{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.004152472,0.0003247221,0.000387493,0.0001980262,0.0009871296,0.0002829338,0.001523136,0.0004983096,0.0001581799],"category_scores_gemma":[0.0004585433,0.000326255,0.000374925,0.0003212637,0.0002459538,0.0001793296,0.000432018,0.01215682,0.00004579869],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002258572,"about_ca_system_score_gemma":0.01276472,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005011169,"about_ca_topic_score_gemma":0.01432378,"domain_scores_codex":[0.9943117,0.0007619058,0.0004820446,0.0004711527,0.0007587784,0.003214447],"domain_scores_gemma":[0.9984241,0.0001317846,0.0006140055,0.0004627857,0.0001723982,0.0001949142],"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":[0.00002937291,0.00004550319,0.0003451088,0.00001650141,0.0001995547,0.00001473245,0.002235791,0.00004763132,0.000009207076,0.967388,0.008880073,0.02078848],"study_design_scores_gemma":[0.0003591494,0.00004460668,0.0002117088,0.0001807633,0.00005957775,0.00002386578,0.001064782,0.000007646643,0.00001605605,0.5334908,0.4642159,0.0003251586],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.05108727,0.01146952,0.003480357,0.06013912,0.003192726,0.0005480114,0.00009029763,0.0002084738,0.8697842],"genre_scores_gemma":[0.6748361,0.05375237,0.0002709127,0.008849716,0.005965974,0.00001981058,0.00002078135,0.00005692799,0.2562274],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.6237488,"threshold_uncertainty_score":0.9999189,"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."}}