{"id":"W4200548621","doi":"10.31219/osf.io/yh3zj","title":"Validity of Automated Learning Progress Assessment in English Written Expression for Students with Learning Difficulties","year":2021,"lang":"en","type":"preprint","venue":"","topic":"Writing and Handwriting Education","field":"Social Sciences","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; Narrative; Curriculum; Grammar; Writing assessment; Quality (philosophy); Mathematics education; Psychology; Expression (computer science); Applied psychology; Computer science; 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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002263744,0.0001933902,0.0004214467,0.0001335893,0.0004409939,0.0004391839,0.0003317999,0.0002823914,0.00003108465],"category_scores_gemma":[0.0009987928,0.0001739098,0.0000884687,0.0001889287,0.0001479647,0.0001285823,0.0002713932,0.0008062089,2.403712e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002175057,"about_ca_system_score_gemma":0.0006463611,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001025222,"about_ca_topic_score_gemma":0.0008270887,"domain_scores_codex":[0.9971188,0.0008836453,0.0004098095,0.0004707843,0.0007723841,0.0003445382],"domain_scores_gemma":[0.998119,0.0004514233,0.0004462977,0.0001515492,0.0007593097,0.00007242906],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.0000259381,0.0006232967,0.9073811,0.0006617073,0.00003857508,0.000003649123,0.08515935,0.004410646,0.000196446,0.000118763,0.0001978283,0.00118269],"study_design_scores_gemma":[0.001354926,0.0002703375,0.4200173,0.006131044,0.00008670937,4.998134e-7,0.5655957,0.003311046,0.001488241,0.00005087921,0.001032066,0.0006612753],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9954234,0.00006475855,0.0003082291,0.000158472,0.0003645326,0.0007464556,0.000003083425,0.0006425427,0.002288473],"genre_scores_gemma":[0.9876468,0.00006296646,0.01000289,0.000004988209,0.0003060738,0.0003400399,0.0001831736,0.00002521585,0.001427888],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4873638,"threshold_uncertainty_score":0.7091836,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03886721325293378,"score_gpt":0.4050019376287716,"score_spread":0.3661347243758378,"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."}}