HPFA: Hypergraph-Based Paired Failure Attribution for LLM Reasoning

2026-08-03

AI Summary

This work addresses the challenge that large language models struggle to accurately pinpoint erroneous steps during reasoning, as existing approaches either incur high computational overhead or overlook the nonlinear logical dependencies within reasoning paths. To this end, the paper proposes a Hypergraph-based Paired Failure Attribution framework (HPFA), which introduces hypergraphs for the first time to model the complex dependencies in reasoning trajectories. By contrasting the hyperedge structures of failed and successful paths, HPFA efficiently narrows the search space to identify root causes of errors. Integrated with attribution-aware data synthesis and supervised fine-tuning, HPFA significantly improves both accuracy and efficiency of error attribution on mathematical reasoning and agent programming tasks. The resulting lightweight attribution model consistently enhances reasoning performance at test time, outperforming baselines lacking either graph-structured modeling or paired-path comparison mechanisms.

Abstract

Reflection is a powerful mechanism for LLM reasoning, yet its effectiveness hinges on accurately attributing failures to specific reasoning steps, a capability that current models notably lack. Existing failure attribution methods either require expensive step-by-step counterfactual testing that scales poorly with trajectory length, or treat reasoning traces as flat sequences that ignore the inherent non-linear logical dependencies. We propose a hypergraph-based paired failure attribution (HPFA) framework that attributes the failure root cause by comparing the hyperedges of the targeted failure reasoning path against a reference successful path. By reducing the search space, our method efficiently localizes root causes and enables scalable synthesis of attribution data for training a lightweight attributor model via supervised fine-tuning and reinforcement learning. Experiments on mathematical reasoning and agentic coding tasks demonstrate that HPFA can dramatically increase attribution accuracy and efficiency, and the trained attributor consistently improves reasoning accuracy at test time, outperforming baselines that lack graph structure or paired analysis.

Problem

• failure attribution
• LLM reasoning
• reasoning steps
• logical dependencies
• root cause localization

Innovation

• hypergraph
• failure attribution
• paired analysis
• reasoning trace
• root cause localization