{"id":"W2289233763","doi":"10.18806/tesl.v32i0.1219","title":"Test de Francais Laval-Montreal: Does It Measure What It Should Measure?","year":2016,"lang":"en","type":"article","venue":"TESL Canada Journal","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Psychology; Test (biology); Construct validity; Task (project management); Test validity; Construct (python library); Pedagogy; Social psychology; Psychometrics; Developmental psychology; Computer science","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02567692,0.000865743,0.0009656294,0.004327856,0.002424922,0.004210868,0.002610336,0.001845509,0.003332851],"category_scores_gemma":[0.1127466,0.0003373422,0.001112076,0.004900685,0.004286651,0.003633816,0.001740076,0.001921233,0.001414474],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02024679,"about_ca_system_score_gemma":0.02849432,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.6028625,"about_ca_topic_score_gemma":0.6502145,"domain_scores_codex":[0.9789103,0.006320973,0.001233968,0.00166473,0.0105619,0.001308303],"domain_scores_gemma":[0.905587,0.02989892,0.01021522,0.006476117,0.0424013,0.005421489],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002445022,0.0001908567,0.6704596,0.0004719468,0.0003187353,0.00008747409,0.002869217,0.001011569,0.0005938693,0.0107583,0.0316466,0.2813473],"study_design_scores_gemma":[0.00006445509,0.0003901408,0.8942094,0.00124427,0.0001221994,0.0001800201,0.002794048,0.002705032,0.00124557,0.005178305,0.09169808,0.0001684033],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6563758,0.03188809,0.036193,0.06879696,0.004822027,0.003149992,0.01570174,0.001447262,0.1816252],"genre_scores_gemma":[0.9559551,0.003609555,0.0218014,0.004242352,0.0004089325,0.002451899,0.002925726,0.0002390474,0.008365996],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6028625,"threshold_uncertainty_score":0.7989527,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03451480174865623,"score_gpt":0.2311048938086571,"score_spread":0.1965900920600009,"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."}}