{"id":"W2004471759","doi":"10.7202/1024862ar","title":"Analyse des différentes facettes influant sur la fidélité de l’épreuve d’expression écrite d’un test de français langue étrangère","year":2014,"lang":"fr","type":"article","venue":"Mesure et évaluation en éducation","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Humanities; Art","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.07018414,0.0004020692,0.0005440397,0.000742191,0.0004612647,0.0005329998,0.0007494644,0.0005446954,0.001155915],"category_scores_gemma":[0.3125679,0.0003474919,0.0002462605,0.002179539,0.00021542,0.0008847503,0.0001363725,0.0006561683,0.0001587818],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002925783,"about_ca_system_score_gemma":0.0006145205,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00170628,"about_ca_topic_score_gemma":0.0004376122,"domain_scores_codex":[0.9781796,0.01790642,0.001063874,0.0008185881,0.001350498,0.0006809669],"domain_scores_gemma":[0.8881526,0.1088256,0.0007484523,0.0009359641,0.001045602,0.0002917657],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001910928,0.0003521156,0.5319726,0.00007646709,0.00004049212,0.000002465784,0.07399828,0.00244868,0.02508674,0.000220031,0.005656086,0.3601269],"study_design_scores_gemma":[0.0007069566,0.0001333766,0.9163865,0.0003772685,0.0001530612,0.00003339367,0.003372812,0.04420763,0.008892976,0.01543371,0.009931381,0.0003709165],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8275726,0.003165982,0.1581767,0.006560956,0.001021001,0.0003298291,0.00005142755,0.0001194626,0.003002106],"genre_scores_gemma":[0.922462,0.0005670044,0.07354683,0.0005588586,0.0003843611,0.00005855875,0.00007412169,0.00004082909,0.002307424],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3844139,"threshold_uncertainty_score":0.9998977,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2584652406471245,"score_gpt":0.4527758650486191,"score_spread":0.1943106244014947,"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."}}