{"id":"W7080182188","doi":"","title":"Write on Paper, Wrong in Practice: Why LLMs Still Struggle with Writing Clinical Notes","year":2025,"lang":"en","type":"article","venue":"ArXiv.org","topic":"Geochemistry and Geologic Mapping","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Sociotechnical system; Documentation; Bespoke; Technical writing; Task (project management); Qualitative research","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":[0.03340689,0.0004620223,0.0005388059,0.001989993,0.01130313,0.01154335,0.002888883,0.004283458,0.007592194],"category_scores_gemma":[0.2343457,0.001205747,0.0005595496,0.002019414,0.01292231,0.01009108,0.009180235,0.005523159,0.003883058],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01174443,"about_ca_system_score_gemma":0.02851221,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02985459,"about_ca_topic_score_gemma":0.0511629,"domain_scores_codex":[0.9490809,0.02571509,0.004721346,0.00354181,0.01297457,0.003966346],"domain_scores_gemma":[0.8166743,0.0999135,0.02708497,0.01034562,0.02743363,0.01854797],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.0001598903,0.0001230108,0.04014547,0.0006225975,0.00004289487,0.006346675,0.7588822,0.0002311576,0.001885111,0.006735339,0.0589909,0.1258347],"study_design_scores_gemma":[0.00005560139,0.0001651516,0.0167565,0.001840351,0.00003487159,0.008361808,0.746607,0.001255936,0.001218861,0.009883537,0.2136322,0.000188194],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6873412,0.007058016,0.0298305,0.2267014,0.002065778,0.0003460253,0.0003429264,0.001297334,0.04501672],"genre_scores_gemma":[0.9502207,0.00255421,0.01325268,0.0192245,0.0002637914,0.0001979333,0.000150052,0.0006305853,0.01350555],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9665931,"threshold_uncertainty_score":0.1766746,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04041609918256068,"score_gpt":0.320519508984694,"score_spread":0.2801034098021333,"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."}}