{"id":"W2962136257","doi":"10.1111/medu.13924","title":"Assessors’ interpretations of narrative data on communication skills in a summative OSCE","year":2019,"lang":"en","type":"article","venue":"Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Summative assessment; Narrative; Objective structured clinical examination; Psychology; Medical education; Communication skills; Medicine; Mathematics education; Formative assessment; Linguistics; Philosophy","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1415291,0.001021083,0.0007981555,0.004536771,0.001841179,0.003533579,0.001362334,0.001102545,0.001874348],"category_scores_gemma":[0.3862905,0.0005489042,0.001050666,0.0017165,0.002489545,0.002749367,0.005890496,0.001619927,0.000946819],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001360425,"about_ca_system_score_gemma":0.002369838,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006949529,"about_ca_topic_score_gemma":0.002074588,"domain_scores_codex":[0.8519864,0.1019594,0.02319517,0.003845334,0.01743764,0.001576068],"domain_scores_gemma":[0.5934706,0.2354462,0.04504391,0.02263194,0.1003029,0.003104532],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.002360861,0.0004681516,0.1800484,0.002801983,0.0005229576,0.001753755,0.392812,0.002173683,0.03045304,0.002807407,0.007817702,0.37598],"study_design_scores_gemma":[0.0003931031,0.005036007,0.3809229,0.007286473,0.0009783225,0.00486738,0.3864715,0.02798045,0.08074071,0.01836377,0.08595794,0.001001387],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8859233,0.001073454,0.09414065,0.00186817,0.0004722723,0.002489001,0.0005864539,0.000447571,0.0129991],"genre_scores_gemma":[0.9118218,0.0006763341,0.08111519,0.0006882407,0.0002043772,0.002828114,0.0003883705,0.0001471383,0.002130435],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1415291,"threshold_uncertainty_score":0.7484862,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01639054828651594,"score_gpt":0.4119734474680303,"score_spread":0.3955828991815144,"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."}}