{"id":"W6962501024","doi":"10.17605/osf.io/39a2m","title":"Accurate Versus Inaccurate Labelers (July 2017)","year":2017,"lang":"en","type":"other","venue":"OSF Preprints (OSF Preprints)","topic":"Diverse Scientific and Economic Studies","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Noise (video); Reliability (semiconductor); Feature (linguistics); Key (lock); Pattern recognition (psychology)","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.006225701,0.0009844122,0.0009024677,0.005602176,0.002595233,0.01285869,0.001202826,0.002439023,0.211515],"category_scores_gemma":[0.03725682,0.0005238291,0.0005101261,0.003967175,0.001542,0.005112839,0.003167889,0.002937533,0.2011594],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00375031,"about_ca_system_score_gemma":0.004866471,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01199037,"about_ca_topic_score_gemma":0.02342953,"domain_scores_codex":[0.9952098,0.0008498774,0.000220398,0.0005169782,0.00273619,0.0004668043],"domain_scores_gemma":[0.9749136,0.005511974,0.001410115,0.004224438,0.0122479,0.001691947],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00009527265,0.00003470562,0.0005307024,0.0000502846,0.000007673196,0.0000188803,0.0000746185,0.0001435844,0.00009180018,0.006670245,0.9564218,0.03586048],"study_design_scores_gemma":[0.00004163356,0.00002074738,0.003959392,0.0002677772,0.00001856507,0.0000741464,0.0004525997,0.001962992,0.001745309,0.03273444,0.9586734,0.00004895704],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.01357551,0.005701919,0.02233672,0.04459263,0.03326501,0.0003484225,0.1475277,0.01595884,0.7166932],"genre_scores_gemma":[0.05582825,0.002652191,0.01406121,0.004502556,0.006699757,0.0003223331,0.09544823,0.01067666,0.8098089],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9937743,"threshold_uncertainty_score":0.7075883,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06390379466558854,"score_gpt":0.2641865076702731,"score_spread":0.2002827130046846,"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."}}