Counterfactual Graph for Multi-Agent LLM Calibration

2026-05-28

AI Summary

In multi-agent large language model systems, communication often induces spurious correlations and false consensus, rendering confidence estimates based on voting agreement unreliable. To address this, this work proposes the CAGE-CAL framework, which introduces a counterfactual graph comparison mechanism—constructing and contrasting agent dependency graphs under conditions with and without communication—to explicitly model pairwise and group-level dependencies. This enables topology-aware confidence calibration beyond simple vote counting. By distinguishing genuine consensus from communication-induced agreement, the method significantly enhances reliability discrimination across five benchmarks, achieves competitive Expected Calibration Error (ECE), and improves the selection of effective communication topologies.

Abstract

Multi-agent LLM systems often treat agreement as evidence: when many agents in a panel give the same answer, that answer is assumed to be more reliable. We show that this assumption can fail after agents communicate. Communication can induce correlated failures and false consensus, so the same vote share may reflect reliable agreement in one topology but over-confidence in another. We propose CAGE-CAL, a counterfactual agent-graph calibration framework for multi-agent LLMs. For each query, CAGE-CAL compares an observed post-communication agent graph with a matched counterfactual no-communication graph, capturing both pairwise failure correlations and group-level dependencies. Rather than simply counting how many agents agree, CAGE-CAL estimates the counterfactual shift between observed and no-communication dependence, and calibrates confidence accordingly. Across five benchmarks, CAGE-CAL improves reliability discrimination with competitive ECE, and its calibrated confidence further improves topology selection over the best fixed-topology strategy.

Problem

• multi-agent LLM
• calibration
• counterfactual
• communication topology
• false consensus

Innovation

• counterfactual calibration
• multi-agent LLMs
• agent communication
• confidence calibration
• graph-based reasoning