{"id":"W2956977098","doi":"10.1007/s11409-019-09198-4","title":"Does domain matter? Monitoring accuracy across domains","year":2019,"lang":"en","type":"article","venue":"Metacognition and Learning","topic":"Decision-Making and Behavioral Economics","field":"Decision Sciences","cited_by":36,"is_retracted":false,"has_abstract":false,"ca_institutions":"Ontario Centre of Excellence for Child and Youth Mental Health; York University","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Overconfidence effect; Task (project management); Calibration; Psychology; Generality; Resolution (logic); Cognitive psychology; Cognition; Domain (mathematical analysis); Artificial intelligence; Metacognition; Domain knowledge; Computer science; Machine learning; Developmental psychology; Statistics; Social psychology; Mathematics","routes":{"ca_aff":true,"ca_fund":true,"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":[0.004998206,0.0004520887,0.0007959456,0.001015403,0.0002878135,0.002678965,0.0007215232,0.001505228,0.002980936],"category_scores_gemma":[0.06481548,0.0003558273,0.0005146249,0.001037161,0.000880775,0.004166236,0.001225532,0.001280873,0.0005728634],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003513384,"about_ca_system_score_gemma":0.0003036043,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001967073,"about_ca_topic_score_gemma":0.001628236,"domain_scores_codex":[0.9981002,0.0006405751,0.0001825883,0.0005701864,0.0002900859,0.0002162747],"domain_scores_gemma":[0.9226865,0.05017784,0.01380275,0.009194714,0.002497807,0.001640409],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002905715,0.000419879,0.9279625,0.0001616798,0.0006617226,0.000104751,0.0008159297,0.005085874,0.005582414,0.001907553,0.0006029583,0.05378892],"study_design_scores_gemma":[0.00005983751,0.0002911153,0.9748744,0.00005719404,0.0002300732,0.000203802,0.0003857854,0.009743501,0.002517353,0.01087584,0.0007213464,0.00003966713],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9946964,0.000390697,0.002016193,0.0002903907,0.00002243572,0.000008620902,0.0002230311,0.00002668511,0.002325539],"genre_scores_gemma":[0.9990643,0.00007497216,0.0004043498,0.00005323887,0.00001578331,0.000004604323,0.0001511198,0.00001450087,0.0002171509],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.004998206,"threshold_uncertainty_score":0.02643335,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0662592283616116,"score_gpt":0.3990462324451357,"score_spread":0.332787004083524,"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."}}