Where Did It All Go Wrong? A Hierarchical Look into Multi-Agent Error Attribution

Adi Banerjee, Anirudh Nair, Tarik Borogovac

Keywords: Multi-Agent Error Attribution, Multi-Agent Systems, Large Language Models, Hierarchical Context, Consensus Voting, Objective Analysis

TL;DR: ECHO improves error attribution in LLM multi-agent systems through hierarchical context analysis and objective consensus voting, outperforming existing methods in complex interaction scenarios.

Abstract: We present ECHO (Error attribution through Contextual Hierarchy and Objective consensus analysis), a novel algorithm for error attribution in LLM multi-agent systems. While existing approaches struggle with accuracy and reliability in complex interaction scenarios, ECHO combines hierarchical context representation, objective analysis-based evaluation, and consensus voting to improve attribution accuracy. Our approach leverages positional-based contextual understanding with objective evaluation criteria. Experimental results demonstrate that ECHO outperforms existing methods across various multi-agent scenarios, particularly for subtle reasoning errors and complex interdependencies. This structured framework provides a more robust solution for error attribution in collaborative AI systems.

Submission Number: 119