{"id":"W4408651722","doi":"10.1037/spq0000691","title":"Improving written-expression curriculum-based measurement feasibility with automated writing evaluation programs.","year":2025,"lang":"en","type":"article","venue":"School Psychology","topic":"Writing and Handwriting Education","field":"Social Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Curriculum; Computer science; Expression (computer science); Medical physics; Mathematics education; Medical education; Information retrieval; Psychology; Programming language; Pedagogy; Medicine","routes":{"ca_aff":true,"ca_fund":false,"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":[0.03087657,0.000480855,0.0004774643,0.001896913,0.0003525519,0.001557182,0.0009760211,0.0006017605,0.002203406],"category_scores_gemma":[0.1305761,0.0002858358,0.0004017038,0.001487934,0.0005083093,0.001785793,0.001565577,0.0008609565,0.0009576894],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005967433,"about_ca_system_score_gemma":0.001567903,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002038023,"about_ca_topic_score_gemma":0.004047564,"domain_scores_codex":[0.9735091,0.01735629,0.001961839,0.001974118,0.004750452,0.0004482758],"domain_scores_gemma":[0.8917776,0.0747169,0.01207765,0.005502022,0.01466551,0.001260218],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.000632603,0.001046972,0.6561944,0.0001990801,0.0001131013,0.00004372755,0.001521535,0.001211898,0.002072161,0.0007298665,0.002484583,0.3337501],"study_design_scores_gemma":[0.0002306125,0.002735805,0.9498851,0.0002668974,0.00008085499,0.000260711,0.001106048,0.03064121,0.007183989,0.001374802,0.00617075,0.00006321813],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9495075,0.0003962953,0.03945533,0.0005550406,0.0001118577,0.001434569,0.0007139812,0.0005745363,0.007250847],"genre_scores_gemma":[0.9514604,0.0001595491,0.04466278,0.0001658526,0.00005546638,0.001615432,0.0005572863,0.0000497289,0.001273548],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03087657,"threshold_uncertainty_score":0.1632928,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06708932981581875,"score_gpt":0.4231695047081389,"score_spread":0.3560801748923201,"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."}}