{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006206288,0.0001564526,0.0001901486,0.0001835782,0.0008308684,0.0001809489,0.0002497102,0.0001958479,0.0001059121],"category_scores_gemma":[0.001820097,0.0001429546,0.00005807793,0.000688536,0.0002080996,0.0001703416,0.00002019818,0.0003031111,0.00002665373],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000565104,"about_ca_system_score_gemma":0.001196209,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007666233,"about_ca_topic_score_gemma":0.0009992194,"domain_scores_codex":[0.9967689,0.0009833709,0.0003728408,0.0005825176,0.0008150887,0.0004772984],"domain_scores_gemma":[0.9980626,0.00006593044,0.0002041735,0.0003807106,0.001141672,0.0001448738],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001631783,0.001588713,0.8113167,0.0001985068,0.00004842421,0.000003656271,0.001120226,0.00007304298,0.01136458,0.001211489,0.01073962,0.1621718],"study_design_scores_gemma":[0.01310155,0.0009586721,0.9007054,0.004513205,0.0005008842,0.000006476389,0.0383835,0.02173392,0.002703061,0.005232696,0.01049377,0.001666854],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9519852,0.0003978613,0.003158866,0.006664255,0.0008379048,0.001757081,0.000001678034,0.001290487,0.03390661],"genre_scores_gemma":[0.9946754,0.000004184361,0.004159853,0.000541824,0.0002446745,0.0002724814,0.00002301816,0.00001279033,0.00006576962],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.160505,"threshold_uncertainty_score":0.6390452,"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."}}