An agent benchmark compressed to 20 tasks predicts full scores with 14-28% less error than the best baselines.

DualViewEval notes that existing benchmark compression models redundancy only in task-model final-score distributions. It analyzes large-scale trajectories instead, identifies six complementary process signals associated with final performance, and jointly exploits outcome and process relations to learn a fixed-size miniset. Across five agent benchmarks and five baselines it wins on all datasets, reaching 24x to 40x compression on APEX-Agents and BFCL with 20 tasks, cutting mean absolute error 14.5-28.2% and improving Kendall's tau up to 7.2% relative to EssenceBench on SWE-bench Verified. Agent evals are expensive; this is how you run them nightly.