{"id":"W4293180673","doi":"10.1002/ev.20490","title":"The importance of implementation: Putting evaluation policy to work","year":2022,"lang":"en","type":"article","venue":"New Directions for Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Work (physics); Government (linguistics); White paper; Public administration; Public relations; Early adopter; Program evaluation; Evaluation methods; Policy analysis; Business; Political science; Marketing","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.03211819,0.0001170788,0.0001602029,0.000489687,0.00185349,0.0001744437,0.0004579157,0.00002514043,0.004750548],"category_scores_gemma":[0.005320163,0.00009303616,0.0001403991,0.003396682,0.00002638577,0.0003234785,0.0001232704,0.0001000461,0.00004121472],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008516531,"about_ca_system_score_gemma":0.002430512,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001021378,"about_ca_topic_score_gemma":0.0008873572,"domain_scores_codex":[0.9924928,0.0009251672,0.001063622,0.0004202659,0.0048392,0.0002589448],"domain_scores_gemma":[0.9951912,0.001343301,0.0006796181,0.0005956315,0.002096661,0.00009358202],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00005482755,0.00003213319,0.01388251,0.000001167026,0.00002573282,1.246013e-8,0.001369331,0.03815974,0.0001446965,0.00326675,0.05568473,0.8873783],"study_design_scores_gemma":[0.003172344,0.000496369,0.185131,0.000009426855,0.0002312033,0.000004388635,0.01039038,0.1072074,0.00114497,0.1185198,0.5734286,0.0002641207],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7828459,0.001607929,0.03320837,0.1342232,0.00710684,0.01612485,0.0001682483,0.0001750146,0.02453962],"genre_scores_gemma":[0.9893297,0.00001105247,0.003197593,0.0007444617,0.0004466729,0.003292273,0.0001131433,0.00001727349,0.002847841],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8871142,"threshold_uncertainty_score":0.999446,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2980319978182459,"score_gpt":0.5961758186242162,"score_spread":0.2981438208059703,"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."}}