PatchBoard: Schema-Grounded State Mutation for Reliable and Auditable LLM Multi-Agent Collaboration

2026-05-27

AI Summary

This work addresses the limitations of existing large language model–based multi-agent systems, which rely on natural language or loosely coupled shared memory, rendering collaboration processes difficult to verify, trace, and audit. The authors propose a schema-constrained transactional collaboration architecture that replaces natural language communication with structured JSON Patch operations to enable verifiable updates over a shared state. The framework enforces constraints through task schemas, role-based write permission contracts, and runtime invariants, with all state transitions jointly validated by a deterministic kernel. Evaluated on the ALFWorld benchmark, the approach achieves an 84.6% success rate with an average token consumption of 45.5k, substantially outperforming LangGraph (30.8%) and Flock (61.6%), thereby demonstrating efficient, reliable, and auditable multi-agent collaboration.

Abstract

LLM multi-agent systems often coordinate through natural-language dialogue or loosely structured shared memory, making intermediate state difficult to validate, attribute, and audit. We introduce PatchBoard, a schema-grounded collaboration architecture that replaces inter-agent dialogue with validated JSON Patch mutations over a shared structured state. An Architect agent constructs a task-specific schema and workflow rules, while a deterministic kernel validates each proposed state mutation against schema constraints, role-specific write contracts, and runtime invariants before committing it transactionally. On 630 matched ALFWorld episodes, PatchBoard achieves an 84.6% success rate, compared with 30.8% for LangGraph and 61.6% for Flock, while reducing tokens per successful task to 45.5k, compared with 368.3k and 64.2k, respectively.

Problem

• LLM multi-agent systems
• state validation
• auditability
• structured collaboration
• intermediate state

Innovation

• schema-grounded
• state mutation
• JSON Patch
• multi-agent collaboration
• auditable LLM systems