{"id":"W4306631416","doi":"10.53967/cje-rce.v45i3.5191","title":"Assessing Learning Skills and Work Habits: What Do Report Card Data Tell Us?","year":2022,"lang":"en","type":"article","venue":"Canadian Journal of Education / Revue canadienne de l éducation","topic":"School Choice and Performance","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"","keywords":"Report card; Academic achievement; Mathematics education; Standardized test; Psychology; Accountability; Set (abstract data type); Construct (python library); Achievement test; Test (biology); Work (physics); Medical education; Pedagogy; Medicine; Computer science","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.0035823,0.00009953911,0.0001613874,0.0004293926,0.001520335,0.0007990507,0.0005161216,0.00006326495,0.0003674242],"category_scores_gemma":[0.002365302,0.0001251951,0.00003618487,0.0007244995,0.00009351132,0.002527955,0.00002582633,0.0005209293,0.000003319989],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003278827,"about_ca_system_score_gemma":0.01961192,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.07001907,"about_ca_topic_score_gemma":0.5545958,"domain_scores_codex":[0.99838,0.0003551989,0.0003972204,0.0002416827,0.0001522472,0.0004736263],"domain_scores_gemma":[0.9974159,0.0001492143,0.0005496352,0.0003533714,0.0004259204,0.001105954],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00000350537,0.00004404222,0.7621729,0.00003054017,0.00002904799,0.00006825606,0.1394711,0.0004221387,0.00001507252,0.0004214714,0.02999715,0.06732474],"study_design_scores_gemma":[0.00006802475,0.0000289344,0.1549366,0.0001928393,0.00004160978,0.0005283532,0.1714365,0.00001102326,0.000002699987,0.0002552979,0.6723421,0.0001560488],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9808306,0.003012255,0.00001623191,0.008085594,0.006486351,0.0001143628,0.000005684555,0.000005989702,0.001442976],"genre_scores_gemma":[0.9939803,0.0004522369,0.0003360496,0.0006316938,0.001778859,0.00001245032,0.00009745416,0.00001696725,0.002693998],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6423449,"threshold_uncertainty_score":0.9997795,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06164262493477819,"score_gpt":0.3397326991640287,"score_spread":0.2780900742292505,"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."}}