{"id":"W4317381695","doi":"10.3138/cjpe.74582","title":"Fostering an Evaluative Culture by Addressing a Breakdown in the Policy Development Cycle","year":2022,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University","funders":"","keywords":"Accountability; Corporate governance; Function (biology); Process (computing); Citizen journalism; Participatory evaluation; Process management; Political science; Public relations; Public administration; Business; Management; Economics; Computer science; Law","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"qualitative","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"grok","categories":[],"domain":null,"study_design":"design_other","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"opus","categories":[],"domain":null,"study_design":"design_other","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"medium","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.03025192,0.0001248817,0.000187106,0.0007420382,0.0007829283,0.0005820196,0.0008868488,0.00003978303,0.001595349],"category_scores_gemma":[0.001335222,0.00008517215,0.00006813429,0.001464962,0.00005734855,0.0007534418,0.00006572332,0.0003822946,0.000013487],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001247388,"about_ca_system_score_gemma":0.009316388,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008079827,"about_ca_topic_score_gemma":0.0191349,"domain_scores_codex":[0.9925865,0.002343662,0.0008942783,0.000222251,0.003631827,0.0003215132],"domain_scores_gemma":[0.9976367,0.0001534475,0.0006351559,0.0002647603,0.001046187,0.0002637367],"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":[0.00001625903,0.00007198569,0.00267746,0.000001984707,0.00001093143,0.000005693202,0.03983489,0.00911664,0.00007942525,0.00009043364,0.003244153,0.9448501],"study_design_scores_gemma":[0.00419081,0.002397599,0.1377844,0.0001227534,0.0001044531,0.000426786,0.1020417,0.1885276,0.0002309133,0.02637086,0.537186,0.0006160518],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9809259,0.001100543,0.0004812242,0.007594812,0.0005778534,0.002876265,0.00002511335,0.00001031275,0.006407946],"genre_scores_gemma":[0.996736,0.00000442471,0.001668508,0.0008939194,0.0001235931,0.0004393445,0.00003951005,0.000008626414,0.00008607125],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9442341,"threshold_uncertainty_score":0.9993173,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4732297645571751,"score_gpt":0.5733556647379928,"score_spread":0.1001259001808177,"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."}}