{"id":"W3138688726","doi":"10.3138/cjpe.69697","title":"Pinpointing Where to Start: A Reflective Analysis on the Introductory Evaluation Course","year":2021,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Course (navigation); Reflective practice; Mathematics education; Process (computing); Psychology; Graduate students; Computer science; Course evaluation; Learning theory; Pedagogy; Higher education; Engineering","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":"qualitative","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"grok","categories":[],"domain":null,"study_design":"design_other","genre":"methods","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":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04227577,0.00015171,0.0003184738,0.0009598667,0.0004537653,0.000710285,0.0004704127,0.00006883236,0.009441528],"category_scores_gemma":[0.01381342,0.000102727,0.0002638288,0.004166788,0.00006717819,0.0004382795,0.00001992659,0.0003218886,0.000183839],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009887069,"about_ca_system_score_gemma":0.01036737,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002525449,"about_ca_topic_score_gemma":0.08043789,"domain_scores_codex":[0.9910305,0.002585319,0.001002921,0.000375063,0.004685971,0.0003202408],"domain_scores_gemma":[0.9849547,0.0006497936,0.0007607357,0.0006342299,0.01254981,0.0004507137],"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":[0.00003187795,0.00008974557,0.01952378,0.000002302498,0.0004222885,0.00000811473,0.004017961,0.07032957,0.0001596682,0.0008994016,0.01645392,0.8880613],"study_design_scores_gemma":[0.001544809,0.001279256,0.4732085,0.0001309586,0.003155759,0.00004226936,0.0169516,0.3899243,0.001052184,0.01784125,0.09440821,0.0004608995],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9339702,0.0009903025,0.002128982,0.05197724,0.001477543,0.0019543,0.00001341451,0.0000111338,0.007476907],"genre_scores_gemma":[0.9961181,0.000006358413,0.001980711,0.001116275,0.0003835979,0.0001218779,0.00001714136,0.0000103101,0.0002456338],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8876005,"threshold_uncertainty_score":0.995243,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3379749603270091,"score_gpt":0.5470026099765712,"score_spread":0.2090276496495621,"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."}}