{"id":"W1773568500","doi":"10.3138/cjpe.0028.007","title":"Evaluator Competencies: The Aotearoa New Zealand Experience","year":2014,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":15,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Aotearoa; Credentialing; Underpinning; Pedagogy; Set (abstract data type); Engineering ethics; Cultural competence; Sociology; Psychology; Medical education; Engineering; Computer science; Medicine; Gender studies","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02505977,0.0003160152,0.0004793534,0.0008065654,0.008559217,0.004520144,0.001101891,0.001640194,0.006971004],"category_scores_gemma":[0.02230706,0.0005403452,0.0003621705,0.001306601,0.00598086,0.003742598,0.007409636,0.004522505,0.0005019557],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0218253,"about_ca_system_score_gemma":0.07818974,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.407992,"about_ca_topic_score_gemma":0.766743,"domain_scores_codex":[0.9900857,0.004304798,0.0004418401,0.0006296918,0.002871167,0.001666763],"domain_scores_gemma":[0.9788955,0.00425814,0.001204657,0.000918322,0.005644283,0.009079137],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0008128946,0.002479373,0.06485031,0.001280561,0.00007002919,0.005088117,0.6245288,0.0006308171,0.004482775,0.02517902,0.05602435,0.2145729],"study_design_scores_gemma":[0.0001773386,0.001252332,0.09953109,0.001727835,0.00005131689,0.00208596,0.3332466,0.0006268318,0.001292559,0.003224399,0.556626,0.0001576566],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.814419,0.003405361,0.00226642,0.05767208,0.0007281904,0.0008876444,0.0001865557,0.00006431926,0.1203705],"genre_scores_gemma":[0.9260659,0.005711401,0.005855849,0.006845317,0.0001105817,0.0006337994,0.0001280816,0.0001201436,0.05452891],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.407992,"threshold_uncertainty_score":0.8112345,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2732667343657929,"score_gpt":0.4978504535242318,"score_spread":0.2245837191584389,"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."}}