{"id":"W2602485715","doi":"10.1016/j.riem.2017.01.116","title":"Trustworthiness of field notes in Family Medicine residency training: does field note content provide evidence to support the validity of our decisions about residents’ competence?","year":2017,"lang":"en","type":"article","venue":"Investigación en Educación Médica","topic":"Innovations in Medical Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Calgary","funders":"","keywords":"Competence (human resources); Trustworthiness; Psychology; Residency training; Medical education; Field training; Field (mathematics); Content validity; Applied psychology; Medicine; Social psychology; Clinical psychology; Psychometrics; Mathematics; Continuing education","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":[0.003716438,0.0003494819,0.0009737398,0.00053352,0.0003630678,0.00004389579,0.001397701,0.0003178953,0.0001823597],"category_scores_gemma":[0.157568,0.0002198707,0.0001510545,0.0007055704,0.0007744587,0.0003654299,0.0002789961,0.0009416242,0.00001512345],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001295437,"about_ca_system_score_gemma":0.002407972,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00665303,"about_ca_topic_score_gemma":0.0008887768,"domain_scores_codex":[0.9954717,0.0002832264,0.001749753,0.0006129258,0.001362803,0.0005196367],"domain_scores_gemma":[0.9919453,0.003649947,0.0009745042,0.002185081,0.0009042251,0.0003408908],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006754365,0.0006158783,0.7857345,0.0007010977,0.000112054,0.00009786178,0.02509305,0.00002227349,0.03283437,0.006073725,0.04870229,0.09933744],"study_design_scores_gemma":[0.001210771,0.001548934,0.9483517,0.009382117,0.0001956459,0.00004666229,0.00789028,0.00008271117,0.02193162,0.007201009,0.001846571,0.0003120004],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7410603,0.000135683,0.000936065,0.2543051,0.001923671,0.0009996903,0.00000856651,0.00002848955,0.0006024551],"genre_scores_gemma":[0.979783,0.0002616878,0.009168936,0.009353899,0.0007961129,0.0001493088,0.00001334979,0.00003225875,0.0004414734],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2449512,"threshold_uncertainty_score":0.9999617,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.212364034716348,"score_gpt":0.4291633286376944,"score_spread":0.2167992939213464,"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."}}