{"id":"W4414856460","doi":"10.1109/tsmc.2025.3614831","title":"Role Assignment for Agent Evaluation Under Uncertainty: A Distributionally Robust Approach","year":2025,"lang":"en","type":"article","venue":"IEEE Transactions on Systems Man and Cybernetics Systems","topic":"Auction Theory and Applications","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Nipissing University","funders":"National Natural Science Foundation of China","keywords":"Ambiguity; Robustness (evolution); Robust optimization; Set (abstract data type); Quality (philosophy); Volatility (finance); Convex optimization","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01110054,0.001758575,0.00276948,0.001569023,0.0009816188,0.003789659,0.004197155,0.002591124,0.004315817],"category_scores_gemma":[0.03636285,0.00100448,0.001721638,0.001712883,0.002705766,0.006409688,0.00534036,0.003867388,0.0007286894],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003207927,"about_ca_system_score_gemma":0.002502971,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003419241,"about_ca_topic_score_gemma":0.002244646,"domain_scores_codex":[0.9891329,0.005954909,0.0004412152,0.001814109,0.001938061,0.0007187795],"domain_scores_gemma":[0.9820099,0.01160695,0.00226439,0.001793342,0.001535525,0.0007897615],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001767114,0.0001330272,0.001289552,0.0001311584,0.00008575843,0.0001397559,0.0003011567,0.7854072,0.001017717,0.1592284,0.001436412,0.05065307],"study_design_scores_gemma":[0.00001226391,0.00004296255,0.0001170712,0.0000129838,0.00001092114,0.00002855966,0.00003824125,0.9404958,0.0002820598,0.05840589,0.0005398174,0.00001338841],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.008749591,0.0002065546,0.9887809,0.0003463089,0.00002454261,0.00006691062,0.00005594197,0.0001307925,0.001638477],"genre_scores_gemma":[0.7085258,0.0004714973,0.2859758,0.0003004643,0.0001226361,0.0003195686,0.000257002,0.0001805168,0.003846686],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01110054,"threshold_uncertainty_score":0.05870599,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09026096889425071,"score_gpt":0.3446426849967998,"score_spread":0.2543817161025491,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}