{"id":"W2063772878","doi":"10.1007/s11409-011-9074-8","title":"Statistical estimates of learners’ judgments about knowledge in calibration of achievement","year":2011,"lang":"en","type":"article","venue":"Metacognition and Learning","topic":"Innovative Teaching and Learning Methods","field":"Psychology","cited_by":24,"is_retracted":false,"has_abstract":false,"ca_institutions":"McGill University; Simon Fraser University","funders":"","keywords":"Calibration; Measure (data warehouse); Psychology; Statistic; Correctness; Metacognition; Mathematics education; Sample (material); Statistics; Computer science; Mathematics; Cognition; Data mining","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":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001996784,0.0001025339,0.0002424995,0.0002344668,0.00006822169,0.000005925005,0.00004586435,0.00007139396,0.001082291],"category_scores_gemma":[0.0003835376,0.00009659093,0.00002660061,0.0002012208,0.000114775,0.00006286246,0.00002373261,0.0004427741,0.00001154077],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000009158376,"about_ca_system_score_gemma":0.00001294612,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002203667,"about_ca_topic_score_gemma":0.00000521194,"domain_scores_codex":[0.997784,0.001410525,0.0003705124,0.0001883052,0.00009410629,0.0001525159],"domain_scores_gemma":[0.9993105,0.0003246642,0.000199908,0.00007195357,0.00006165373,0.00003130849],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004911646,0.001193272,0.5523043,0.000246253,0.0002279333,0.00001097948,0.1006312,0.00005729751,0.01513303,0.05167749,0.00007294599,0.277954],"study_design_scores_gemma":[0.001413283,0.0005869091,0.9823823,0.000153011,0.00007389515,0.00000383097,0.007453143,0.001832245,0.004034145,0.001317248,0.0005678487,0.0001821463],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8417931,0.0002054002,0.1341575,0.00001164539,0.0001083228,0.0000980783,0.000003615893,0.00002768033,0.02359468],"genre_scores_gemma":[0.9775364,0.000006406156,0.02188454,0.00002354769,0.00001456812,0.00001263501,0.00003242179,0.00001327716,0.0004762201],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4300779,"threshold_uncertainty_score":0.9998308,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1019453062351795,"score_gpt":0.3849430238929029,"score_spread":0.2829977176577234,"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."}}