{"id":"W4366449208","doi":"10.3138/cjpe.23.011","title":"L’examen de la qualité des évaluations fédérales: une méta-évaluation réussie?","year":2008,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université Laval","funders":"","keywords":"Credibility; Treasury; Valuation (finance); Quality (philosophy); Relevance (law); Political science; Psychology; Accounting; Business; Philosophy; Epistemology","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04576758,0.0002067951,0.0003572161,0.001125815,0.0008639415,0.0004803896,0.000637438,0.0001729761,0.002859352],"category_scores_gemma":[0.009944887,0.0001712361,0.0002373949,0.001891267,0.0005705472,0.001166319,0.00001693697,0.0003524766,0.0001788303],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001082944,"about_ca_system_score_gemma":0.01522051,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001662685,"about_ca_topic_score_gemma":0.01303783,"domain_scores_codex":[0.9902869,0.003584474,0.001547412,0.0003034954,0.003788516,0.0004891738],"domain_scores_gemma":[0.9924003,0.001175958,0.001039435,0.0004453339,0.004199058,0.0007398976],"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":[0.00002262559,0.0001114751,0.1526268,0.000006231061,0.00005714027,0.00001870221,0.0123677,0.01413113,0.0001458432,0.001079948,0.006209887,0.8132225],"study_design_scores_gemma":[0.001746136,0.0005503846,0.7861068,0.00008264019,0.000187967,0.000457292,0.003573942,0.09723967,0.0001757377,0.07249853,0.03709116,0.0002897894],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9770349,0.0006864899,0.007699292,0.002045351,0.0005022644,0.001050064,0.000009870124,0.00002609359,0.01094571],"genre_scores_gemma":[0.9767848,0.0000854379,0.02190998,0.0002509445,0.00028622,0.0001339614,0.00002665129,0.00002094725,0.0005010146],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8129327,"threshold_uncertainty_score":0.9983948,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5449930809225783,"score_gpt":0.5670762013901169,"score_spread":0.02208312046753869,"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."}}