{"id":"W4391621038","doi":"10.2139/ssrn.4713695","title":"Would a Two-benchmark Regime Be Better?","year":2024,"lang":"en","type":"article","venue":"SSRN Electronic Journal","topic":"European Union Policy and Governance","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Benchmark (surveying); Econometrics; Computer science; Mathematics; Statistics; Geology; Geodesy","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":[0.01662677,0.0006728965,0.002379783,0.00125176,0.005080163,0.01004016,0.002326917,0.01293858,0.04217755],"category_scores_gemma":[0.04160812,0.0003978681,0.001145316,0.00121968,0.0054185,0.009752981,0.003133151,0.01194298,0.004141753],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003416289,"about_ca_system_score_gemma":0.004627187,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0039252,"about_ca_topic_score_gemma":0.004620411,"domain_scores_codex":[0.9877973,0.004581254,0.0005722061,0.002126901,0.002257573,0.002664811],"domain_scores_gemma":[0.9868287,0.005491882,0.001512195,0.002383756,0.001902842,0.001880523],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0003479212,0.0002120063,0.0009832343,0.0001432563,0.00008446164,0.00009105697,0.0002456363,0.0008611359,0.0003507261,0.9426982,0.03072395,0.02325844],"study_design_scores_gemma":[0.000436786,0.0001652353,0.001795921,0.0001792598,0.00005238843,0.00007323688,0.0005715268,0.001931139,0.0004061344,0.9610802,0.03325307,0.00005505466],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.1012062,0.002027407,0.02761917,0.4553551,0.0102932,0.0003152757,0.001070166,0.0003926954,0.4017209],"genre_scores_gemma":[0.8683947,0.0005743387,0.01068896,0.06757816,0.002844874,0.0004220309,0.000217974,0.0001992292,0.04907969],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.04217755,"threshold_uncertainty_score":0.141098,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01575473948652081,"score_gpt":0.3063121208676065,"score_spread":0.2905573813810857,"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."}}