The authors organized RCA tools into three abstraction levels: L1 retrieves raw metrics, L2 converts metrics into evidence using methods such as anomaly detection or causal analysis, and L3 directly produces diagnostic candidates.
Diagnostic tools speed agentic root cause analysis, but accuracy varies
Across 375 microservice failures, higher-level tools reduced diagnosis time, while their accuracy benefits depended on the task, model and system.
Chinese Tech
Yu Luo · Rongchen Gao · Zhenhui Zhou · Changchang Liu · Yuliang You · Yongqian Sun · +4 more
Nankai University · Huawei Technologies Co., Ltd. · Tsinghua University
Research Digest··2 min read
Luo and colleagues tested how different tool designs affect LLM agents diagnosing failures in microservice systems.
Why this paper
From Huawei Technologies Co., Ltd. and 2 others
In one line
Higher-level diagnostic tools reduce localization accuracy but improve failure type identification and speed in agentic RCA.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§