{"id":"W4401363543","doi":"10.3138/cjpe-2023-0021","title":"Breaking the Mould? Expanding our Conceptualization of Evaluator Competencies","year":2024,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Conceptualization; Psychology; Engineering ethics; Sociology; Computer science; Engineering; Artificial intelligence","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.02087729,0.0001167946,0.0002310049,0.0006766078,0.000245119,0.0006441518,0.0005074079,0.00006605505,0.001107358],"category_scores_gemma":[0.002915028,0.0000711175,0.0001672586,0.001224098,0.000110757,0.0007439646,0.00001531348,0.0002023622,0.00004446034],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003343034,"about_ca_system_score_gemma":0.005539183,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003259716,"about_ca_topic_score_gemma":0.003917518,"domain_scores_codex":[0.9948405,0.0007820912,0.001113961,0.0001791091,0.002861598,0.000222723],"domain_scores_gemma":[0.9953874,0.0004076203,0.0006356931,0.0002594069,0.003104075,0.0002058579],"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":[0.00001115534,0.00002036827,0.01033862,0.00002043278,0.00007397297,0.000005278891,0.009593023,0.009002294,0.0005832355,0.006538028,0.00611918,0.9576944],"study_design_scores_gemma":[0.001127634,0.0008452901,0.04374561,0.0006283842,0.0004627538,0.0001700738,0.04613392,0.7582368,0.001492944,0.0243635,0.1224642,0.0003288928],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9558064,0.00813821,0.01360678,0.01013778,0.005654345,0.002206736,0.00002594653,0.00003686604,0.00438695],"genre_scores_gemma":[0.9982228,0.00001763726,0.001161144,0.0001154787,0.0002787496,0.00003681098,0.000008188823,0.00001078603,0.0001484004],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9573655,"threshold_uncertainty_score":0.9998057,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4040203586186757,"score_gpt":0.5589097674439346,"score_spread":0.1548894088252589,"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."}}