{"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":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00124901,0.0002606686,0.0002893856,0.00009925151,0.001162588,0.0008376872,0.0003607153,0.00007962795,0.02969694],"category_scores_gemma":[0.0008616106,0.0001451801,0.0001271442,0.00003450052,0.0001164194,0.0006558141,0.00002835592,0.0008162964,0.00003314473],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004366526,"about_ca_system_score_gemma":0.001138888,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.2513172,"about_ca_topic_score_gemma":0.9004955,"domain_scores_codex":[0.9976925,0.0002123143,0.0003822453,0.0002256409,0.0008472952,0.0006399993],"domain_scores_gemma":[0.9983686,0.0004941553,0.0002159947,0.0002411061,0.0003035542,0.000376622],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00004141804,0.00007068738,0.002222664,0.0000266288,0.000168887,0.0002999607,0.02265522,0.00001989249,0.0009168245,0.000971064,0.8719488,0.100658],"study_design_scores_gemma":[0.0006420112,0.00007201519,0.001107207,0.0006866421,0.00005960496,0.0001470326,0.008678379,0.00001946878,0.0001257606,0.0005431279,0.9875828,0.000335957],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.1516865,0.1100957,0.001227257,0.4632355,0.01649083,0.0007977784,0.0003426466,0.0009266173,0.2551972],"genre_scores_gemma":[0.9411541,0.0003404971,0.00005067766,0.03309387,0.003200133,0.00000565773,0.000001582768,0.00006458359,0.02208893],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7894676,"threshold_uncertainty_score":0.97119,"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."}}