Agent Waste Index – what 341,054 public coding-agent runs wasted

Metermaid AI scored 341,054 public coding-agent trajectories from 11 public datasets, including SWE-smith, SWE-rebench, Open-SWE-Traces, mini-coder-trajs and agentic-coding-trajectories, for loops, blind retries, oversized tool output, abandoned runs and cost by run length, capping each dataset/config/split at the first 20,000 rows. The company published the results as the Agent Waste Index edition 1.1, with the pipeline, detectors and a `sweep` command that rebuilds the Index from the hub available in its GitHub audit repository. Metermaid AI said dataset publishers who find a number wrong for their data should open an issue, and that the method notes carry a corrections list.