Researchers proposed Error-Propagation Modeling for Failure Attribution (EMFA) to improve failure attribution in LLM-based multi-agent systems. EMFA constructs a structured representation of the failed trajectory, models error propagation, and identifies the decisive error.