{"id":"W2081701430","doi":"10.3138/cjpe.29.1.62","title":"Assessing the Quality of Aboriginal Program Evaluations","year":2014,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université Laval","funders":"","keywords":"Treasury; Popularity; Quality (philosophy); Context (archaeology); Government (linguistics); Program evaluation; Strengths and weaknesses; Corporate governance; Replicate; Business; Political science; Public administration; Psychology; Finance; Geography; Statistics","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":[0.07163566,0.0001375327,0.0003356489,0.0005315152,0.0004094114,0.0008416753,0.000776155,0.00008141744,0.00143015],"category_scores_gemma":[0.009443286,0.00008536712,0.0002177209,0.00123799,0.0002650796,0.001071939,0.00001056032,0.0002832312,0.00004165031],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002706665,"about_ca_system_score_gemma":0.008181978,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008393119,"about_ca_topic_score_gemma":0.01281561,"domain_scores_codex":[0.9904861,0.002856651,0.001828106,0.000214108,0.004308457,0.0003065838],"domain_scores_gemma":[0.9886031,0.001130484,0.001952101,0.0005300196,0.007432273,0.0003520031],"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":[0.000005855326,0.00007095182,0.045118,0.000005480016,0.00002419898,2.456503e-7,0.0007468679,0.002536395,0.0000763465,0.003177668,0.0008760187,0.9473619],"study_design_scores_gemma":[0.001042499,0.0006837851,0.7191836,0.00006230642,0.0001573284,0.00001821178,0.002133999,0.136143,0.0001590932,0.0241391,0.1161164,0.0001608],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9720997,0.0003101069,0.008894664,0.005399509,0.001274047,0.001951466,0.000005694016,0.00001695206,0.01004792],"genre_scores_gemma":[0.9889395,0.00000503398,0.01043086,0.0001517881,0.0002690737,0.0001151357,0.00001013598,0.00000908745,0.00006944371],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9472012,"threshold_uncertainty_score":0.9994827,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5441703005540053,"score_gpt":0.6713453947850856,"score_spread":0.1271750942310803,"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."}}