{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00134559,0.000056449,0.0001309149,0.00005121105,0.00007108342,0.0000778892,0.0001183993,0.00003973145,0.0002272701],"category_scores_gemma":[0.0002390627,0.00004098998,0.00001745463,0.00004292852,0.00009115072,0.0001362316,0.00005660434,0.0002062813,7.651242e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004348722,"about_ca_system_score_gemma":0.0001857451,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007155342,"about_ca_topic_score_gemma":0.0001942583,"domain_scores_codex":[0.9985842,0.00003575135,0.0003546905,0.0001099043,0.0008256151,0.00008990662],"domain_scores_gemma":[0.9996168,0.0001238409,0.00006587538,0.00007010808,0.0001033418,0.00002005877],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.00001398644,0.00103055,0.1334924,0.005200245,0.0004139599,0.000009648617,0.08067076,0.00001521873,0.001480621,0.7625028,0.001537598,0.0136322],"study_design_scores_gemma":[0.004661885,0.0005666366,0.3666468,0.1333916,0.001056498,0.00002227301,0.04717103,0.03259442,0.2146608,0.01680049,0.1800065,0.002420922],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9863764,0.0004678407,0.00002771734,0.002368836,0.00006555427,0.000131643,0.000004137628,0.00001935457,0.0105385],"genre_scores_gemma":[0.9994541,0.00003191697,0.0001666689,0.00008673316,0.0001228828,0.0000155146,0.000002127436,0.000005421634,0.0001146316],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7457023,"threshold_uncertainty_score":0.2488448,"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."}}