{"id":"W2097486283","doi":"10.1002/ev.20045","title":"Rethinking the Relationship Among Monitoring, Evaluation, and Results‐Based Management: Observations From Canada","year":2013,"lang":"en","type":"article","venue":"New Directions for Evaluation","topic":"Public Policy and Administration Research","field":"Social Sciences","cited_by":24,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Accountability; Government (linguistics); Monitoring and evaluation; Dual (grammatical number); Performance management; Process management; Political science; Public administration; Business; Marketing","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":["sts"],"consensus_categories":[],"category_scores_codex":[0.00461092,0.00009442323,0.00007597301,0.0001132477,0.002654754,0.0004121889,0.0001933034,0.00008620907,0.0002461306],"category_scores_gemma":[0.007940806,0.00008622287,0.00003714042,0.0006512862,0.0001263044,0.0005952404,0.0000187503,0.0001328787,0.000005621489],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005385614,"about_ca_system_score_gemma":0.002649392,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.7622146,"about_ca_topic_score_gemma":0.7844575,"domain_scores_codex":[0.9971409,0.000683565,0.0003566166,0.0002756471,0.001293481,0.0002497954],"domain_scores_gemma":[0.9965351,0.001920546,0.0001845301,0.0002725837,0.0009356948,0.0001516121],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00006711613,0.0001502549,0.3769011,0.00002958827,0.000266086,3.859907e-7,0.02375833,0.00439118,0.00008181138,0.1051739,0.2825956,0.2065847],"study_design_scores_gemma":[0.0009438709,0.00001142865,0.8559817,0.00003602444,0.0001069895,6.639898e-8,0.002209753,0.02283179,0.00003714984,0.07689896,0.04083137,0.0001108672],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.860548,0.0003120295,0.001613235,0.07823433,0.001278153,0.005756032,0.00009611248,0.0001519057,0.05201024],"genre_scores_gemma":[0.9912386,0.00002582625,0.001838183,0.0001694771,0.0005342852,0.001320036,0.0002169307,0.000009943825,0.004646721],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4790806,"threshold_uncertainty_score":0.9986436,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2437606918516431,"score_gpt":0.4279213200316784,"score_spread":0.1841606281800354,"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."}}