{"id":"W4378376220","doi":"10.5334/pme.925","title":"Quality of Narratives in Assessment: Piloting a List of Evidence-Based Quality Indicators","year":2023,"lang":"en","type":"article","venue":"Perspectives on Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Sherbrooke","funders":"Université de Sherbrooke","keywords":"Checklist; Narrative; Quality (philosophy); Context (archaeology); Inter-rater reliability; Psychology; Medical education; Medicine; Computer science; History; Developmental psychology; Linguistics","routes":{"ca_aff":true,"ca_fund":true,"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":["metaresearch"],"category_scores_codex":[0.6198888,0.001934519,0.002707196,0.01333332,0.004411091,0.007383128,0.00463225,0.003147517,0.002325662],"category_scores_gemma":[0.7176949,0.002041094,0.00436109,0.0101168,0.004966883,0.01620789,0.01324022,0.005148997,0.000932951],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01587209,"about_ca_system_score_gemma":0.05890993,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00307386,"about_ca_topic_score_gemma":0.005162782,"domain_scores_codex":[0.4455346,0.3297824,0.156073,0.007746924,0.05609387,0.004769144],"domain_scores_gemma":[0.1967882,0.4619424,0.07279336,0.04774614,0.2117098,0.009020205],"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":[0.001643813,0.002040866,0.07851243,0.04328125,0.0006624167,0.0006666572,0.100329,0.002496106,0.004590416,0.00845704,0.01403951,0.7432806],"study_design_scores_gemma":[0.007344738,0.02087096,0.234439,0.1623708,0.004479397,0.003213436,0.1442664,0.03302803,0.05823055,0.06133636,0.2678771,0.002543444],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2879827,0.009660807,0.3645153,0.03304252,0.002587382,0.2837284,0.003055629,0.001763123,0.01366421],"genre_scores_gemma":[0.2020584,0.002821295,0.6665887,0.001528505,0.0001634045,0.1251863,0.001027322,0.0001889031,0.0004371099],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.3801112,"threshold_uncertainty_score":0.4687449,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1384609392390447,"score_gpt":0.5218332806893698,"score_spread":0.3833723414503251,"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."}}