Multi-Pattern Multi-Model Benchmark Runs on Live Network Configuration Tasks

This dataset is a export of 350 benchmark run artifacts from a comparative study of LLM agent orchestration patterns on live network configuration tasks (FRRouting under Containerlab). Five tasks span RIP, OSPF, and BGP protocols at graded difficulty; five orchestration patterns (single-agent, sequential, debate, group chat, concurrent) are evaluated with four language models. The bundle is intended for **reproducibility, independent re-analysis, and citation** alongside the companion manuscript and related publications.