{"id":"W2512095165","doi":"10.3138/cjpe.274","title":"Evaluate This! A Case for Developing Evaluation Competencies","year":2016,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Saskatchewan","funders":"","keywords":"Competition (biology); Psychology; Business; Medical education; Public relations; Political science; Medicine; Ecology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05449253,0.0001920546,0.0003415507,0.000709017,0.0004894453,0.0004520881,0.0005054364,0.0001108558,0.003913481],"category_scores_gemma":[0.01280846,0.0001180008,0.0002225537,0.000749348,0.0001775007,0.001178433,0.00001700644,0.0001150575,0.0001272464],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001566738,"about_ca_system_score_gemma":0.01283316,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003066562,"about_ca_topic_score_gemma":0.02646582,"domain_scores_codex":[0.992824,0.001156079,0.001497194,0.0003337152,0.003743916,0.0004451538],"domain_scores_gemma":[0.9837123,0.001074615,0.001204151,0.0003925724,0.01318376,0.0004326463],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00002671645,0.00001994477,0.003724236,0.00000539601,0.00004142454,0.00001623783,0.0006614036,0.0003435607,0.0001848193,0.001083594,0.004237496,0.9896552],"study_design_scores_gemma":[0.01335881,0.002619088,0.04398893,0.0006078176,0.00115797,0.003604011,0.004560872,0.4163085,0.001879115,0.1457634,0.365082,0.001069495],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9254388,0.001235894,0.03716546,0.02273854,0.003605374,0.006586937,0.00005053155,0.00003215231,0.003146303],"genre_scores_gemma":[0.9771815,0.00001267117,0.02144517,0.0002544582,0.0003356656,0.0004372447,0.00001065066,0.00001735211,0.0003052743],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9885857,"threshold_uncertainty_score":0.9969971,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5723191234277859,"score_gpt":0.562928248525481,"score_spread":0.009390874902304946,"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."}}