{"id":"W2166249030","doi":"10.3138/cmlr.65.5.703","title":"Assessing AIM: A Study of Grade 8 Students in an Ontario School Board","year":2009,"lang":"en","type":"article","venue":"Canadian Modern Language Review/ La Revue canadienne des langues vivantes","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":31,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Perception; Psychology; Sample (material); Test (biology); Mathematics education; Medical education; French; Language proficiency; Pedagogy; AP French Language; Language education; Medicine; Humanities","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0008146428,0.0003162646,0.0006819209,0.000421947,0.000271395,0.0003227674,0.0005702421,0.00007056718,0.000487113],"category_scores_gemma":[0.0002825192,0.0003034765,0.0001080542,0.000133845,0.00009902312,0.0005175706,0.00002634724,0.0005786482,0.00001182987],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008176784,"about_ca_system_score_gemma":0.0002662397,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.8833521,"about_ca_topic_score_gemma":0.9980781,"domain_scores_codex":[0.997712,0.0004218661,0.000545861,0.0004512402,0.000160061,0.0007089545],"domain_scores_gemma":[0.9983987,0.00007960519,0.0001910797,0.0005434421,0.0001041999,0.0006829811],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.00001116397,0.0005076047,0.1113433,0.0009292125,0.00009442357,0.002301071,0.8436939,0.00006758337,0.0005572211,0.0009735331,0.0003412124,0.03917981],"study_design_scores_gemma":[0.001949332,0.001307128,0.712959,0.01506018,0.0004263111,0.0001460208,0.224201,0.0005319227,0.00001027051,0.0006865226,0.04099268,0.001729656],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9826146,0.01203778,0.000004553538,0.0001295284,0.0001073558,0.0005411826,0.00002722339,0.00005616006,0.004481644],"genre_scores_gemma":[0.9968397,0.0001935666,0.0001075956,0.0007219043,0.0002060534,0.00002347664,0.00006395499,0.0000436462,0.00180013],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6194929,"threshold_uncertainty_score":0.9999417,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03999005893972341,"score_gpt":0.2966461545881899,"score_spread":0.2566560956484665,"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."}}