{"id":"W4296187046","doi":"10.1787/911cc792-en","title":"Evaluation Framework and Practices: A comparative analysis of five OECD countries","year":2022,"lang":"en","type":"article","venue":"OECD Journal on Budgeting","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Institutionalisation; Quality assurance; Government (linguistics); Function (biology); Regional science; Political science; Quality (philosophy); Public administration; Economic growth; Public economics; Accounting; Business; Economics; Sociology; Operations management","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"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":[0.06069158,0.0004399998,0.0007176286,0.01496582,0.003041281,0.008537906,0.0009335753,0.001043166,0.001945737],"category_scores_gemma":[0.07755077,0.0003425232,0.0008763064,0.02573604,0.003529534,0.003004294,0.00433975,0.0008654935,0.0001981793],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02655912,"about_ca_system_score_gemma":0.02450443,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.07332706,"about_ca_topic_score_gemma":0.03505556,"domain_scores_codex":[0.9529756,0.02830788,0.004317521,0.001177645,0.00959581,0.003625458],"domain_scores_gemma":[0.907778,0.05123717,0.008692373,0.003739683,0.02536964,0.003183146],"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":[0.001356236,0.0004506272,0.302025,0.004139394,0.000572248,0.001683946,0.04031605,0.01373357,0.0007103923,0.2421531,0.01270032,0.3801592],"study_design_scores_gemma":[0.0001942946,0.0005359509,0.6363848,0.008424406,0.0005333473,0.001034949,0.1030452,0.002724403,0.00148529,0.01517223,0.2302682,0.0001968201],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7762027,0.03317812,0.009533816,0.006193387,0.0001729167,0.0009816763,0.002520213,0.0001952728,0.1710218],"genre_scores_gemma":[0.983498,0.008350135,0.004869099,0.000536459,0.00002239338,0.0003705754,0.0008402946,0.00005348909,0.001459593],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07332706,"threshold_uncertainty_score":0.3209716,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2789534152173669,"score_gpt":0.5586174930213613,"score_spread":0.2796640778039944,"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."}}