Reproduce “Free to Look, Free to Reason: Simple LLM Harnesses Outdiagnose Hand-Crafted Root Cause Analysis Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tingzhu Bi, Xun Zhang, Jinglin Li, Ping Wang, Meng Ma
The authors compare simple model-directed harnesses with prescribed agents for microservice root-cause analysis. Across 90 cases from eight datasets, these harnesses let models inspect prepared evidence, query telemetry with SQL, or write analysis code. GPT-6 Luna’s Codex harness reaches 68.9% PR@1 over three passes, versus FoA’s 54.4% in one reported run; this does not establish that prescriptions caused the gap.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.