StateFuse: Deterministic Conflict-Preserving Memory for Multi-Agent Systems

2026-03-25 · Source: cs.CL updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

StateFuse is a conflict-aware replicated memory contract for multi-agent systems, built upon standard OpSet/CRDT merge principles. It introduces an agent-facing semantics layer featuring immutable history, explicit conflict objects, and dual correction handles: "claim_id" for exact edits and "claim_ref" for semantic, cross-replica targets. The system also incorporates deterministic predicate contracts and projection-time resolution that prevents rewriting replicated state. Evaluated on a 282-question MemoryAgentBench slice, StateFuse matched conflict-preserving baselines on answer accuracy (97.5%) but uniquely surfaced contradictions. In a controlled agent loop, it enabled safer abstention and correction compared to early collapse, demonstrating its value as a public memory contract for contradiction surfacing and auditable correction, rather than a universal accuracy enhancer.

Key takeaway

For AI Architects and Machine Learning Engineers designing memory layers for multi-agent systems, you should prioritize conflict-preserving memory contracts like StateFuse. This approach ensures explicit contradiction surfacing and auditable correction, which is materially safer than collapsing disagreements prematurely. By enabling conservative abstention and semantic correction, your systems can make more informed decisions and maintain a clear, verifiable history. This reduces the risk of hidden errors and unsafe actions.

Key insights

StateFuse offers a conflict-aware memory contract for multi-agent systems, prioritizing explicit contradiction surfacing and auditable correction over hidden overwrite.

Principles

• Maintain immutable history and explicit conflict surfacing.
• Use dual "claim_id" and "claim_ref" for corrections.
• Ensure projection-time resolution cannot mutate base memory.

Method

StateFuse stores immutable operations (Evidence, Claim, Retraction, Decision). Materialization transforms the op-set into a queryable state, grouping claims and emitting ConflictSet objects. Projection derives task-scoped views without altering base memory.

In practice

• Implement "claim_ref" for robust cross-replica corrections.
• Expose explicit conflict objects to agent decision policies.
• Adopt conservative abstention for ambiguous agent tasks.

Topics

• Agent Memory
• CRDT
• Conflict Resolution
• Multi-Agent Systems
• Replicated Data
• Semantic Correction

Code references

• nZiben/statefuse

Best for: Research Scientist, AI Engineer, AI Scientist, Machine Learning Engineer, AI Architect

Related on AIssential

• SubtleMemory: A Benchmark for Fine-Grained Relational Memory Discrimination in Long-Horizon AI Agents — AI agents struggle to discriminate subtle, context-dependent relations (complementary, nuanced, contradictory) among long-term memories.
• SuperLocalMemory: Privacy-Preserving Multi-Agent Memory with Bayesian Trust Defense Against Memory Poisoning — Local-first AI memory with Bayesian trust and adaptive re-ranking defends against poisoning and personalizes retrieval without cloud or LLMs.
• From Data Models to Mind Models: Designing AI Memory at Scale — Agentic memory enables AI systems to learn, adapt, and retain knowledge, crucial for complex, stateful operations.
• [AINews] The Claude Code Source Leak — The Claude Code leak reveals advanced agent architecture, emphasizing memory management, subagents, and robust operational features.
• PROJECTMEM: A Local-First, Event-Sourced Memory and Judgment Layer for AI Coding Agents — AI coding agents benefit from local, event-sourced memory and pre-action judgment to prevent repeated errors and improve efficiency.
• Privacy-Preserving RAG via Multi-Agent Semantic Rewriting: Achieving Confidentiality Without Compromising Contextual Fidelity — A multi-agent framework sanitizes RAG content via semantic rewriting, achieving privacy without compromising contextual fidelity.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CL updates on arXiv.org.