{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.006087221,0.0001627687,0.0005279248,0.001283947,0.0000678492,0.00000794818,0.0002153859,0.0001946053,0.0007119806],"category_scores_gemma":[0.04386181,0.0001337753,0.0001095446,0.003710671,0.0006153768,0.0001506103,0.00003248677,0.0006298056,0.00001002295],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000562562,"about_ca_system_score_gemma":0.004923144,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004292309,"about_ca_topic_score_gemma":0.00002456243,"domain_scores_codex":[0.9957399,0.0006324137,0.00119813,0.0004382313,0.001750786,0.0002405779],"domain_scores_gemma":[0.9969812,0.001054833,0.0006800493,0.0004909764,0.000650684,0.0001422279],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006796799,0.01043807,0.6818501,0.002399242,0.0001104297,0.00000288172,0.075131,0.00002518502,0.006344714,0.08890428,0.003387202,0.1307272],"study_design_scores_gemma":[0.0009206317,0.0003173367,0.8810538,0.002138262,0.00001341244,9.796095e-7,0.1130524,0.0004431611,0.0007827118,0.0009741041,0.0001729229,0.0001302432],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9660275,0.0002059188,0.0005882948,0.02961745,0.0004213405,0.0005091526,0.000003970999,0.00006194649,0.002564407],"genre_scores_gemma":[0.995223,0.0001265892,0.003157331,0.0009464283,0.0002037843,0.0001292137,0.00007574179,0.00001809432,0.0001197915],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1992037,"threshold_uncertainty_score":0.9641922,"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."}}