{"id":"W4413198593","doi":"10.3102/ip.25.2233403","title":"Examining the Internal Structure and Measurement Invariances of Curriculum-based French Fluency Measures","year":2025,"lang":"en","type":"article","venue":"","topic":"Educational Technology and Assessment","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Fluency; Curriculum; Computer science; Artificial intelligence; Mathematics education; Psychology; Pedagogy","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":[0.0003491259,0.00006811817,0.0000794083,0.00007160276,0.00008873754,0.0000391055,0.0005645904,0.0000370337,0.00001461385],"category_scores_gemma":[0.00008058101,0.00004081124,0.00001414702,0.000243179,0.00008252807,0.000090574,0.0001097592,0.0001099593,3.656859e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002389141,"about_ca_system_score_gemma":0.0002009938,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000103538,"about_ca_topic_score_gemma":0.00007493496,"domain_scores_codex":[0.9993109,0.00004840682,0.0001431185,0.0001739025,0.0002385981,0.00008514129],"domain_scores_gemma":[0.9994708,0.00006546555,0.00005916828,0.0002473709,0.000143,0.00001422497],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.00000141385,0.0001079119,0.3157032,0.0000397165,0.00008572775,9.050597e-7,0.0003035931,0.00009800484,0.005983651,0.6054103,0.002438787,0.06982682],"study_design_scores_gemma":[0.0005234812,0.0001314278,0.7729592,0.0003192119,0.0000304628,0.000005666217,0.0005832571,0.008473597,0.09121054,0.1237519,0.001793298,0.0002178809],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3478322,0.001408341,0.6361588,0.009102502,0.0005772365,0.00018017,0.000001324596,0.00008050362,0.004658918],"genre_scores_gemma":[0.980602,0.00000539782,0.01899925,0.0003212758,0.000009503033,0.000009815966,2.289627e-7,9.306111e-7,0.0000516134],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6327698,"threshold_uncertainty_score":0.1664234,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02812102469365024,"score_gpt":0.2741693927075776,"score_spread":0.2460483680139274,"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."}}