{"id":"W2040697633","doi":"10.1007/s11092-011-9117-3","title":"Consistency of report card grades and external assessments in a Canadian province","year":2011,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":21,"is_retracted":false,"has_abstract":false,"ca_institutions":"Trent University; University of Toronto","funders":"Ministère de l’Éducation, Gouvernement de l’Ontario","keywords":"Report card; Psychology; Consistency (knowledge bases); Reading (process); Medical education; Grade level; Variance (accounting); Mathematics education; Medicine; Pedagogy; Mathematics; Political science","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true,"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":[0.009342493,0.0001706397,0.0002825066,0.0004062161,0.0001592485,0.0001403287,0.0002504874,0.00006430442,0.001696792],"category_scores_gemma":[0.001176474,0.0001457244,0.00004921446,0.0003820764,0.0002227257,0.0007954681,0.00007602631,0.0001438439,0.000005375001],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005645155,"about_ca_system_score_gemma":0.007392032,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.08166616,"about_ca_topic_score_gemma":0.09856468,"domain_scores_codex":[0.995932,0.0003557408,0.001071249,0.0006503976,0.00172735,0.0002632492],"domain_scores_gemma":[0.9970948,0.0006007866,0.000468155,0.0004224507,0.001174137,0.0002397363],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000009294571,0.0003574523,0.9734425,0.00001433155,0.00001147376,6.843753e-7,0.0005256787,0.000001388412,0.0001928558,0.01390222,0.0001823596,0.01135979],"study_design_scores_gemma":[0.0003273549,0.00006852171,0.898832,0.00002333757,0.0000225805,0.00001262335,0.002236972,0.0009575937,0.00006236765,0.09699287,0.0003325962,0.000131237],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9810081,0.0001813513,0.00008878965,0.00220075,0.0004796952,0.000840222,0.00002696683,0.000004825616,0.01516935],"genre_scores_gemma":[0.9936734,0.00001462146,0.005470767,0.0001742941,0.0000596811,0.0001849013,0.00004496698,0.000006483035,0.0003708813],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08309064,"threshold_uncertainty_score":0.9992158,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2160058372944961,"score_gpt":0.5025141811864667,"score_spread":0.2865083438919707,"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."}}