The Error You See Is Not the Error You Made: Progression-aware Reasoning Origin for Reasoning Error Localization

2026-09-27

AI Summary

This study addresses the challenge of error localization in multi-step reasoning with large language models, where error propagation causes the first rejected step to diverge from the true error origin. To tackle this, we propose PRO, a framework that reformulates verification as a progress-aware error source localization problem. We formally prove the insufficiency of relying solely on forward rejection evidence and introduce a training-free attribution paradigm that jointly models preceding support and backward compatibility to identify signal conflict regions, leveraging interventional evidence to distinguish genuine error sources from their propagated manifestations. Extensive experiments across open-domain, medical, and structured reasoning tasks demonstrate that PRO significantly outperforms strong existing baselines in error localization accuracy, validating both the faithfulness and effectiveness of the proposed approach.

Abstract

Verifying multi-step LLM reasoning requires more than determining whether a trace is correct: a useful verifier should identify where the reasoning first goes wrong. However, existing holistic methods provide little positional evidence, while forward sequential verification often treats the first rejected step as the error source. Under error propagation, this assumption can fail, since an earlier mistake may remain locally plausible and become observable only through its downstream consequences. We therefore rethink reasoning verification as a progression-aware error-source localization problem: rather than asking only where a reasoning trace first appears inconsistent, we ask which earlier step best explains how that inconsistency emerges along the trajectory. Based on this view, we propose Progression-aware Reasoning Origin (PRO), a training-free framework for first-error localization. PRO jointly models incoming support from the preceding context and outgoing compatibility with subsequent reasoning, selectively refines regions where these signals disagree, and finally performs detector-conditioned source attribution with intervention-based evidence to distinguish the true error origin from its propagated manifestations. We further formalize the gap between forward rejection and structural exposure, showing why incoming-side evidence alone is insufficient for reliable localization under error propagation. Experiments across open-form, medical, and structured reasoning tasks demonstrate consistent improvements over strong verification baselines, supporting progression-aware source attribution as a more faithful formulation of reasoning verification.

Problem

• reasoning error localization
• multi-step reasoning
• error propagation
• large language models
• reasoning verification

Innovation

• Reasoning Error Localization
• Progression-aware Verification
• Training-free Framework
• Error Propagation
• Source Attribution