{"id":"W3125406353","doi":"","title":"Testing Multiple Forecasters","year":2007,"lang":"en","type":"preprint","venue":"RePEc: Research Papers in Economics","topic":"Forecasting Techniques and Applications","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Test (biology); Calibration; Contrast (vision); Computer science; Set (abstract data type); Machine learning; Process (computing); Artificial intelligence; Data mining; Econometrics; Statistics; Mathematics","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":[0.03207367,0.001722762,0.002155806,0.001750279,0.001036951,0.003756479,0.002552242,0.003620429,0.02730045],"category_scores_gemma":[0.2103774,0.0007256479,0.002298027,0.001527743,0.0014001,0.00612591,0.002127825,0.003299344,0.004172633],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009871389,"about_ca_system_score_gemma":0.00180593,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002234089,"about_ca_topic_score_gemma":0.001625383,"domain_scores_codex":[0.9764436,0.01201445,0.001317112,0.006512479,0.002562223,0.001150125],"domain_scores_gemma":[0.7065988,0.2523665,0.01061084,0.0214461,0.006669389,0.00230836],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.007693682,0.001506865,0.3686742,0.0007002868,0.006007002,0.001756724,0.001968534,0.1118269,0.003509543,0.05151295,0.02125487,0.4235884],"study_design_scores_gemma":[0.001253405,0.003686286,0.07963182,0.0003619496,0.001812918,0.001221716,0.002025776,0.7559922,0.008467247,0.1342836,0.01097895,0.0002840105],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.7898602,0.001335472,0.175357,0.006467174,0.002446044,0.0003882945,0.002626857,0.001015908,0.02050316],"genre_scores_gemma":[0.9793479,0.0001655258,0.01524618,0.0003614391,0.0003434109,0.0001563364,0.001557461,0.00007702927,0.002744738],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.03207367,"threshold_uncertainty_score":0.1696238,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2973178101706639,"score_gpt":0.4520874961241753,"score_spread":0.1547696859535113,"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."}}