📄 中文摘要
在先进的人工智能推理系统中，符号图网络是一种常见的架构模式，专门化的智能体或模块通过委托边连接，动态执行图中路由任务。现有调度器优化负载和适应性，但对几何结构缺乏建模，未能考虑故障在树状与循环结构中的传播差异。在树状委托中，单个故障可能导致指数级级联，而在密集循环图中，故障通常自我限制。识别这一可观察性差距并量化其系统级成本后，提出了一种轻量级的缓解方案。该研究制定了在线几何控制，以在时间索引执行图上进行路径风险评估，并结合路径局部故障历史进行优化。
级联感知的多智能体路由：时空侧车与几何切换
📄 English Summary
Cascade-Aware Multi-Agent Routing: Spatio-Temporal Sidecars and Geometry-Switching
A common architectural pattern in advanced AI reasoning systems is the symbolic graph network, where specialized agents or modules are connected by delegation edges to route tasks through a dynamic execution graph. Current schedulers optimize load and fitness but are geometry-blind, failing to model how failures propagate differently in tree-like versus cyclic regimes. In tree-like delegation, a single failure can cascade exponentially, while in dense cyclic graphs, failures tend to self-limit. This research identifies the observability gap, quantifies its system-level cost, and proposes a lightweight mitigation. It formulates online geometry control for route-risk estimation on time-indexed execution graphs, incorporating route-local failure history.