{"id":"W2201706766","doi":"10.1016/j.spinee.2015.06.043","title":"Assessing calibration in an external validation study","year":2015,"lang":"en","type":"letter","venue":"The Spine Journal","topic":"Spine and Intervertebral Disc Pathology","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":false,"ca_institutions":"McMaster University; Population Health Research Institute","funders":"","keywords":"Medicine; Calibration; Medical physics; Statistics","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":["research_integrity"],"consensus_categories":[],"category_scores_codex":[0.001133606,0.0002237539,0.0004346955,0.0003008788,0.00008821652,0.0002766149,0.0002501248,0.0002811078,0.0003717845],"category_scores_gemma":[0.0001265093,0.0001350937,0.0001140386,0.0001545052,0.00004163677,0.0006952245,0.00005118733,0.002805095,0.0000307962],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001803972,"about_ca_system_score_gemma":0.000262669,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004273383,"about_ca_topic_score_gemma":0.00001589882,"domain_scores_codex":[0.9977951,0.0005647752,0.0005955432,0.0002182831,0.0005515213,0.0002747444],"domain_scores_gemma":[0.9990257,0.000001426601,0.0003517962,0.0003468591,0.0001914712,0.00008275817],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.0000118054,0.000655738,0.004197208,0.00005587469,0.00009954735,0.005720966,0.001286752,0.00004260022,0.01038532,0.000005191507,0.970993,0.006546038],"study_design_scores_gemma":[0.005095687,0.005761282,0.9679052,0.0009295105,0.000616827,0.008889234,0.001605734,0.002535217,0.0003199249,0.0008771201,0.004970877,0.0004933877],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7792138,0.0002125654,0.001939515,0.2163039,0.0009749403,0.0004529031,0.000001702807,0.0000270968,0.0008735718],"genre_scores_gemma":[0.5317683,0.00003092295,0.001468443,0.4333845,0.02878007,0.00003247746,0.0002603198,0.0001316427,0.004143334],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9660221,"threshold_uncertainty_score":0.9994954,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1169164735841542,"score_gpt":0.4061365181796991,"score_spread":0.289220044595545,"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."}}