{"id":"W4413048167","doi":"10.1080/10401334.2025.2542859","title":"Still Not Clear? Exploring the Impact of Clarifying Assessment Items on Assessor Cognition in Medical Education","year":2025,"lang":"en","type":"article","venue":"Teaching and Learning in Medicine","topic":"Innovations in Medical Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Chinook Regional Hospital; London Health Sciences Centre; Western University","funders":"","keywords":"Formative assessment; Context (archaeology); Psychology; Cognition; Interpretation (philosophy); Perception; Task (project management); Applied psychology; Medical education; Function (biology); Social psychology; Medicine; Pedagogy; Computer science","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.007115678,0.0001476399,0.0003855,0.0007691876,0.0001421802,0.00001270789,0.0001031807,0.000138117,0.00008852341],"category_scores_gemma":[0.01716817,0.00009686527,0.00003659388,0.0006183276,0.0001719366,0.0000907889,0.00003296078,0.003889936,0.000001373547],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000310571,"about_ca_system_score_gemma":0.0007072263,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001210246,"about_ca_topic_score_gemma":0.00001606031,"domain_scores_codex":[0.9976263,0.0006670192,0.0006544082,0.0002538714,0.0005857911,0.000212617],"domain_scores_gemma":[0.9985344,0.0009290419,0.0001715651,0.0001995174,0.0001049588,0.00006055732],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00025243,0.0008559201,0.2111092,0.000307098,0.00006204467,0.000007872188,0.01207036,0.0006139821,0.001336268,0.004271839,0.001206189,0.7679068],"study_design_scores_gemma":[0.002075649,0.0007730477,0.9483952,0.007237107,0.00006126043,0.00001865673,0.01287528,0.02613851,0.00002987392,0.0002314297,0.002060354,0.0001036675],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9740244,0.000129635,0.0007170458,0.01747273,0.0004568454,0.0003138133,1.81139e-7,0.00003982858,0.006845551],"genre_scores_gemma":[0.996646,0.00008277096,0.0007165921,0.001883126,0.0002626941,0.00007856393,0.00004644583,0.00001441779,0.0002693491],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7678032,"threshold_uncertainty_score":0.9984081,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05036742462057025,"score_gpt":0.4468184040305332,"score_spread":0.396450979409963,"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."}}