{"id":"W1836271141","doi":"10.1111/medu.12080","title":"Checklists improve experts’ diagnostic decisions","year":2013,"lang":"en","type":"article","venue":"Medical Education","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":93,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; University Health Network","funders":"","keywords":"Checklist; Context (archaeology); Interpretation (philosophy); Computer science; Cognition; Diagnostic accuracy; Medical physics; Medicine; Psychology; Cognitive psychology; Radiology; Psychiatry","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":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0002505561,0.0001164929,0.0002230295,0.00006275604,0.00004903516,0.00002940792,0.0001126917,0.0001973877,0.007686453],"category_scores_gemma":[0.5659319,0.00008633298,0.00009242405,0.0001671324,0.00009265108,0.00006088688,0.00004026794,0.000223758,0.001550506],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000682389,"about_ca_system_score_gemma":0.001417145,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004567135,"about_ca_topic_score_gemma":0.00000637996,"domain_scores_codex":[0.9985008,0.00002987269,0.0003567183,0.0002756667,0.0005925174,0.0002443726],"domain_scores_gemma":[0.9733502,0.0243934,0.0000997233,0.0005994131,0.00036022,0.00119704],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.00000943652,0.001499822,0.02210869,0.000009660219,0.0000147608,0.000009702184,0.0001742001,1.335176e-7,0.0001098101,0.0003481671,0.515525,0.4601906],"study_design_scores_gemma":[0.002509787,0.0005203492,0.8824908,0.004967599,0.0001481746,0.0001770561,0.0005204818,0.001412547,0.0004739515,0.00252001,0.1038688,0.0003904994],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9406537,0.0006638164,0.0003680169,0.04988636,0.002158749,0.0005600487,0.000001337451,0.0001053281,0.005602685],"genre_scores_gemma":[0.971925,0.0002734278,0.001430096,0.02168458,0.001035632,0.0004083034,0.00008835031,0.00002081807,0.003133829],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.860382,"threshold_uncertainty_score":0.9992269,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01513996075955962,"score_gpt":0.3605869921277853,"score_spread":0.3454470313682256,"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."}}