{"id":"W2560370233","doi":"10.3138/cjpe.306","title":"Measuring Evaluation Capacity in Ontario Public Health Units","year":2016,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":17,"is_retracted":false,"has_abstract":true,"ca_institutions":"École Nationale d'Administration Publique","funders":"","keywords":"Business; Capacity building; Process (computing); Evaluation methods; Knowledge management; Process management; Public health; Program evaluation; Public relations; Nursing; Medicine; Political science; Computer science; Public administration; Engineering","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"empirical","about_ca_system":false,"about_ca_topic":true,"confidence":"high","status":"direct model label, unvalidated"},{"model":"grok","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":true,"confidence":"high","status":"direct model label, unvalidated"},{"model":"opus","categories":["metaresearch"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":true,"confidence":"low","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.07772779,0.0001388744,0.000297573,0.001486377,0.0002146469,0.0003679996,0.0004881195,0.00008400792,0.004825947],"category_scores_gemma":[0.0103476,0.00009199527,0.00008091174,0.001375168,0.00008610649,0.00149792,0.00001236262,0.0002651272,0.00009180222],"about_ca_system_candidate":true,"about_ca_system_consensus":true,"about_ca_system_score_codex":0.006303264,"about_ca_system_score_gemma":0.05175488,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.04558,"about_ca_topic_score_gemma":0.9618332,"domain_scores_codex":[0.9908561,0.002211713,0.001492516,0.0002781567,0.004652212,0.0005093546],"domain_scores_gemma":[0.9899289,0.0002872569,0.001061718,0.0003435439,0.00760336,0.0007752446],"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":[0.000005374739,0.00002848447,0.2023553,0.000001376586,0.00001046393,0.000001053058,0.002440505,0.0003909313,0.00002261427,0.0002684385,0.0006094369,0.793866],"study_design_scores_gemma":[0.002396837,0.0004616596,0.9092016,0.0001561122,0.00002557551,0.00003316335,0.0009303371,0.01110099,0.00004331442,0.01139984,0.06407487,0.0001757353],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9806124,0.0002379205,0.001374864,0.01262183,0.00103512,0.001430817,0.000004369182,0.000007589852,0.002675085],"genre_scores_gemma":[0.9982151,0.000006416866,0.001182246,0.000234648,0.00009159034,0.00008308589,0.000006704334,0.000008786118,0.0001714259],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9162532,"threshold_uncertainty_score":0.9979886,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9060914930214881,"score_gpt":0.5182209136786268,"score_spread":0.3878705793428613,"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."}}