{"id":"W4283362807","doi":"10.31244/jero.2022.01.03","title":"Validity of automated learning progress assessment in English written expression for students with learning difficulties","year":2022,"lang":"en","type":"article","venue":"Journal for Educational Research Online","topic":"Reading and Literacy Development","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"Social Sciences and Humanities Research Council of Canada; Institute of Education Sciences; U.S. Department of Education","keywords":"Spelling; Curriculum; Narrative; Grammar; Writing assessment; Psychology; Quality (philosophy); Mathematics education; Applied psychology; Pedagogy; Linguistics","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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":[0.03427045,0.0005187802,0.0005766626,0.002457703,0.0004977726,0.001394635,0.001181034,0.0006325958,0.0007650043],"category_scores_gemma":[0.1224469,0.0003917119,0.0007364278,0.001099813,0.0008245584,0.001543252,0.002114952,0.0007086479,0.0005134402],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006845519,"about_ca_system_score_gemma":0.001068532,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00243384,"about_ca_topic_score_gemma":0.004143191,"domain_scores_codex":[0.9729677,0.01262833,0.003245434,0.00267514,0.007946587,0.0005368766],"domain_scores_gemma":[0.8750256,0.06305107,0.02605858,0.008762782,0.02469695,0.002404978],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006625193,0.0006740332,0.9083392,0.0001044252,0.0001445809,0.00005119928,0.002355056,0.0008754894,0.001259159,0.0001408004,0.0004597,0.08493391],"study_design_scores_gemma":[0.0001000751,0.002650938,0.9818643,0.0001194004,0.00005147238,0.0001684861,0.001246693,0.009457599,0.002302949,0.0002561256,0.001731346,0.0000506294],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9949664,0.0001265043,0.002457495,0.00006251853,0.00003947276,0.00027681,0.0001860881,0.00006851646,0.001816147],"genre_scores_gemma":[0.9945667,0.00004885943,0.004233629,0.00003807286,0.0000152586,0.0003705043,0.0002201672,0.00001987972,0.0004868777],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03427045,"threshold_uncertainty_score":0.1812416,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1084629845957615,"score_gpt":0.5313951478204307,"score_spread":0.4229321632246692,"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."}}