Roles with Rails: Contract-Preserving Role Evolution in Multi-Agent Structured Reasoning

2026-05-27

AI Summary

This work addresses the challenge that role evolution in multi-agent systems often disrupts structured contracts, leading to capability gaps, communication incompatibilities, or output protocol failures. It formally defines, for the first time, the problem of contract-preserving role evolution and introduces a credit-guided role-pool evolution framework. This framework enforces strict maintenance of capability coverage, communication compatibility, verification logic, aggregation mechanisms, and output protocols during role adaptation through terminal protection and conditional repair mechanisms. Integrating credit-guided retrieval, credit-ranked communication DAGs, a contextual bandit controller, and LLM-proposed edit validation, the approach demonstrates significant performance gains on three major large language models and multiple real-world reasoning benchmarks while preserving contractual integrity.

Abstract

Role-based LLM multi-agent systems need adaptive role pools, yet adapting such systems is not merely a matter of prompt optimization: roles often carry structural obligations, including capability coverage, message compatibility, validation, final-answer aggregation, and parser-compatible output protocols. Existing systems either fix the role inventory and lose adaptivity, or allow unconstrained generation to induce role drift, removing structurally necessary roles and breaking answer contracts. We formulate this as contract-preserving role evolution, requiring every committed edit to preserve five structural contracts (capability, communication, validation, aggregation, output protocol). We instantiate this formulation in SERO, a Self-Evolving Role Orchestration framework that evolves a typed role-card pool through credit-guided retrieval, a credit-ranked communication DAG with a protected terminal aggregator and conditional validator repair, and a contextual-bandit controller whose LLM-proposed edits are committed only when they preserve the contracts and improve task score. Experiments on real-world reasoning benchmarks across three LLM backbones confirm the value of contract-preserving role evolution.

Problem

• role evolution
• multi-agent systems
• contract preservation
• structured reasoning
• LLM orchestration

Innovation

• contract-preserving evolution
• multi-agent LLM systems
• role orchestration
• structural contracts
• self-evolving framework