{"id":"W4366704821","doi":"10.58213/education.v4i1.33","title":"Examining performance indicators for a written expression test based on the curriculum","year":2022,"lang":"en","type":"article","venue":"IAAR Journal of Education - ISSN 2583-6846 Peer-Reviewed Journal","topic":"Writing and Handwriting Education","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Ministry of Advanced Education","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Automaticity; Handwriting; Fluency; Reading (process); Acronym; Psychology; Spelling; Test (biology); Task (project management); Quality (philosophy); Curriculum; Cognitive psychology; Computer science; Linguistics; Mathematics education; Cognition; Artificial intelligence; Pedagogy","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":["sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0129865,0.0002170255,0.0003737094,0.0005914076,0.004172245,0.0003957994,0.0009372214,0.00007748551,0.00125415],"category_scores_gemma":[0.008013553,0.0001621234,0.0002912409,0.0008792712,0.0001325466,0.0004474783,0.00004263712,0.00125793,0.00001181507],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006124643,"about_ca_system_score_gemma":0.003255777,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002842246,"about_ca_topic_score_gemma":0.000004634354,"domain_scores_codex":[0.9954491,0.001112865,0.001068753,0.000220308,0.00168629,0.0004627345],"domain_scores_gemma":[0.9942961,0.001511325,0.002150088,0.0002655701,0.001434626,0.0003422998],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000114335,0.002770048,0.05941032,0.0002150878,0.00006028327,0.000006148326,0.02211726,0.0005913124,0.0005357041,0.001171383,0.7477674,0.1652407],"study_design_scores_gemma":[0.0007849849,0.0009314686,0.008920924,0.001683295,0.0001482077,0.0002103456,0.06985085,0.000250372,0.0002237011,0.0004786335,0.9161665,0.0003507193],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9023051,0.005018119,0.0004701695,0.07665452,0.007433071,0.0008697702,0.00002115455,0.00003854892,0.0071896],"genre_scores_gemma":[0.9888238,0.000789811,0.002107215,0.001797428,0.004063902,0.0001101836,0.00001197682,0.00003429756,0.002261331],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1683991,"threshold_uncertainty_score":0.9996588,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0475548252586622,"score_gpt":0.3470929565802417,"score_spread":0.2995381313215795,"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."}}