{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.03118654,0.0001127504,0.0004069622,0.0008235444,0.0008892583,0.0003356398,0.0003245012,0.00003455047,0.0128465],"category_scores_gemma":[0.003128179,0.00008562035,0.0001338667,0.0008049009,0.00006681443,0.0004768224,0.0001160248,0.0005352324,0.00006141159],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001824284,"about_ca_system_score_gemma":0.0004039083,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001381079,"about_ca_topic_score_gemma":0.00002190246,"domain_scores_codex":[0.992253,0.002036484,0.0008555296,0.0002688907,0.004420545,0.0001655258],"domain_scores_gemma":[0.9908607,0.004736788,0.00295921,0.0002444363,0.00111087,0.00008798259],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0006931798,0.0004010312,0.196331,0.00001105085,0.002217017,0.00001425384,0.05176181,0.6666053,0.0003385353,0.01213914,0.006994845,0.06249286],"study_design_scores_gemma":[0.001112073,0.0008362589,0.123993,0.00003860748,0.001819367,0.00004289169,0.05511893,0.7387169,0.0002065874,0.01722055,0.06059799,0.0002968814],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9867507,0.0005170259,0.001168497,0.002872792,0.0003606959,0.0001813226,0.00001999797,0.000006894993,0.008122131],"genre_scores_gemma":[0.9970924,0.00007905113,0.001703938,0.0008203969,0.00008260916,0.00001662046,0.000005989898,0.000003893119,0.0001951256],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07233799,"threshold_uncertainty_score":0.9975973,"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."}}