{"id":"W4205154527","doi":"10.1097/acm.0000000000004507","title":"Written-Based Progress Testing: A Scoping Review","year":2021,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":24,"is_retracted":false,"has_abstract":true,"ca_institutions":"Medical Council of Canada; Université de Sherbrooke","funders":"","keywords":"Facilitator; Formative assessment; Test (biology); Inclusion (mineral); Consistency (knowledge bases); Thematic analysis; MEDLINE; Data extraction","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":[0.001580857,0.0001105434,0.0003166119,0.00004023409,0.0002484467,0.0000108503,0.0002972121,0.0001171145,0.001108643],"category_scores_gemma":[0.003268892,0.00009172235,0.00003269186,0.0008528719,0.0005052378,0.00008497213,0.00004675278,0.000354454,0.00005273643],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005223002,"about_ca_system_score_gemma":0.000648756,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006543074,"about_ca_topic_score_gemma":0.0000300827,"domain_scores_codex":[0.9980575,0.0002144691,0.0003614752,0.0002641389,0.0007545416,0.0003478668],"domain_scores_gemma":[0.9987917,0.0005051514,0.0001458482,0.0001653307,0.0002255233,0.0001664694],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"systematic_review","study_design_scores_codex":[0.00002245186,0.0000881931,0.6206216,0.01147349,0.0001078337,0.0003627288,0.02280844,0.000001444968,0.001634339,0.02326201,0.2414205,0.07819704],"study_design_scores_gemma":[0.004782869,0.000348676,0.01903068,0.713478,0.0007793867,0.00002618247,0.02160632,0.0001623406,0.000478116,0.003130646,0.2348267,0.001350076],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.0155075,0.503437,0.0003286262,0.3088339,0.001664166,0.002331222,0.000001517294,0.0005004418,0.1673956],"genre_scores_gemma":[0.7477843,0.1454131,0.003447406,0.07659221,0.009358838,0.0004224377,0.00008070895,0.00008210897,0.01681891],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7322768,"threshold_uncertainty_score":0.9998045,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1411212440806743,"score_gpt":0.4682054334345254,"score_spread":0.3270841893538511,"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."}}