{"id":"W2612271425","doi":"10.7202/1040137ar","title":"Évaluateurs évalués : évaluation diagnostique des compétences en évaluation des correcteurs d’une épreuve d’expression écrite à forts enjeux","year":2017,"lang":"fr","type":"article","venue":"Mesure et évaluation en éducation","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Humanities; Valuation (finance); Mathematics; Art; Economics","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":[0.02134745,0.001243671,0.0009848694,0.005008899,0.00104115,0.003882608,0.0009828705,0.00107615,0.007439815],"category_scores_gemma":[0.08754975,0.0005597275,0.001539782,0.002366486,0.001646884,0.003522737,0.002671677,0.001401968,0.002357945],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002359721,"about_ca_system_score_gemma":0.002590644,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006198639,"about_ca_topic_score_gemma":0.009078403,"domain_scores_codex":[0.9733671,0.01231835,0.002318721,0.002149911,0.009070787,0.0007751197],"domain_scores_gemma":[0.9052833,0.05737324,0.008776708,0.005418162,0.02155536,0.001593227],"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":[0.001505339,0.0008306257,0.264028,0.002257022,0.0006690366,0.0002730391,0.02326171,0.002518164,0.01416671,0.01399787,0.008260499,0.668232],"study_design_scores_gemma":[0.0002645238,0.002514034,0.8502095,0.001719206,0.0005808198,0.0008403325,0.01619591,0.01706028,0.034392,0.01811645,0.0576776,0.000429414],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.753267,0.002304927,0.1590746,0.001045508,0.0003623019,0.002876893,0.001994482,0.001404831,0.0776694],"genre_scores_gemma":[0.8663894,0.001144433,0.1007965,0.0002905988,0.00008373213,0.003541309,0.001229875,0.0003490134,0.02617514],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02134745,"threshold_uncertainty_score":0.1128975,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5793917652197416,"score_gpt":0.527240377785965,"score_spread":0.05215138743377656,"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."}}