Reasoning Consensus: Structural Ensembling of LLM Reasoning via Weighted DAG Aggregation

2026-07-30

AI Summary

This work addresses the opacity of large language model (LLM) reasoning by proposing the first approach to ensemble not just final answers but the underlying reasoning structures themselves. The method constructs an interpretable consensus reasoning graph by extracting directed acyclic graphs (DAGs) from multiple chains of thought and fusing them with weights proportional to the number of independent verifications each component receives. This enables systematic analysis and validation of diverse reasoning pathways, including those discarded during standard inference. Experimental results across six benchmarks demonstrate consistent superiority over majority voting under equivalent computational budgets, with up to a 3.1% accuracy gain on MuSR-MM. Moreover, the consensus subgraph outperforms the majority answer path in 54.4%–65.4% of comparisons, highlighting its enhanced reliability and interpretability.

Abstract

Large Language Models (LLMs) explore problems through chain-of-thought, but this exploration is buried in unstructured prose. On high-stakes tasks, users cannot tell which steps are well-supported, which alternatives were seriously considered, or how the final conclusion compares to those the model discarded. We propose a framework that ensembles the reasoning structure, not just the answers, of multiple LLMs by weighted merging of Directed Acyclic Graphs (DAGs) extracted from reasoning chains. We weight each step by how many traces independently attest to it, to return "Consensus Reasoning". Across six benchmarks spanning statutory interpretation, graduate-level science, narrative multi-hop reasoning, and first-order logic, our ensemble outperforms a matched-budget majority-vote baseline, with a maximum accuracy gain of 3.1% on MuSR-MM (narrative multi-hop reasoning). On a single model, the framework matches or exceeds self-consistency at the same trace budget while additionally exposing an inspectable consensus reasoning graph. Ensemble weights correlate with LLM-judge rankings of reasoning quality at Spearman $ρ= 0.30$-$0.51$, and consensus subgraphs are preferred over alternatives leading to the majority-vote answer in 54.4-65.4% of head-to-head comparisons across five of six datasets. We observe that our framework can also be used to analyze diverse reasoning perspectives for a problem.

Problem

• reasoning transparency
• structured reasoning
• LLM interpretability
• reasoning consensus
• chain-of-thought analysis

Innovation

• Reasoning Consensus
• Weighted DAG Aggregation
• Structural Ensembling
• Chain-of-Thought
• Interpretable Reasoning