{"id":"W4389672983","doi":"10.56645/jmde.v19i46.971","title":"Building Spaces for Dialogues to Rethink Evaluator Competencies: Lessons from the Webinars Organized by the Evaluation Centre for Complex Health Interventions","year":2023,"lang":"en","type":"article","venue":"Journal of MultiDisciplinary Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Psychological intervention; General partnership; Sustainability; Political science; Public relations; Capacity building; Diversity (politics); Sociology; Engineering ethics; Psychology; Engineering","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.08628207,0.0002617324,0.0005913465,0.0003780634,0.001835139,0.0005200599,0.001274733,0.00009390596,0.0005963356],"category_scores_gemma":[0.01733043,0.0001451176,0.000570278,0.001404734,0.0001154299,0.0006531226,0.0002419302,0.0002642117,0.00008261446],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006926871,"about_ca_system_score_gemma":0.001696581,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004315579,"about_ca_topic_score_gemma":0.0006291792,"domain_scores_codex":[0.9880425,0.003295754,0.002231157,0.0005143845,0.005463848,0.0004523673],"domain_scores_gemma":[0.9805071,0.01023758,0.002357464,0.0006663407,0.006035125,0.0001964357],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0008063603,0.0003799325,0.003163186,0.00005562135,0.0003204088,3.781619e-7,0.04256512,0.06134799,0.01423991,0.001134653,0.6319903,0.2439962],"study_design_scores_gemma":[0.004473466,0.0009183857,0.07942727,0.0003612,0.0004513563,0.000005364103,0.03314406,0.8088829,0.0002997275,0.05463023,0.01717115,0.0002349179],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5937175,0.0009798642,0.06203971,0.3342324,0.00239981,0.005964953,0.0005957481,0.00004128589,0.00002874977],"genre_scores_gemma":[0.9708983,0.0001259333,0.02588683,0.000873149,0.0007850747,0.0004724551,0.000566364,0.00004302751,0.0003488221],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7475349,"threshold_uncertainty_score":0.9994643,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4878210702307514,"score_gpt":0.5796790978903889,"score_spread":0.09185802765963741,"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."}}