{"id":"W4410720035","doi":"10.31234/osf.io/a7kdx_v1","title":"The Forecasting Proficiency Test: A General Use Assessment of Forecasting Ability","year":2024,"lang":"en","type":"preprint","venue":"","topic":"Forecasting Techniques and Applications","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"Open Philanthropy Project","keywords":"Test (biology); Computer science; Econometrics; Economics; Geology","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":[0.002047486,0.0006298934,0.000475169,0.001980839,0.0003334186,0.001163624,0.000578992,0.00100777,0.0058999],"category_scores_gemma":[0.01814691,0.0001570169,0.000579176,0.001052529,0.0005168279,0.001707914,0.0008300989,0.0009384513,0.002451422],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000349729,"about_ca_system_score_gemma":0.000537472,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002664139,"about_ca_topic_score_gemma":0.002571148,"domain_scores_codex":[0.9988849,0.0002302843,0.0001316018,0.0001833573,0.0004150798,0.0001548126],"domain_scores_gemma":[0.9906741,0.00376388,0.002323295,0.0008160808,0.001479536,0.0009431191],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0005150651,0.001269598,0.8097547,0.0001194457,0.0002180856,0.0002539957,0.001176183,0.003103901,0.00225165,0.002585592,0.02594933,0.1528023],"study_design_scores_gemma":[0.00005941662,0.0009346994,0.9732416,0.00005979851,0.00005224481,0.0004161724,0.0004596469,0.007758155,0.002433289,0.004414553,0.01010695,0.00006343986],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9564978,0.000315409,0.01033179,0.001009327,0.0001484646,0.0004344579,0.006517113,0.0004653751,0.02428022],"genre_scores_gemma":[0.9809701,0.0003014632,0.007131364,0.0002601309,0.00007353722,0.0005452627,0.006056178,0.00006791857,0.004593986],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0058999,"threshold_uncertainty_score":0.01973712,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3578384490180099,"score_gpt":0.4638868639794628,"score_spread":0.1060484149614528,"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."}}