RUC-AIBOX/AEWM-Action-Judge-Bench

AEWM Action Judge Benchmark Can a model distinguish productive agent decisions from unproductive ones before execution? The AEWM Action Judge Benchmark evaluates this capability on 3,000 annotated decisions across Search, Terminal, and Software Engineering (SWE). Given a task, its observed interaction history, and a proposed reasoning-action pair, a model predicts whether the decision is Critical, Exploratory, or Noisy. The benchmark tests decision-level judgment in long-horizon… See the full description on the dataset page: https://huggingface.co/datasets/RUC-AIBOX/AEWM-Action-Judge-Bench.

<iframe src="https://silytics.si/embed/item/63174e24-6a68-4241-b498-839e265d0aed" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 18 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.