What Defines Good Reasoning in LLMs? Dissecting Reasoning Steps with Multi-Aspect Evaluation

TL;DR

This paper proposes a granular, multi-aspect evaluation method called CaSE for assessing reasoning quality in LLMs, focusing on relevance and coherence, which improves model robustness and performance.

Contribution

It introduces CaSE, a causal stepwise evaluation technique for measuring reasoning quality, and demonstrates its effectiveness in improving LLM performance through better data curation.

Findings

CaSE reliably measures relevance and coherence of reasoning steps.

Curating training data with CaSE improves LLM final-answer accuracy.

The method aligns well with human judgments on reasoning quality.

Abstract

Evaluating large language models (LLMs) on final-answer correctness is the dominant paradigm. This approach, however, provides a coarse signal for model improvement and overlooks the quality of the underlying reasoning process. We argue that a more granular evaluation of reasoning offers a more effective path to building robust models. We decompose reasoning quality into two dimensions: relevance and coherence. Relevance measures if a step is grounded in the problem; coherence measures if it follows logically from prior steps. To measure these aspects reliably, we introduce causal stepwise evaluation (CaSE). This method assesses each reasoning step using only its preceding context, which avoids hindsight bias. We validate CaSE against human judgments on our new expert-annotated benchmarks, MRa-GSM8K and MRa-MATH. More importantly, we show that curating training data with CaSE-evaluated relevance and coherence directly improves final task performance. Our work provides a scalable framework for analyzing, debugging, and improving LLM reasoning, demonstrating the practical value of moving beyond validity checks.