses, revealing significant performance variations and introducing a new metric for assessing model consistency and robustness.

Contribution

It introduces RCScore and CRS metrics to quantify instruction sensitivity and stylistic self-consistency in large language models, addressing limitations of traditional evaluation methods.

Findings

Instruction style can alter accuracy by up to 16.7% points.

Deterministic decoding yields more stylistically stable outputs.

Model scale positively correlates with cross-style consistency.

Abstract

Current LLM evaluations often rely on a single instruction template, overlooking models' sensitivity to instruction style-a critical aspect for real-world deployments. We present RCScore, a multi-dimensional framework quantifying how instruction formulation affects model responses. By systematically transforming benchmark problems into multiple instruction styles, RCScore reveals performance variations undetected by conventional metrics. Our experiments across ten LLMs on four reasoning benchmarks demonstrate that instruction style can shift accuracy by up to 16.7% points. We introduce Cross-Response Similarity (CRS), a method applying RCScore metrics to measure stylistic self-consistency, and establish its strong correlation with task accuracy, suggesting consistency as a valuable proxy for model reliability. Additional findings show that deterministic decoding produces more stylistically stable outputs, and model scale correlates positively with cross-style consistency. RCScore offers a principled approach to assess instruction robustness.