{"id":"W2946581393","doi":"10.7202/1059172ar","title":"Approches psychométrique et didactique de la validité d’une évaluation externe en mathématiques : quelles complémentarités et quelles divergences ?","year":2019,"lang":"fr","type":"article","venue":"Mesure et évaluation en éducation","topic":"Mathematics Education and Teaching Techniques","field":"Social Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Humanities; Philosophy; Mathematics","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1469562,0.002032535,0.002819103,0.007171583,0.003221899,0.02038199,0.004280072,0.003706688,0.008293128],"category_scores_gemma":[0.4152301,0.001527385,0.002541576,0.005525116,0.02153236,0.02041404,0.00945852,0.008761504,0.00132564],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.009416444,"about_ca_system_score_gemma":0.01313643,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0064152,"about_ca_topic_score_gemma":0.004896313,"domain_scores_codex":[0.7442113,0.1681922,0.014667,0.01839694,0.05216423,0.002368273],"domain_scores_gemma":[0.504374,0.3586896,0.02148705,0.05405743,0.05862198,0.002769811],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.0003358161,0.0001657193,0.01244312,0.002050558,0.000653991,0.0001554206,0.020405,0.004446718,0.001610794,0.789656,0.001708201,0.1663686],"study_design_scores_gemma":[0.0001543244,0.0003835104,0.009795094,0.002146539,0.0003685163,0.0003422817,0.007537641,0.02010942,0.003751015,0.9239069,0.03130875,0.000195911],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.03962411,0.00375906,0.8973101,0.01157825,0.0004568289,0.0006237476,0.0002987039,0.0003318621,0.04601745],"genre_scores_gemma":[0.5513435,0.001581102,0.4359221,0.001546644,0.0004060707,0.002190206,0.0004116475,0.0004130986,0.00618571],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.1469562,"threshold_uncertainty_score":0.7771881,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.061065612239627,"score_gpt":0.4275992748591815,"score_spread":0.3665336626195546,"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."}}