{"id":"W4417276784","doi":"10.63564/jnep.v15n12p29","title":"Evaluating a zero-shot GenAI assistant for clinical record writing in nursing education","year":2025,"lang":"","type":"article","venue":"Journal of Nursing Education and Practice","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Formative assessment; Practicum; Nurse education; Reliability (semiconductor); Scale (ratio); Nursing Outcomes Classification; Summative assessment; Clinical judgment","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"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":[0.05042877,0.0009065989,0.0005155956,0.0009429838,0.000670461,0.001938918,0.001966509,0.001008417,0.003949513],"category_scores_gemma":[0.1529422,0.0003942532,0.0005803989,0.0005495224,0.001159092,0.001815885,0.003194116,0.0009567934,0.001296869],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001294485,"about_ca_system_score_gemma":0.003066535,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003966472,"about_ca_topic_score_gemma":0.00108099,"domain_scores_codex":[0.9456853,0.04373138,0.002697392,0.002094551,0.00510596,0.0006854036],"domain_scores_gemma":[0.8310565,0.135038,0.007657329,0.008358585,0.01422898,0.003660657],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.003930084,0.003362894,0.02237311,0.003491266,0.0000876256,0.000632619,0.03640235,0.002234134,0.03380988,0.00110097,0.003626002,0.8889491],"study_design_scores_gemma":[0.007864507,0.1613311,0.2047696,0.007383076,0.001242235,0.012958,0.08027606,0.07398509,0.2285601,0.007456913,0.2125521,0.00162141],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8449634,0.0006845312,0.1352993,0.0007515412,0.0005208366,0.006913167,0.0002137314,0.003502711,0.007150779],"genre_scores_gemma":[0.5102185,0.0004787574,0.48004,0.0006753139,0.00013651,0.004345275,0.0002747084,0.0002333713,0.003597594],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9495713,"threshold_uncertainty_score":0.266696,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6511088525085303,"score_gpt":0.7018804436322489,"score_spread":0.05077159112371854,"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."}}