{"id":"W2912476504","doi":"10.1111/medu.13801","title":"Accuracy of self‐monitoring: does experience, ability or case difficulty matter?","year":2019,"lang":"en","type":"article","venue":"Medical Education","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":30,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Seniority; Aptitude; Context (archaeology); Self-monitoring; Psychology; Psychological intervention; Confidence interval; Social psychology; Applied psychology; Medicine; Developmental psychology; Psychiatry","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","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0003131354,0.0001016438,0.0002761592,0.00003919973,0.00002950851,0.000009437282,0.00009252934,0.0001447219,0.005653044],"category_scores_gemma":[0.06638946,0.00005675446,0.00007380129,0.000167859,0.0000991546,0.00005463172,0.0000358999,0.0001961162,0.0001958397],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006476766,"about_ca_system_score_gemma":0.00112251,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003184282,"about_ca_topic_score_gemma":0.000005360264,"domain_scores_codex":[0.9986254,0.00005099823,0.000419208,0.0002656397,0.0004726969,0.0001660516],"domain_scores_gemma":[0.9939387,0.005019915,0.0001295723,0.0004274375,0.0001646143,0.0003197315],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001043477,0.002345562,0.947831,0.0001555902,0.00001752216,0.00005773813,0.002215134,1.53798e-7,0.0001767013,0.00002528428,0.004676376,0.04239457],"study_design_scores_gemma":[0.001427362,0.00029145,0.9822521,0.0020669,0.00008186716,0.0008041789,0.003681375,0.00009261143,0.001941903,0.00005375811,0.007150698,0.0001558121],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9932438,0.00007374323,0.00001170356,0.003458557,0.001923456,0.0003059684,0.000002277314,0.00004563482,0.0009348813],"genre_scores_gemma":[0.9969398,0.00005899835,0.0006205069,0.000873744,0.0004329318,0.00005009327,0.00001619541,0.000009431602,0.0009983034],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.06607632,"threshold_uncertainty_score":0.9952559,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01824775107992284,"score_gpt":0.3958603817714509,"score_spread":0.3776126306915281,"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."}}