Fault Localization

Fault localization aims to pinpoint where a failure originates, turning symptoms such as failing tests, agent errors, or abnormal sensor readings into actionable diagnoses. Software methods rank suspicious code using test coverage, program graphs, and learned or LLM-guided analysis, while physical-system methods use telemetry, signal patterns, and model residuals to locate faulty components. The work spans repository repair, AI agent traces, power grids, robotics, and industrial monitoring, with evaluations ranging from code-location accuracy to fault-zone identification under noise or changing conditions. A central challenge is that a plausible location may not establish causality or improve repair, and results can depend on localization granularity, available evidence, and distribution shifts.