{"id":"W2951521963","doi":"10.2139/ssrn.2953963","title":"Evaluating Strategic Forecasters","year":2017,"lang":"en","type":"article","venue":"SSRN Electronic Journal","topic":"Auction Theory and Applications","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Business","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":["sts"],"consensus_categories":[],"category_scores_codex":[0.01196745,0.00008975391,0.0001293553,0.00009133297,0.002071779,0.0008225854,0.001307304,0.00004348002,0.0003064044],"category_scores_gemma":[0.0009826996,0.00006412228,0.0001204653,0.0001020581,0.0001215847,0.0005376478,0.00007148572,0.000905026,0.0005467599],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001788801,"about_ca_system_score_gemma":0.001044532,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000008864984,"about_ca_topic_score_gemma":0.0001448491,"domain_scores_codex":[0.9974635,0.0001533061,0.0003814447,0.0002313696,0.0006969489,0.001073381],"domain_scores_gemma":[0.9982275,0.0001893736,0.000638456,0.0006514874,0.0002073108,0.00008585832],"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":[0.00002013968,0.00001511155,0.0005895394,1.97799e-7,0.00002056667,6.202782e-7,0.00006647556,0.0001420365,0.0004257152,0.8196377,0.0000697529,0.1790121],"study_design_scores_gemma":[0.0002686347,0.0001208825,0.0005950783,0.000003272281,0.000009938748,0.0002501691,0.0040951,0.00127924,0.00008281445,0.9921793,0.001036876,0.0000787043],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9272103,0.0002315682,0.04002774,0.003270609,0.0003892532,0.0001068752,0.000001717715,0.00002001112,0.02874188],"genre_scores_gemma":[0.9878826,0.00008504592,0.0002347256,0.00005501902,0.0003241006,0.000005101975,3.591564e-7,0.000007687902,0.01140543],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1789334,"threshold_uncertainty_score":0.9992274,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3029690398647003,"score_gpt":0.4914819184839186,"score_spread":0.1885128786192183,"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."}}