复合AI系统可靠性：150起生产事故的故障分类与韧性模式目录
原标题：Compound AI System Reliability: A Failure Taxonomy and Resilience Pattern Catalog from 150 Production Incidents
AI 摘要
该研究分析了来自开源复合AI项目及匿名企业部署的150份生产事故报告，构建了涵盖检索、生成、工具、编排和集成五类共23种故障模式的分类体系。针对每类故障，作者提出韧性模式并通过受控故障注入实验测量效果：断路器减少89%级联传播，输出质量门在用户受影响前捕获73%的静默退化，组件隔离减少64%爆炸半径。采用三种以上韧性模式的系统相比无结构监控基线将平均恢复时间（MTTR）降低71%。
正文节选
Compound AI System Reliability: A Failure Taxonomy and Resilience Pattern Catalog from 150 Production Incidents Abstract Deploying compound AI systems reliably and safely requires understanding failure modes that emerge at component boundaries, not within individual models. Cascading errors propagate across component boundaries, silent quality degradation evades standard monitoring, and coordination failures yield incorrect collective behavior from individually correct parts. We analyze 150 prod