LatticeMind: A Conflict-Aware Memory Primitive for Multi-Agent Systems

2026-08-08

AI Summary

Existing multi-agent large language model systems lack a persistent mechanism for handling contradictory claims at write time, making it difficult to track the adoption status, disputes, and update rationale of statements. This work proposes a conflict-aware structured memory mechanism that detects contradictions during the write phase via lightweight symbolic checks and invokes the large language model for coordination only when semantic conflicts are irreconcilable—enabling write-time conflict resolution in multi-agent systems for the first time. By integrating symbolic verification with semantic coordination, the approach supports traceable and updatable memory management. On the source-agnostic ConflictBank benchmark, it achieves an accuracy of 0.97, substantially outperforming the strongest baseline (0.61, p<10⁻⁶); ablation studies confirm that removing either component causes a 12–14 percentage point performance drop.

Abstract

Multi-agent LLM systems often fail not for lack of candidate answers, but because they have no persistent mechanism for deciding which incompatible claim should currently be trusted. Majority vote, debate, and judge-based selection choose an output without recording which claim wins, which is contested, or why a later update supersedes it. We present \term{LatticeMind}, a conflict-aware structured memory that handles contradiction at write time. It maintains explicit item status, applies cheap symbolic conflict checks, and invokes LLM reconciliation only for unresolved semantic cases. On a label-blind ConflictBank evaluation that removes source-name hints, LatticeMind reaches 0.97 accuracy versus 0.61 for the strongest aggregation baseline, with the gap significant at $p<10^{-6}$ by paired McNemar test. Ablations show that removing the checker or the reconciler costs 12 to 14 points. On four secondary planning benchmarks the picture is mixed: LatticeMind beats naive merge on three of four, but does not replace deliberation methods on tasks rewarding iterative search.

Problem

• multi-agent systems
• conflict resolution
• structured memory
• contradiction handling
• trust decision

Innovation

• conflict-aware memory
• multi-agent LLM systems
• structured memory primitive
• symbolic conflict checking
• LLM reconciliation