{"id":"W4409512100","doi":"10.1186/s12909-025-07134-6","title":"Quantify difference between physicians and medical students in clinical reasoning: evidence from eye-tracking","year":2025,"lang":"en","type":"article","venue":"BMC Medical Education","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"Inner Mongolia Medical University","keywords":"Medical education; Tracking (education); Eye tracking; MEDLINE; Educational measurement; Medicine; Psychology; Curriculum; Computer science; Pedagogy; Artificial intelligence","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"],"consensus_categories":[],"category_scores_codex":[0.003108675,0.0001912008,0.000696622,0.000121646,0.00006405242,0.00006213703,0.0004379936,0.0005336099,0.0002467429],"category_scores_gemma":[0.4494343,0.0001630913,0.0001168467,0.0003830983,0.0003650393,0.00007035297,0.000220824,0.0009796302,0.0000336265],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000876513,"about_ca_system_score_gemma":0.004322512,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00117304,"about_ca_topic_score_gemma":0.0005165437,"domain_scores_codex":[0.9957588,0.000440384,0.001095494,0.0006760902,0.001714094,0.0003151925],"domain_scores_gemma":[0.9369374,0.0607038,0.0002797685,0.0006967025,0.0002086231,0.001173745],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00006927302,0.0009275749,0.7983242,0.00005136813,0.00002899835,0.00001127702,0.0001885051,9.546625e-8,0.000003514455,0.0001584523,0.001238717,0.198998],"study_design_scores_gemma":[0.00148754,0.0001085815,0.9576137,0.03873967,0.0001513408,0.000002497306,0.0002749485,0.0007357484,0.00001692933,0.0005765086,0.0001633641,0.0001291751],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9835712,0.00189497,0.002265128,0.01051696,0.001163212,0.0003194865,0.000002333334,0.00005824046,0.0002084182],"genre_scores_gemma":[0.9906021,0.001733882,0.001304751,0.004854764,0.001007577,0.00004473851,0.00006753114,0.00001422036,0.0003704467],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4463256,"threshold_uncertainty_score":0.766795,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06997983774980288,"score_gpt":0.5007527503252808,"score_spread":0.4307729125754779,"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."}}