作者位次 #3/7ACL 2026Poster· Natural Language Processing
Seeing the Whole Elephant: A Benchmark for Failure Attribution in LLM -based Multi-Agent Systems
Mengzhuo Chen · Junjie Wang · Fangwen Mu · Yawen Wang · Zhe Liu · Huanxiang Feng · Qing Wang
摘要
Failure attribution, i.e., identifying the responsible agent and decisive step of a failure, is particularly challenging in LLM-based multi-agent systems (MAS) due to their natural-language reasoning, nondeterministic outputs, and intricate interaction dynamics. A reliable benchmark is therefore...