{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003476195,0.0002549199,0.0004009789,0.0003016482,0.0006958469,0.000557626,0.000353817,0.0001848338,0.00003465197],"category_scores_gemma":[0.00004464394,0.0002096083,0.0001690187,0.0006012896,0.0001302079,0.0001214658,0.000004548645,0.0001654217,0.00005831496],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003340071,"about_ca_system_score_gemma":0.0001605268,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001026606,"about_ca_topic_score_gemma":0.00001910696,"domain_scores_codex":[0.9962927,0.0005544985,0.0009578629,0.0007407141,0.001166845,0.0002873816],"domain_scores_gemma":[0.9974833,0.0007763366,0.0003121795,0.0006506068,0.0006286112,0.0001490236],"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":[0.00005744179,0.0002293676,0.00001224767,0.00004690404,0.0001114249,1.078794e-7,0.0001475837,0.875232,0.0002974815,0.1161844,0.00214558,0.005535451],"study_design_scores_gemma":[0.001450637,0.0001362874,0.0002359596,0.0001509012,0.0002571089,0.00002527262,0.01311396,0.927669,0.0002488542,0.01187378,0.0444535,0.0003847035],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.006944654,0.0005989338,0.9825253,0.0003314232,0.001454046,0.002512916,0.000328601,0.00007403415,0.005230103],"genre_scores_gemma":[0.9758033,0.00002001945,0.0001442692,0.00007051224,0.0001046512,0.002354952,0.00003886167,0.00001608917,0.02144735],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.982381,"threshold_uncertainty_score":0.854758,"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."}}