{"id":"W4416559616","doi":"10.31234/osf.io/a7kdx_v8","title":"The Forecasting Proficiency Test: A General Use Assessment of Forecasting Ability","year":2025,"lang":"","type":"article","venue":"","topic":"Forecasting Techniques and Applications","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Probabilistic logic; Probabilistic forecasting; Variance (accounting); Variety (cybernetics); Bayesian probability; Test (biology)","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":[0.002006209,0.0006358187,0.0004872646,0.002044922,0.0003680864,0.001121866,0.0006366929,0.0009916357,0.00510249],"category_scores_gemma":[0.0159528,0.0001611248,0.000668537,0.001047735,0.0005044597,0.001778766,0.0008599377,0.0009923625,0.002123429],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003888738,"about_ca_system_score_gemma":0.0005925914,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003182004,"about_ca_topic_score_gemma":0.00328319,"domain_scores_codex":[0.9989021,0.0002204668,0.0001535207,0.0001732229,0.0003980603,0.0001526725],"domain_scores_gemma":[0.9910453,0.003224807,0.002593264,0.0006651231,0.001465574,0.001006029],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000417828,0.001187711,0.8621499,0.00009590926,0.0002072772,0.000245531,0.0008923003,0.002732163,0.001532865,0.001641008,0.02027542,0.1086221],"study_design_scores_gemma":[0.00006190347,0.001150032,0.9754652,0.00006435678,0.00005931375,0.0004917398,0.0004964589,0.008221917,0.002107489,0.003163811,0.008649872,0.00006799032],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.964912,0.0003063928,0.007756047,0.0009300699,0.0001311408,0.0004443598,0.005501143,0.000373314,0.0196456],"genre_scores_gemma":[0.9838574,0.0003267766,0.005826832,0.0002877316,0.00006494117,0.0005397212,0.005433985,0.00004950486,0.003613208],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.00510249,"threshold_uncertainty_score":0.01706952,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2132503898232042,"score_gpt":0.4378919946550607,"score_spread":0.2246416048318565,"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."}}