Multi-agent debate can improve large language model reasoning by eliciting diverse hypotheses and critiques, yet its performance is often constrained by weak moderation. Common pipelines rely on fixed budgets, agreement-based stopping, or untrained judges, leading to redundant deliberation and unreliable evidence aggregation. We cast moderation as a meta-cognitive process, monitoring debate utility, controlling deliberation, and adjudicating a final answer, and introduce Meta-Moderator, a learnable framework that dynamically regulates debate and decides when to finalize an answer. Meta-Moderator is trained independently of the debaters via outcome-driven policy optimization, making debate regulation an explicit capability rather than an incidental effect of prompting. Across five benchmarks, Meta-Moderator outperforms widely used decision layers and transfers across tasks and system configurations. Further analyses show that it allocates debate more selectively and reduces mis-aggregation after informative hypotheses appear.
Figures & tables
Figure 1: Three common issues in standard multi-agent debate: redundancy, drift, and mis-aggregation.
Figure 2: Overview of Meta-Moderator. The moderator monitors debate states and adjudicates a final answer when the debate stops; it is trained via outcome-driven policy optimization and used as a drop-in decision layer.
GSM8K
AMC
MATH
StratQA
MMLU
Meta-Moderator ∗
y∗∈Ht(↑)
432
21
180
386
385
y^=y∗…(↓)
33
8
25
42
77
Meta-Moderator
y∗∈Ht(↑)
431
24
187
369
351
y^=y∗…(↓)
12
1
8
9
15
Table 5: Analysis of debate coverage and adjudication.
Figure 3: Average debate rounds across benchmarks, comparing the untrained Meta-Moderator ∗ and the trained Meta-Moderator under the same maximum debate budget.
Appendix figures & tables1 asset
Supplementary material from the paper’s appendix.
Appendix
Figure 4: Case study comparing a failure of the untrained Meta-Moderator ∗ with a successful decision by the trained Meta-Moderator.
School of Cyber Science and Technology, University of Science and Technology of China, Hefei, China · School of Artificial Intelligence, Shenzhen Technology University, Shenzhen, China
School of Computer Science & Technology, Beijing Jiaotong University, Beijing, China · Beijing Key Laboratory of Traffic Data Mining and Embodied Intelligence, Beijing, China