How Benchmarks Mis-Score Computer-Use Agents
For researchers and developers of computer-use agents, this paper provides a framework and evidence to improve benchmark reliability, which is crucial for accurate evaluation and progress in the field.
The paper identifies and categorizes reliability issues in computer-use agent benchmarks, auditing 150 failure-scored trajectories and finding that 15.3% of FAIL verdicts are incorrect (10.7% evaluator false negatives, 4.7% broken tasks), with verification/feedback and planning failures dominating genuine failures.
Computer-use agents (CUA) are being deployed to browse the web and operate desktop software, yet their benchmark scores are still commonly produced by brittle scripted oracles. A score is the output of a pipeline in which tasks can be stale, trajectories can omit decisive visual evidence, evaluators can reject valid alternatives, and aggregate reports can hide the cause of failure. We organize these problems into a reliability framework spanning task construction, trajectory observation, scoring, and reporting. We then audit 150 public failure-scored trajectories from five web, enterprise-workflow, and desktop-control benchmarks, find that 15.3\% of FAIL verdicts are wrong: 10.7\% are evaluator false negatives and 4.7\% are broken tasks. For genuine failures, a three-tier diagnostic taxonomy shows that verification/feedback and planning failures dominate execution/grounding errors, while a single scalar success rate can not explain. We connect these findings to newer long-horizon CUA benchmarks and derive stage-specific design rules for CUA evaluation.