📄 中文摘要
混合推理系统结合了学习组件与基于模型的推理，越来越多地应用于工具增强的决策循环中。然而，在部分可观测性和持续证据不匹配的情况下，其运行行为仍然缺乏深入理解。实际中，故障往往表现为内部推理动态的逐渐偏离，而非孤立的预测错误。该研究从卡尔曼思想的角度研究混合推理系统的运行稳定性，将推理建模为由内部创新信号驱动的随机推理过程，并引入可测量的运行现象——认知漂移。稳定性通过可检测性、有限偏离和可恢复性来定义，而非任务级别的表现。
基于卡尔曼思想的混合推理系统的运行稳定性与恢复
📄 English Summary
Kalman-Inspired Runtime Stability and Recovery in Hybrid Reasoning Systems
Hybrid reasoning systems that integrate learned components with model-based inference are increasingly utilized in tool-augmented decision loops. However, their runtime behavior under partial observability and sustained evidence mismatch remains inadequately understood. Failures often manifest as gradual divergence in internal reasoning dynamics rather than isolated prediction errors. This study examines runtime stability in hybrid reasoning systems from a Kalman-inspired perspective, modeling reasoning as a stochastic inference process driven by an internal innovation signal. Cognitive drift is introduced as a measurable runtime phenomenon. Stability is defined in terms of detectability, bounded divergence, and recoverability, rather than task-level performance.