{"id":"W1511097611","doi":"10.7202/1027906ar","title":"Évaluer les capacités des élèves à résoudre des problèmes dans le cadre d’une évaluation externe, en France","year":2014,"lang":"fr","type":"article","venue":"Éducation et francophonie","topic":"Innovative Teaching and Learning Methods","field":"Psychology","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Humanities; Philosophy; Political science","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":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.01092597,0.0005822771,0.000597424,0.0003062336,0.001260754,0.0001548623,0.0004556403,0.0006519117,0.0022572],"category_scores_gemma":[0.002309898,0.0006269654,0.0002081227,0.001163142,0.001384059,0.0005030287,0.00005403353,0.001434157,0.0008717907],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004040861,"about_ca_system_score_gemma":0.0004893802,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01773601,"about_ca_topic_score_gemma":0.001113313,"domain_scores_codex":[0.9844515,0.01250515,0.0008315982,0.0009421842,0.000456095,0.0008134512],"domain_scores_gemma":[0.996428,0.001448308,0.0005564232,0.0008335274,0.0005791584,0.0001546434],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00004965625,0.0007670845,0.04961877,0.0004448151,0.0001470634,0.000002867968,0.4076844,0.001970635,0.01037011,0.02261819,0.005693181,0.5006332],"study_design_scores_gemma":[0.001494404,0.000290818,0.8242747,0.000685459,0.00008238474,0.00005830337,0.02058794,0.001648752,0.003000499,0.03207697,0.1150525,0.000747281],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.602618,0.003545451,0.3373758,0.004294379,0.005026593,0.0003955616,0.00002030055,0.0001889674,0.04653485],"genre_scores_gemma":[0.8565416,0.0001582195,0.0647304,0.0005905188,0.001436578,0.0001880224,0.0000956218,0.000138354,0.07612069],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7746559,"threshold_uncertainty_score":0.9999061,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05793218869663264,"score_gpt":0.3707037330175957,"score_spread":0.3127715443209631,"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."}}