{"id":"W2326223763","doi":"10.7202/1034582ar","title":"Fiabilité d’un dispositif d’évaluation de l’habileté à déterminer le résultat d’une chaîne d’opérations chez des élèves québécois du secondaire","year":2016,"lang":"fr","type":"article","venue":"Mesure et évaluation en éducation","topic":"Mathematics Education and Teaching Techniques","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université Laval; Université du Québec à Rimouski","funders":"","keywords":"Humanities; Mathematics; Political science; Philosophy","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006800182,0.001647746,0.001527843,0.002587619,0.001823461,0.004304429,0.001443721,0.001626116,0.008033762],"category_scores_gemma":[0.02118318,0.0008414063,0.001677253,0.002588245,0.00135267,0.002204827,0.001759361,0.001492053,0.001341133],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005396215,"about_ca_system_score_gemma":0.005250703,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.2004446,"about_ca_topic_score_gemma":0.2639161,"domain_scores_codex":[0.9935477,0.0013039,0.0002809083,0.001208192,0.003123034,0.0005362282],"domain_scores_gemma":[0.9795144,0.008878755,0.001614392,0.001358971,0.008009386,0.0006240875],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002641646,0.0006431546,0.5006447,0.003193856,0.00131049,0.0006723106,0.009861699,0.07092728,0.06139009,0.004942213,0.004425844,0.3393466],"study_design_scores_gemma":[0.00006720169,0.001884821,0.8488191,0.0007504863,0.0006973054,0.0002502183,0.009002006,0.09063563,0.0267938,0.00309999,0.01759089,0.0004084793],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9064957,0.001548636,0.06814604,0.0005964936,0.0001417275,0.0004712417,0.001902918,0.000648051,0.02004908],"genre_scores_gemma":[0.9489754,0.0003804494,0.04077443,0.0001033963,0.00001632767,0.000325746,0.001153642,0.0001460947,0.008124509],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7995554,"threshold_uncertainty_score":0.3985557,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.033833876987283,"score_gpt":0.3422327486793048,"score_spread":0.3083988716920217,"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."}}