LLM-agent-free Jev SRE diagnosis pipeline passes 76.2% of 105 diagnoses in 14.6s median
tianyin_xu · x · 2026-10-17
The SREGym team (Yiming Su et al., with Tianyin Xu) published a follow-up on Jev-driven SRE incident diagnosis:
- Idea: The previous post used Jev as a decision aid for an LLM agent; this one drops the agent entirely. A programmatic collector reads Kubernetes objects, events, recent pod logs, and resource usage, grouping observations by component (e.g., a Deployment) and summarizing failure signs.
- Pipeline: Jev picks the likeliest component to inspect → the collector gathers numbered evidence items → Jev judges whether the component is the origin, a downstream victim, or unrelated, and selects supporting observations → the pipeline assembles a diagnosis report.
- Results: Across 21 SREGym-Lite fault types, the pipeline passes 80 of 105 diagnoses (76.2%) with a median diagnosis time of 14.6 seconds.
- Framing: The authors argue much of incident debugging is intelligent graph traversal — following a breadcrumb trail with only local views and a fast branch explorer; unsuitable for GPS navigation, which needs the bigger picture. The post also analyzes where diagnoses went wrong.
Related event: SREGym's Jev Diagnostic Pipeline Passes 76.2% of Faults Without LLM Agents(2 posts)→
More from coding & agent
- Building a notes chatbot: why semantic search fails on dates and membership queries — bzarnal · 2026-10-17
- 14B judge no better than 8B: one prompt tweak lifted grading from 3/14 to 10/14 — Lost-Masterpiece2400 · 2026-10-17
- 10 open-source repos for training small models to do your agent's job — victor_explore · 2026-10-17
- Robotics engineer challenges coding agent hype: 'They can't do visuomotor control' — AlexTensor · 2026-10-17
- Notion AI lead: much of what runs on GPUs should run as CPU code — threepointone · 2026-10-17
- Dev uses Google Argon procedural generation scripts to build real 3D game assets — Promptmethus · 2026-10-17