预测智能体何时该推理？可靠性路由的行为压力测试
原标题：When Should Forecasting Agents Reason? Behavioral Stress Tests for Reliability Routing
AI 摘要
该研究将预测智能体的机制选择（检索、推理、参考市场先验或使用历史类比）视为可观测行为，在 ForecastBench 式二元预测任务上进行压力测试。作者提出 ReliabilityRoute，基于历史覆盖率、市场先验可用性、证据强度等可靠性特征进行路由，而非硬编码来源名称。实验显示固定路由在 2025-10-26 评估集上将 Brier 分数从 0.1876 降至 0.1846，而 walk-forward 自适应规则在 16 个 LLM 版本上平均 Brier 从 0.1867 降至 0.1839。核心结论是更多推理并不总是更好，预测智能体应首先判断哪种证据来源值得控制。
正文节选
When Should Forecasting Agents Reason? Behavioral Stress Tests for Reliability Routing Abstract Forecasting agents increasingly combine language-model reasoning, retrieval, ensembling, and calibration, but it remains unclear when each behavior should be trusted. We study this question on ForecastBench-style binary forecasting tasks, treating the choice to retrieve, reason, defer to a market prior, or use a historical analog as an observable agent behavior rather than a hidden implementation deta