{"id":"W4402918453","doi":"10.1080/02602938.2024.2404634","title":"Authentic assessment: from panacea to criticality","year":2024,"lang":"en","type":"article","venue":"Assessment & Evaluation in Higher Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":27,"is_retracted":false,"has_abstract":true,"ca_institutions":"Ontario Tech University","funders":"","keywords":"Panacea (medicine); Psychology; Criticality; Pedagogy; Mathematics education","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"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":[0.002343541,0.000215657,0.0002682106,0.0006807264,0.00009390739,0.0001866106,0.0001436571,0.0001664734,0.01181993],"category_scores_gemma":[0.000382907,0.0002148305,0.00007140153,0.001729059,0.00005498118,0.0003732805,0.0000396454,0.0005017162,0.0003609374],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002648064,"about_ca_system_score_gemma":0.003642146,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000242251,"about_ca_topic_score_gemma":0.00001153062,"domain_scores_codex":[0.9965072,0.0003008,0.0007675593,0.0006770948,0.001438571,0.0003087299],"domain_scores_gemma":[0.9982798,0.0002088346,0.00007415086,0.0006040218,0.0007093041,0.0001239244],"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":[0.00004532932,0.004238204,0.1553859,0.0006849106,0.0002403712,0.000008485202,0.002604673,0.0001326707,0.005715792,0.2427939,0.1452451,0.4429046],"study_design_scores_gemma":[0.0004960314,0.00009284514,0.9086346,0.0006931523,0.0003225308,0.00000338665,0.0006947176,0.01644737,0.00006986879,0.01967533,0.0526465,0.0002236798],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8427647,0.0005035729,0.004650086,0.1048408,0.01161184,0.002290393,0.00001458931,0.0002277837,0.03309628],"genre_scores_gemma":[0.9645715,0.00001284762,0.02364283,0.004905685,0.001134209,0.001494644,0.001236819,0.00004073068,0.002960742],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7532488,"threshold_uncertainty_score":0.9890834,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0740774940749726,"score_gpt":0.5029731262640805,"score_spread":0.4288956321891079,"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."}}