{"id":"W3031759560","doi":"10.3138/cjpe.69030","title":"Kylie Hutchinson, ed. (2018). <i>Evaluation Failures: 22 Tales of Mistakes Made and Lessons Learned</i> .","year":2020,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Innovations in Medical Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Psychology","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":[],"consensus_categories":[],"category_scores_codex":[0.003111254,0.0001405061,0.0003029906,0.0003371499,0.0001283012,0.00005848734,0.0001271513,0.0001389624,0.0005616183],"category_scores_gemma":[0.005929621,0.0001309433,0.00007918136,0.00055513,0.000235802,0.000226086,0.000005669726,0.0003814332,0.000008942019],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003562946,"about_ca_system_score_gemma":0.00491629,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003979646,"about_ca_topic_score_gemma":0.0006098129,"domain_scores_codex":[0.9975493,0.0002167615,0.0007415388,0.0001988623,0.001058085,0.00023543],"domain_scores_gemma":[0.9964584,0.00004727596,0.0005830609,0.0001786606,0.002280236,0.0004523297],"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":[0.00006104126,0.00009390626,0.004782101,0.0001775348,0.000103831,0.000006906474,0.00331198,0.0001434755,0.002030483,0.0006960993,0.0244744,0.9641182],"study_design_scores_gemma":[0.01509704,0.006426771,0.1739946,0.002152898,0.005378109,0.0008043015,0.02658831,0.1223324,0.005265228,0.007657079,0.6332458,0.001057486],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9050521,0.003664267,0.001100973,0.08619331,0.0005929573,0.001981051,0.00001615352,0.00002170489,0.001377487],"genre_scores_gemma":[0.9868133,0.00005829354,0.01148574,0.0009135681,0.0004934248,0.00005647679,0.0001214181,0.00002250657,0.00003527823],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9630607,"threshold_uncertainty_score":0.8721288,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1614933911372405,"score_gpt":0.4195411450148336,"score_spread":0.2580477538775932,"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."}}