{"id":"W4393902965","doi":"10.3138/cjpe-2024-0004","title":"Developing Evaluation Capacity Building Competencies: Participant Reflections From the 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":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University; University of Waterloo; University of Alberta","funders":"","keywords":"Capacity building; Capacity development; Psychology; Business; Environmental resource management; Environmental science; Economic growth; Economics","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.03962088,0.001255139,0.001028547,0.00158862,0.01829979,0.01175484,0.004321632,0.008394639,0.002839743],"category_scores_gemma":[0.09979644,0.001162951,0.0008462839,0.0008236015,0.01508778,0.007047627,0.01385947,0.0166288,0.0005804612],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007757266,"about_ca_system_score_gemma":0.0130957,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01467325,"about_ca_topic_score_gemma":0.03384944,"domain_scores_codex":[0.9588882,0.03007218,0.0008589815,0.001570779,0.003768491,0.004841317],"domain_scores_gemma":[0.9227225,0.04596205,0.002564524,0.002351765,0.01292306,0.01347621],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00008087114,0.0005049874,0.003045201,0.0001039945,0.00001062111,0.003118983,0.9628761,0.0003442855,0.0008801348,0.004780687,0.009930288,0.01432377],"study_design_scores_gemma":[0.00001847038,0.0002370566,0.0008856237,0.0001592468,0.000006036342,0.0009963831,0.9332724,0.0006319584,0.0007226864,0.001852987,0.06113582,0.00008135985],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8987003,0.0007176431,0.0105548,0.06251436,0.001420164,0.0009766342,0.0001674666,0.0001263293,0.02482233],"genre_scores_gemma":[0.9696973,0.0006756545,0.006922972,0.008464115,0.0003096612,0.0007301589,0.0001015602,0.0001874803,0.01291111],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9603791,"threshold_uncertainty_score":0.2095378,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8696201595344439,"score_gpt":0.5799551477667225,"score_spread":0.2896650117677214,"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."}}