{"id":"W1997246891","doi":"10.1002/lary.23437","title":"Multi‐institutional evaluation of a sinus surgery checklist","year":2012,"lang":"en","type":"article","venue":"The Laryngoscope","topic":"Sinusitis and nasal conditions","field":"Medicine","cited_by":21,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"McNemar's test; Checklist; Test (biology); Observational study; Medicine; Task (project management); Psychology; Statistics; Mathematics; Cognitive psychology; Internal medicine","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":[],"consensus_categories":[],"category_scores_codex":[0.001281288,0.0000740967,0.0001677666,0.00005991604,0.0001073175,0.000004999399,0.00004245445,0.00004125417,0.000892439],"category_scores_gemma":[0.0003427758,0.00004561172,0.00009441585,0.0001833787,0.0001229786,0.0001142489,0.00002106344,0.00008709494,0.0001186445],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005900265,"about_ca_system_score_gemma":0.0002179043,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008898453,"about_ca_topic_score_gemma":0.00001285812,"domain_scores_codex":[0.9991026,0.00009285173,0.0001768596,0.00007286247,0.0003979997,0.0001568372],"domain_scores_gemma":[0.9993545,0.0001207196,0.00006508972,0.0001929617,0.0001983205,0.00006836971],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0005729123,0.003981742,0.6873417,0.0003483534,0.0007899114,0.00004478939,0.002799819,0.000102295,0.1883261,0.01027611,0.07707278,0.02834349],"study_design_scores_gemma":[0.000724123,0.0000223045,0.9897509,0.00008668393,0.0002912893,0.0001415673,0.00004617381,0.000677307,0.005048411,0.00002377712,0.003122129,0.00006535161],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.990149,0.001702342,0.0001189037,0.0003778471,0.0003030667,0.0002822744,0.00002449103,0.00002272331,0.007019347],"genre_scores_gemma":[0.9984933,0.00001252037,0.0003057207,0.000390904,0.0002841497,0.00002905995,0.00008889085,0.000008172241,0.0003872617],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3024092,"threshold_uncertainty_score":0.9771582,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08853174877171685,"score_gpt":0.3432825347763225,"score_spread":0.2547507860046057,"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."}}