{"id":"W6944970400","doi":"10.2312/evs.20231036","title":"RiskFix: Supporting Expert Validation of Predictive Timeseries Models in High-Intensity Settings","year":2023,"lang":"en","type":"article","venue":"Open MIND","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Hospital for Sick Children; University of Toronto","funders":"","keywords":"Workflow; Domain (mathematical analysis); Subject-matter expert; Bottleneck; Software; Domain knowledge; Ground truth","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02117406,0.003724498,0.001405745,0.001818979,0.0006807052,0.003514254,0.006333685,0.002895152,0.01101999],"category_scores_gemma":[0.1127125,0.002045632,0.002078277,0.0007189757,0.001566603,0.005945915,0.006714604,0.003395956,0.004535447],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001193175,"about_ca_system_score_gemma":0.003091306,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00452815,"about_ca_topic_score_gemma":0.00344588,"domain_scores_codex":[0.9891382,0.004160134,0.001343395,0.002387963,0.002390798,0.0005795292],"domain_scores_gemma":[0.9197419,0.06302045,0.003073253,0.00787879,0.005082374,0.001203263],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.007888578,0.001765811,0.03715476,0.002771375,0.001179587,0.003222947,0.01328247,0.2489557,0.02842225,0.03172148,0.09898259,0.5246524],"study_design_scores_gemma":[0.0005798846,0.0003485152,0.003947468,0.0003728452,0.0001059385,0.0005960181,0.000303889,0.9065016,0.02659513,0.02931507,0.03111969,0.0002139776],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02755457,0.0002078662,0.7516708,0.0004250904,0.0001137517,0.0004900382,0.001306901,0.2159782,0.002252792],"genre_scores_gemma":[0.3448174,0.0003121049,0.6171077,0.0006873505,0.0001096916,0.00142685,0.006651964,0.02536866,0.003518192],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.02117406,"threshold_uncertainty_score":0.1119804,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2161478756960769,"score_gpt":0.455522485278761,"score_spread":0.2393746095826841,"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."}}