{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01815563,0.0003447365,0.0004706857,0.002360853,0.005689996,0.003392787,0.002544047,0.0005498048,0.002972344],"category_scores_gemma":[0.03643117,0.0005912656,0.00034958,0.003297391,0.003820051,0.001332575,0.004857802,0.0008137372,0.0002338003],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.06919046,"about_ca_system_score_gemma":0.1080654,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.884464,"about_ca_topic_score_gemma":0.915966,"domain_scores_codex":[0.9809694,0.007104488,0.001106353,0.00108859,0.005741033,0.003989986],"domain_scores_gemma":[0.9435164,0.01310909,0.008763645,0.002334098,0.02364382,0.008632962],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004194576,0.0005019456,0.8581892,0.0006092588,0.0001114656,0.0002381477,0.07226549,0.001534028,0.001516621,0.001346101,0.003927798,0.0593406],"study_design_scores_gemma":[0.00003291716,0.0002369251,0.9336769,0.0002318797,0.0000271977,0.00003351982,0.05512895,0.001092246,0.0006745586,0.0002487689,0.008577416,0.000038719],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9904447,0.0003497356,0.0005609811,0.0009344448,0.0000143305,0.0003745178,0.0002889868,0.00002653654,0.007005837],"genre_scores_gemma":[0.9980913,0.0001460985,0.0007336879,0.00007041125,0.000004935684,0.0001462337,0.0001342803,0.000005012755,0.0006681128],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9818444,"threshold_uncertainty_score":0,"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."}}