{"id":"W2128363176","doi":"10.3109/0142159x.2011.551559","title":"Criteria for good assessment: Consensus statement and recommendations from the Ottawa 2010 Conference","year":2011,"lang":"en","type":"article","venue":"Medical Teacher","topic":"Innovations in Medical Education","field":"Medicine","cited_by":559,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Summative assessment; Formative assessment; Accreditation; Consistency (knowledge bases); Medical education; Health care; Coherence (philosophical gambling strategy); Statement (logic); Medicine; Psychology; Computer science; Political science; Mathematics education","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.4601737,0.005162985,0.01150328,0.02986555,0.01560879,0.02075921,0.03403722,0.03264016,0.006992188],"category_scores_gemma":[0.4859276,0.005672056,0.01680056,0.02280941,0.02218596,0.01631561,0.02653894,0.03569626,0.008575211],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.05421475,"about_ca_system_score_gemma":0.2472021,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.1145464,"about_ca_topic_score_gemma":0.1161079,"domain_scores_codex":[0.5821559,0.1627519,0.1430939,0.008803454,0.089708,0.01348681],"domain_scores_gemma":[0.3360023,0.2016196,0.02555239,0.02490589,0.3888863,0.0230335],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0005210195,0.0004812332,0.004432616,0.02121525,0.0005338674,0.0008480721,0.006860815,0.002373599,0.0008484136,0.04411963,0.6325647,0.2852007],"study_design_scores_gemma":[0.000640498,0.0003221596,0.010039,0.1068293,0.0009689169,0.001217501,0.0106437,0.004078612,0.002015417,0.06318638,0.7989261,0.001132495],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"other","genre_scores_codex":[0.003448979,0.1000736,0.1843638,0.5821496,0.04397484,0.03037468,0.004380343,0.002320032,0.04891411],"genre_scores_gemma":[0.03540755,0.04299,0.8032877,0.0585724,0.00280741,0.03800977,0.00666058,0.001622391,0.01064222],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.4601737,"threshold_uncertainty_score":0.665702,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1056044521845296,"score_gpt":0.4178424629684081,"score_spread":0.3122380107838785,"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."}}