{"id":"W4417045320","doi":"10.31234/osf.io/rgf2u_v1","title":"Psychometric Measurement of Forecasters Using the Wisdom of Crowds","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":"Ground truth; Crowds; Event (particle physics); Aggregate (composite); Measure (data warehouse); Baseline (sea); Outcome (game theory)","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":[0.01587023,0.001107315,0.0009487764,0.003205585,0.0009931376,0.00319778,0.001118441,0.001377531,0.004213385],"category_scores_gemma":[0.1394918,0.0003318815,0.0008963725,0.002621162,0.001844646,0.003567618,0.002640926,0.001842172,0.000887347],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001175985,"about_ca_system_score_gemma":0.001157028,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00525918,"about_ca_topic_score_gemma":0.003535422,"domain_scores_codex":[0.9918109,0.004234714,0.0005026993,0.001352004,0.001867537,0.0002321663],"domain_scores_gemma":[0.9022053,0.06752747,0.01027773,0.01164467,0.006083338,0.002261423],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001276178,0.001006091,0.4998392,0.000697876,0.001309392,0.0003790991,0.01153395,0.1172292,0.002728882,0.04343242,0.007876889,0.3126909],"study_design_scores_gemma":[0.00026222,0.000970155,0.2563362,0.0003413438,0.0003299877,0.0002681319,0.005914339,0.5258106,0.003865856,0.1980242,0.007402025,0.0004750356],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7952645,0.0004777182,0.18088,0.001474364,0.0002328655,0.0004459109,0.001531836,0.0006601401,0.01903249],"genre_scores_gemma":[0.9798201,0.00009769088,0.01886841,0.00007906263,0.00006025842,0.0001798662,0.0003843749,0.00003400535,0.0004763544],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01587023,"threshold_uncertainty_score":0.08393079,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2997204130172303,"score_gpt":0.4336905674516686,"score_spread":0.1339701544344383,"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."}}