ahenrij/flakestorm

FlakeStorm是一个标记数据集，包含从公共开源项目中收集的真实不可靠（即间歇性）GitLab CI/CD作业失败记录。每条记录包含完整的原始作业日志以及两个级别的失败注释：粗粒度的GitLab原生`failure_reason`和细粒度的`category`标签，涵盖30种不同的失败类型（基础设施临时故障、依赖问题、不稳定测试、超时等）。该数据集旨在训练和评估从原始日志输出中自动诊断CI管道故障的模型，这是实现自我修复管道和智能警报的关键一步。

FlakeStorm is a labeled dataset of real unreliable (i.e., intermittent) GitLab CI/CD job failures collected from public open-source projects. Each record contains the full raw job log alongside two levels of failure annotation: a coarse GitLab-native `failure_reason` and a fine-grained `category` label covering 30 distinct failure types (infrastructure transients, dependency issues, flaky tests, timeouts, etc.). The dataset is designed for training and evaluating models that automatically diagnose CI pipeline failures from raw log output, a key step toward self-healing pipelines and intelligent alerting.