{"id":"W2945068297","doi":"10.3138/cjpe.42239","title":"20 Years Later: Reflections on the CES Student Evaluation Case Competition","year":2019,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Memorial University of Newfoundland; University of Waterloo","funders":"","keywords":"Competition (biology); Teamwork; Psychology; Computer-assisted web interviewing; Medical education; Social psychology; Applied psychology; Marketing; Management; Medicine; Business; 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":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.02542967,0.0001257617,0.0001888533,0.0006562268,0.0003478065,0.000628839,0.0004072994,0.00007548431,0.0105732],"category_scores_gemma":[0.001241578,0.00008346226,0.0001386233,0.000826827,0.00006906046,0.0005393895,0.00001277835,0.0002896513,0.0007281159],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007010905,"about_ca_system_score_gemma":0.002861454,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002891715,"about_ca_topic_score_gemma":0.02711431,"domain_scores_codex":[0.9936551,0.001454591,0.0008274062,0.0002327917,0.00358929,0.0002408125],"domain_scores_gemma":[0.9948979,0.000471921,0.0006834847,0.0004266595,0.003262417,0.0002576057],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00004867686,0.0001991413,0.03151545,0.000004813307,0.000137412,0.00006047638,0.007351899,0.04024589,0.0002253376,0.006259571,0.008096869,0.9058545],"study_design_scores_gemma":[0.004671789,0.004088218,0.4377646,0.0002280897,0.0006627877,0.001541105,0.02856543,0.2349272,0.0003385663,0.03249306,0.25407,0.0006491061],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9789716,0.0001321133,0.00008415222,0.004943853,0.001819527,0.002079197,0.000009155751,0.000009293989,0.01195114],"genre_scores_gemma":[0.9985334,0.000006963136,0.000326697,0.0004500573,0.0002140786,0.0001177929,0.00001766826,0.00001009043,0.0003232264],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9052054,"threshold_uncertainty_score":0.9906383,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4493941165396864,"score_gpt":0.5776639702248278,"score_spread":0.1282698536851415,"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."}}