{"id":"W2113150213","doi":"10.55016/ojs/ajer.v54i3.55237","title":"Alignment of Scores on Large-Scale Assessments and Report-Card Grades","year":2008,"lang":"en","type":"article","venue":"Alberta Journal of Educational Research","topic":"Educational Assessment and Pedagogy","field":"Social Sciences","cited_by":15,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Report card; Psychology; Scale (ratio); Rating scale; Mathematics education; Statistics; Clinical psychology; Applied psychology; Developmental psychology; Pedagogy; Geography; Mathematics; Cartography","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01784433,0.000462784,0.0005956298,0.002627639,0.0005380095,0.001712744,0.0008196212,0.0004127286,0.001475738],"category_scores_gemma":[0.09380451,0.0003488451,0.0003650714,0.002088106,0.0006934659,0.0008821578,0.002313901,0.0008191972,0.0006022032],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008454747,"about_ca_system_score_gemma":0.0008234273,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003568147,"about_ca_topic_score_gemma":0.01063434,"domain_scores_codex":[0.9810947,0.008180538,0.002182667,0.00299922,0.004935369,0.00060755],"domain_scores_gemma":[0.8916075,0.04867319,0.02317473,0.01007352,0.02309841,0.003372728],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004664312,0.0001609746,0.9540253,0.00008645967,0.0001605844,0.00003120942,0.002964533,0.0005309,0.001250794,0.0002500143,0.0007345302,0.0393382],"study_design_scores_gemma":[0.00001168953,0.0002837869,0.9967704,0.00001399846,0.00001883192,0.00002346099,0.0005994265,0.0005627883,0.0009952211,0.0001280545,0.000575458,0.00001697083],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9927301,0.0001057129,0.002818056,0.0000613277,0.00003850819,0.0001212869,0.0006443902,0.0001104067,0.003370274],"genre_scores_gemma":[0.9969818,0.00003177847,0.001725523,0.00001542097,0.00001036736,0.00009487824,0.000580314,0.00001974486,0.0005401404],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01784433,"threshold_uncertainty_score":0.09437096,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2037117002265996,"score_gpt":0.5313738724859143,"score_spread":0.3276621722593147,"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."}}