谁能否决智能体？LLM评审制度中的证据门控权限
原标题：Who May Overrule the Agent? Evidence-Gated Authority in LLM Review Institutions
AI 摘要
研究将多智能体系统中的评审机制视为制度管辖权问题：在提案、任务信息、模型和评审输出都相同的条件下，评审者何时有权让自己的建议成为最终决定。作者比较了两种制度：一种是广泛覆盖权，另一种是证据门控，即评审者必须给出机制上有效的约束违反记录和合规修复，才能替换原提案。在两种本地运行的开源权重模型、96个空间政策任务和96任务组合迁移基准上，证据门控大幅减少对正确提案的破坏，同时保留大部分违规修复；但前瞻性验证显示，Qwen 在显式保留、替换、升级接口下消除了正确提案破坏，Mistral 仍替换全部48个正确提案，说明接口设计会改变评审行为，却不是与模型无关的保障。
正文节选
Who May Overrule the Agent? Evidence-Gated Authority in LLM Review Institutions Abstract Multi-agent language-model systems often add a reviewer to improve an upstream agent’s decision. Yet a reviewer can both repair errors and replace decisions that were already correct. We study this as a problem of institutional jurisdiction: conditional on the same proposal, task information, model, and review output, when is the reviewer authorized to make its recommendation binding? We compare broad overri