{"id":"W2797014392","doi":"10.1177/0829573518765010","title":"Developing Proficiency in Standardized Cognitive Assessment Scoring: How Much Is Enough?","year":2018,"lang":"en","type":"article","venue":"Canadian Journal of School Psychology","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Psychology; Standardized test; Wechsler Intelligence Scale for Children; Wechsler Adult Intelligence Scale; Clinical psychology; Cognition; Wechsler Preschool and Primary Scale of Intelligence; Intelligence quotient; Medical education; Applied psychology; Mathematics education; Psychiatry; Medicine","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":[0.001864078,0.0001388885,0.0003047434,0.0006043536,0.0004839866,0.0001972229,0.0005001852,0.0001550866,0.001282074],"category_scores_gemma":[0.0006014453,0.0001372798,0.00007912709,0.00072175,0.0005704838,0.0003962816,0.00001631583,0.0004520912,0.00004860055],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007896223,"about_ca_system_score_gemma":0.006400033,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002296444,"about_ca_topic_score_gemma":0.06094644,"domain_scores_codex":[0.9979848,0.0002843792,0.0003968008,0.0002411855,0.0004174306,0.0006753833],"domain_scores_gemma":[0.9981021,0.0001052325,0.0002911607,0.0001151059,0.0007422212,0.0006441532],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00006624753,0.00004075784,0.9458689,0.000005733625,0.00008109767,0.0002126691,0.01514074,8.400416e-8,0.0001230557,0.009976985,0.02248483,0.005998885],"study_design_scores_gemma":[0.002756823,0.0005158911,0.8538849,0.0002430938,0.00003377582,0.00002462242,0.04668323,0.000001805044,0.0001015082,0.004347713,0.09106576,0.0003409179],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8712488,0.0003027262,0.002323115,0.04876441,0.002694938,0.0003631707,0.00001039587,0.00001007895,0.07428236],"genre_scores_gemma":[0.9935599,0.0001065132,0.002492862,0.002467144,0.0008687414,0.000006003179,9.274768e-7,0.00001264209,0.0004852412],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1223111,"threshold_uncertainty_score":0.9996309,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1039526991943376,"score_gpt":0.4554083521238081,"score_spread":0.3514556529294705,"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."}}