{"id":"W3138870977","doi":"10.3138/cjpe.69577","title":"Competency-Based Evaluation Education: Four Essential Things to Know and Do","year":2021,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University; University of Alberta","funders":"","keywords":"Pace; Lifelong learning; Perspective (graphical); Context (archaeology); Psychology; Knowledge management; Pedagogy; Computer science; Medical education; Medicine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium","status":"direct model label, unvalidated"},{"model":"grok","categories":[],"domain":null,"study_design":"design_other","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"opus","categories":[],"domain":null,"study_design":"not_applicable","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":[0.01744519,0.000143932,0.0002497338,0.0007667732,0.0003142126,0.0009984743,0.000329294,0.0000851383,0.005250778],"category_scores_gemma":[0.00589044,0.0001252747,0.0001114848,0.001181874,0.00006627875,0.000829813,0.00001868421,0.0001950396,0.00009218212],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004811035,"about_ca_system_score_gemma":0.04616685,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001837855,"about_ca_topic_score_gemma":0.008656777,"domain_scores_codex":[0.9940124,0.001044102,0.000925693,0.00032911,0.00342932,0.0002594423],"domain_scores_gemma":[0.9866195,0.0002540859,0.0005629082,0.0003615936,0.01148398,0.0007179878],"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":[0.00001312005,0.00006571486,0.00981653,0.000006657208,0.00001777557,0.000004635895,0.0009462174,0.00326453,0.0001839578,0.001329744,0.007210107,0.977141],"study_design_scores_gemma":[0.004121224,0.001209954,0.2261313,0.0005254904,0.0006962849,0.0003683431,0.007597805,0.3328786,0.001593329,0.07103066,0.3531037,0.0007432946],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9107004,0.004596017,0.007503259,0.05248715,0.005382741,0.003292026,0.000013955,0.00002309676,0.0160013],"genre_scores_gemma":[0.9714199,0.00001117382,0.02632474,0.001512276,0.0003530431,0.0001190874,0.00003159964,0.00001153792,0.0002166183],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9763977,"threshold_uncertainty_score":0.9956586,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1726522912841817,"score_gpt":0.5158863399418671,"score_spread":0.3432340486576854,"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."}}