{"id":"W1825023912","doi":"","title":"Using Large-scale Assessments to Determine Student Grades","year":2013,"lang":"en","type":"article","venue":"Canadian Journal of Education / Revue canadienne de l éducation","topic":"Educational Assessment and Pedagogy","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Prince Edward Island","funders":"","keywords":"Scale (ratio); Tweaking; Psychology; Mathematics education; Medical education; Guideline; Medicine; Computer science; Geography; Cartography","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01039956,0.000875145,0.0006158225,0.005471423,0.002572573,0.002607119,0.001854205,0.0003335642,0.004229864],"category_scores_gemma":[0.03245004,0.0003145537,0.0004696787,0.004608778,0.001101652,0.001281351,0.001965969,0.0009873798,0.001694376],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.008211728,"about_ca_system_score_gemma":0.02241069,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.556817,"about_ca_topic_score_gemma":0.8665482,"domain_scores_codex":[0.9882951,0.001969541,0.001215237,0.001052754,0.006803764,0.0006636576],"domain_scores_gemma":[0.9482911,0.005950517,0.004440045,0.002998344,0.03554094,0.002778956],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0002915583,0.0006721091,0.4591591,0.0006411885,0.0001414189,0.0001404914,0.01361679,0.001792028,0.007134036,0.002698398,0.02543102,0.4882818],"study_design_scores_gemma":[0.00003989011,0.0003765171,0.948519,0.0003312641,0.00004967132,0.00007671412,0.00768773,0.002001905,0.005272231,0.001248353,0.03425385,0.0001428663],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7349129,0.001416051,0.05526186,0.001674133,0.0005609925,0.006384123,0.007815284,0.002635294,0.1893393],"genre_scores_gemma":[0.9267198,0.0009003889,0.04774381,0.0001968744,0.00004906544,0.001530071,0.003018877,0.0001611141,0.01967995],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9896004,"threshold_uncertainty_score":0.891586,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1356944071166592,"score_gpt":0.4296797860607543,"score_spread":0.2939853789440952,"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."}}