On 90 validated microservice outages, three simple LLM harnesses hit 65 to 77 percent PR@1 against 16 to 53 percent for five published root-cause-analysis agents
AspireAI community · arXiv (Tingzhu Bi, Xun Zhang, Jinglin Li et al.)
Root cause analysis of microservice failures attracts LLM agents built by hand, with standard operating procedures, fixed stages and role agents, and each is evaluated on its own data against few baselines, so it was unknown whether that design beats simply letting a capable model investigate on its own. The paper compares three simple harnesses that prescribe none of the model's calls (one call over an evidence package, a SQL agent, and the Codex coding agent) with five published LLM RCA agents and with statistical and causal methods, on RCAtlas, 90 validated cases from eight public datasets, across six models and three reasoning efforts. The simple harnesses are the most accurate, at PR@1 65-77% on three mid-tier models, ahead of the published agents at 16-53% and the non-LLM methods at 6-39%, on every model where both families were run. The published agents fix from a tenth to all of their model calls in the program, and a case study shows those prescriptions skipping the decisive evidence. A stronger model mainly improves reliability, a multi-turn harness mainly improves aptitude, and more reasoning effort changes accuracy by a few points at most while multiplying the time by up to seven. Among runs more accurate than the best non-LLM method, every Pareto-optimal run in cost or time uses a simple harness. The limitation: this is one RCA benchmark of 90 cases, and the result is about root cause analysis in microservice operations rather than a general claim about agent design.
Read the story and explore more from the AspireAI community.
Two sparse-autoencoder analyses have resisted scaling: matching semantically similar features across layers, and compressing large feature circuits into interpretable supernodes. This paper argues both are the same underlying problem, estimating semantic distances between SAE features that sit on different activation manifolds. Its representation treats each feature not as a single decoder vector but as an activation-weighted distribution over the hidden states expressing it, projects those distributions into a shared reference space, and compares them with Wasserstein distance. The representation is proved invariant to activation rescaling and stable under perturbation, and recovers true matches under finite-sample margin conditions. Empirically it outperformed decoder-vector and LLM-based baselines and captured subtle functional distinctions between related features, and it compressed large feature circuits into interpretable supernodes automatically. The limitation is that this is interpretability tooling: the paper demonstrates better feature matching and circuit compression, not that downstream interpretability claims become more accurate.