{"id":"W4393862188","doi":"10.3138/cjpe-2024-0012","title":"Strengthening Evaluation Capacity Building Practice Through Competition: The Max Bell School of Public Policy’s Evaluation Capacity Case Challenge","year":2024,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta; University of Ottawa; Queen's University; McGill University","funders":"","keywords":"Competition (biology); Capacity building; Political science; Architectural engineering; Economics; Engineering; Economic growth","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","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1047157,0.0003012019,0.0004276051,0.001338922,0.0008181958,0.001447219,0.0007122868,0.0002034479,0.00562904],"category_scores_gemma":[0.03645203,0.0002191361,0.0003020657,0.00258753,0.0003390133,0.004158949,0.00004088896,0.0008102994,0.0001179289],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0027868,"about_ca_system_score_gemma":0.01966661,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004366387,"about_ca_topic_score_gemma":0.02477061,"domain_scores_codex":[0.9820954,0.006348348,0.002226947,0.0005306072,0.008238618,0.000560148],"domain_scores_gemma":[0.9792621,0.0019567,0.001802474,0.0007152905,0.01570233,0.0005611286],"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":[0.00002478319,0.000155693,0.00104477,0.00005032038,0.0002596292,0.00003555239,0.009822521,0.01722323,0.0001527928,0.0251027,0.001638203,0.9444898],"study_design_scores_gemma":[0.00191834,0.0008940854,0.005100744,0.0004265259,0.001362845,0.002448057,0.01328542,0.8111745,0.0003055327,0.083402,0.07925088,0.0004310563],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9290416,0.006165314,0.00961997,0.03238584,0.003033485,0.004896387,0.00007216036,0.00004756741,0.01473767],"genre_scores_gemma":[0.9911991,0.00008760161,0.00714907,0.0002582991,0.0009318211,0.0002777748,0.00003381633,0.00002677569,0.00003578997],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9440587,"threshold_uncertainty_score":0.9995894,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6298851995534387,"score_gpt":0.5347281576265688,"score_spread":0.09515704192686991,"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."}}