{"id":"W1746934302","doi":"10.3138/cjpe.0028.012","title":"Evaluator Competencies and Professionalizing the Field: Where Are We Now?","year":2014,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":16,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Professionalization; Globe; Narrative; Field (mathematics); Core competency; Happening; Engineering ethics; Sociology; Epistemology; Pedagogy; Psychology; Management; Engineering; Social science; Linguistics; History; Philosophy","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium","status":"direct model label, unvalidated"},{"model":"grok","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"opus","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.02026254,0.0001140676,0.000211937,0.0002988071,0.0004910763,0.0004654196,0.0004542965,0.00007229306,0.001622805],"category_scores_gemma":[0.003451496,0.00006473122,0.00008447996,0.0004251733,0.0001153982,0.0004499189,0.00002314728,0.0002571399,0.00004676583],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001213795,"about_ca_system_score_gemma":0.002199836,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001291561,"about_ca_topic_score_gemma":0.02758917,"domain_scores_codex":[0.9955854,0.001108238,0.0006959777,0.00017314,0.002208029,0.0002291916],"domain_scores_gemma":[0.9950384,0.001183563,0.0007570415,0.0002833003,0.002409653,0.0003279956],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001332565,0.00001839766,0.07032406,0.00001081768,0.00002184494,0.000001459945,0.001724097,0.000585368,0.00002581445,0.001286701,0.05593716,0.870051],"study_design_scores_gemma":[0.0008045443,0.0004739039,0.1159027,0.0002300156,0.00008985761,0.00004401563,0.005221462,0.1249041,0.00004314844,0.02176272,0.7303718,0.0001517375],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.661557,0.007519471,0.003806208,0.3147537,0.003389842,0.002640774,0.000008162366,0.00002454613,0.006300315],"genre_scores_gemma":[0.9957798,0.0001020698,0.00211079,0.001277078,0.0003069304,0.00005351042,0.000001839706,0.000008042242,0.0003600093],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8698992,"threshold_uncertainty_score":0.9992899,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2766781532871869,"score_gpt":0.51874720227499,"score_spread":0.2420690489878032,"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."}}