{"id":"W3191701502","doi":"10.1080/0142159x.2021.1957088","title":"Ottawa 2020 consensus statement for programmatic assessment – 1. Agreement on the principles","year":2021,"lang":"en","type":"article","venue":"Medical Teacher","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":115,"is_retracted":false,"has_abstract":true,"ca_institutions":"The Wilson Centre; University of Toronto","funders":"","keywords":"Operationalization; Medical education; Curriculum; Program evaluation; Engineering ethics; Medicine; Psychology; Political science; Pedagogy; Engineering","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"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":[0.003866223,0.000135594,0.0001903854,0.00001616631,0.0004948003,0.0001359265,0.0003431294,0.00009678915,0.01101056],"category_scores_gemma":[0.001253738,0.00008767218,0.0001205672,0.0001728853,0.0002732927,0.00002391205,0.00009959526,0.0002499265,0.00008431805],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002461123,"about_ca_system_score_gemma":0.0007199071,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001047988,"about_ca_topic_score_gemma":0.00108017,"domain_scores_codex":[0.9966165,0.0005358532,0.0003874056,0.0002872428,0.001723628,0.000449363],"domain_scores_gemma":[0.9988264,0.0004747662,0.0001154181,0.0002405453,0.0001150104,0.0002277943],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00003706308,0.002579601,0.0900499,0.00009598891,0.0005817222,0.0001343552,0.01532624,0.000006110052,0.00003554557,0.4634986,0.3822259,0.04542902],"study_design_scores_gemma":[0.0007923783,0.0001229305,0.005421168,0.00006623169,0.00004630975,6.656558e-7,0.01821395,0.00104638,0.00002012316,0.001236625,0.9728657,0.0001675335],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.3000464,0.0002135977,0.001242553,0.3118043,0.001328619,0.002795127,0.000007051676,0.0001538391,0.3824086],"genre_scores_gemma":[0.9142879,0.0001350095,0.003159609,0.002266363,0.001243004,0.0009615988,0.00004149887,0.00002814251,0.07787691],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6142415,"threshold_uncertainty_score":0.9898935,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07826089378203037,"score_gpt":0.4097405264132591,"score_spread":0.3314796326312287,"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."}}