{"id":"W6926610684","doi":"10.25318/3710023001-eng","title":"Mean scores of Grade 8 students on the Pan-Canadian Assessment Program mathematics subdomains assessments","year":2022,"lang":"en","type":"dataset","venue":"Statistics Canada Dissemination","topic":"Computational Physics and Python Applications","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Online assessment; Program evaluation; Standards-based assessment; Statistical analysis","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004032852,0.0002728616,0.0002640267,0.0001864802,0.0004506946,0.0002586883,0.001907955,0.00005425222,0.0004271853],"category_scores_gemma":[0.00006894057,0.0002437622,0.0000447987,0.0005944037,0.00005061709,0.00006506604,0.0002686479,0.0003860387,0.000002747478],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008772647,"about_ca_system_score_gemma":0.002351965,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.3352529,"about_ca_topic_score_gemma":0.8449491,"domain_scores_codex":[0.9966651,0.0001316133,0.0004917695,0.0004013372,0.002016588,0.0002935793],"domain_scores_gemma":[0.9969596,0.001140892,0.0005645349,0.0009125713,0.0002596634,0.0001627709],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[5.08242e-7,0.0003027323,0.00001841818,0.00005500977,0.0000548826,0.00001019647,0.00004104671,0.0000746737,0.000001433317,0.1808238,0.8134649,0.005152436],"study_design_scores_gemma":[0.0002655501,0.0003858187,0.03460137,0.0001697606,0.0001622654,0.000005910347,0.0002746944,0.0147155,0.00002258294,0.04277578,0.9057744,0.0008464319],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"dataset","genre_gemma":"dataset","genre_scores_codex":[0.0001132247,0.000008922624,0.02935679,0.0004465436,0.0003735196,0.00107633,0.9684857,0.00001833686,0.0001206524],"genre_scores_gemma":[0.007083406,0.00002326127,0.0193294,0.0001973061,0.00003366651,0.0005642232,0.9726636,0.00002128386,0.00008383417],"genre_candidate":"dataset","genre_consensus":"dataset","teacher_disagreement_score":0.5096962,"threshold_uncertainty_score":0.9940334,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01772105602771094,"score_gpt":0.3492534200867212,"score_spread":0.3315323640590102,"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."}}