{"id":"W4394933359","doi":"10.1371/journal.pone.0302299","title":"Assessment of confidence in medical writing: Development and validation of the first trustworthy measurement tool","year":2024,"lang":"en","type":"article","venue":"PLoS ONE","topic":"Academic Writing and Publishing","field":"Arts and Humanities","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"York University; McMaster University; Impact","funders":"","keywords":"Cronbach's alpha; Confidence interval; Reliability (semiconductor); Popularity; Computer science; Analytics; MEDLINE; Internal consistency; Psychology; Medical physics; Medicine; Psychometrics; Data science; Clinical psychology; Social 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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07172617,0.000574787,0.00111908,0.005298624,0.0009791519,0.002510011,0.001613679,0.001575703,0.001213949],"category_scores_gemma":[0.1737226,0.0005932135,0.001920854,0.002349393,0.001384091,0.003150657,0.003624674,0.00192016,0.0005391555],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001274123,"about_ca_system_score_gemma":0.005341174,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005991234,"about_ca_topic_score_gemma":0.000868717,"domain_scores_codex":[0.9482465,0.02104565,0.01101085,0.00175218,0.01700791,0.0009369107],"domain_scores_gemma":[0.800158,0.1142354,0.02234024,0.01062643,0.04901589,0.003624002],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0007222635,0.001730848,0.6769581,0.002355712,0.0003796722,0.0002872596,0.01427699,0.001067105,0.005613604,0.002224936,0.00296656,0.291417],"study_design_scores_gemma":[0.0004315314,0.006896863,0.9306616,0.002881521,0.0004926486,0.001472255,0.0110107,0.01489494,0.01245002,0.004189319,0.01434968,0.0002688755],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9066413,0.001321392,0.07035425,0.001509491,0.0002860762,0.01041385,0.001015551,0.000218795,0.008239265],"genre_scores_gemma":[0.8432294,0.0005323678,0.144581,0.0002462927,0.00008079968,0.009981131,0.0007193607,0.00004464941,0.0005849401],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9282738,"threshold_uncertainty_score":0.3793288,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1011174596008173,"score_gpt":0.2602164426142926,"score_spread":0.1590989830134752,"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."}}