{"id":"W2965597477","doi":"10.15530/urtec-2019-396","title":"Why Most Models Fail: The Importance of Blind Validation","year":2019,"lang":"en","type":"article","venue":"Proceedings of the 7th Unconventional Resources Technology Conference","topic":"Simulation Techniques and Applications","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":false,"ca_institutions":"Encana (Canada)","funders":"","keywords":"Computer science","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.139608,0.002055258,0.003815964,0.00331823,0.002451055,0.0102449,0.005609363,0.006839063,0.003252532],"category_scores_gemma":[0.5255932,0.001818289,0.002051088,0.001646325,0.01012001,0.01779219,0.007060064,0.008025148,0.00202769],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002616507,"about_ca_system_score_gemma":0.01085807,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006765267,"about_ca_topic_score_gemma":0.004741994,"domain_scores_codex":[0.8739692,0.08640976,0.006863716,0.007732661,0.02280837,0.002216335],"domain_scores_gemma":[0.4912266,0.375275,0.02192984,0.06810381,0.04060982,0.002854803],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.002961669,0.0008956114,0.06747524,0.003179958,0.003580413,0.0008306411,0.005408119,0.2258749,0.006617638,0.2499111,0.0540392,0.3792256],"study_design_scores_gemma":[0.0004442813,0.0003497124,0.006348615,0.001990085,0.0003407844,0.0002467388,0.001086948,0.3160867,0.006650085,0.6437507,0.02232504,0.0003802781],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"commentary","genre_scores_codex":[0.0781868,0.005130795,0.8361377,0.05645312,0.00262726,0.0005295888,0.000915479,0.004170082,0.01584918],"genre_scores_gemma":[0.7967653,0.001543387,0.1866007,0.009613533,0.0006718909,0.0004199068,0.0006036707,0.001704367,0.002077133],"genre_candidate":"commentary","genre_consensus":null,"teacher_disagreement_score":0.860392,"threshold_uncertainty_score":0.7383263,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07921849459820626,"score_gpt":0.3407682537686444,"score_spread":0.2615497591704381,"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."}}