{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1797911,0.001211183,0.001464634,0.001693439,0.004438883,0.003066964,0.002782635,0.001279517,0.001842764],"category_scores_gemma":[0.4369418,0.001137938,0.0006412638,0.002778712,0.001561576,0.005501316,0.0006964104,0.001388887,0.0005981548],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001375287,"about_ca_system_score_gemma":0.003401617,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004699538,"about_ca_topic_score_gemma":0.002826488,"domain_scores_codex":[0.9385826,0.04765933,0.003408085,0.002715918,0.006045607,0.001588462],"domain_scores_gemma":[0.9129876,0.07288583,0.004311705,0.003392318,0.005850052,0.0005724126],"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":[0.0001162977,0.0004929483,0.1369423,0.0001667451,0.00009520353,0.00000925751,0.1194309,0.008427281,0.003035672,0.0005086358,0.007618758,0.7231559],"study_design_scores_gemma":[0.001450002,0.0004510893,0.7341303,0.001569572,0.0004331553,0.00007911128,0.00963157,0.04641419,0.003295536,0.1992791,0.002314827,0.0009515036],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7571529,0.006200525,0.1919337,0.01645799,0.01765001,0.002531452,0.00008859128,0.0002762298,0.007708583],"genre_scores_gemma":[0.8321702,0.006659945,0.150949,0.0003269989,0.001301961,0.0004371368,0.0004138661,0.0001394312,0.007601425],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7222044,"threshold_uncertainty_score":0.9991071,"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."}}