{"id":"W4411270008","doi":"10.2139/ssrn.5293276","title":"Would a Two-Benchmark Regime Be Better?","year":2025,"lang":"en","type":"preprint","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); Computer science; Econometrics; Economics; 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":[0.01520972,0.0007022066,0.002509736,0.001227331,0.004655159,0.009068037,0.002115383,0.01202937,0.04250275],"category_scores_gemma":[0.03904603,0.0004039842,0.001114108,0.001192543,0.005284586,0.01011964,0.003189459,0.0106298,0.004124928],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003090826,"about_ca_system_score_gemma":0.003702917,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002899829,"about_ca_topic_score_gemma":0.003375441,"domain_scores_codex":[0.9890197,0.004327114,0.0005049599,0.001990192,0.00188564,0.002272279],"domain_scores_gemma":[0.9873389,0.00531893,0.001491852,0.002509026,0.001654132,0.001687177],"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":[0.0003553418,0.0001882118,0.0008110465,0.000141843,0.00007750071,0.00008282102,0.0002065561,0.0008872676,0.0003299595,0.9509599,0.02459952,0.02135998],"study_design_scores_gemma":[0.0003420637,0.0001122262,0.001187059,0.0001235805,0.00003859743,0.00005526997,0.0003632197,0.001763258,0.0003236529,0.9743334,0.02132029,0.0000374684],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"other","genre_gemma":"commentary","genre_scores_codex":[0.122575,0.002205377,0.03582897,0.3927267,0.009621697,0.0003323972,0.001239614,0.0004241652,0.4350461],"genre_scores_gemma":[0.8919848,0.0005629021,0.01090613,0.04750171,0.002572495,0.0003881796,0.0002097527,0.0001949984,0.04567902],"genre_candidate":"commentary","genre_consensus":null,"teacher_disagreement_score":0.04250275,"threshold_uncertainty_score":0.1421858,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01984496260887892,"score_gpt":0.3189799638165419,"score_spread":0.299135001207663,"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."}}