{"id":"W2542250253","doi":"10.1145/3012004","title":"The Challenge of Test Data Quality in Data Processing","year":2017,"lang":"en","type":"article","venue":"Journal of Data and Information Quality","topic":"Data Quality and Management","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"Natural Sciences and Engineering Research Council of Canada; Vienna Science and Technology Fund","keywords":"Citation; Test (biology); Computer science; Data quality; Library science; Quality (philosophy); World Wide Web; Data science; Engineering; Operations management","routes":{"ca_aff":true,"ca_fund":true,"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":[0.403006,0.001740443,0.004965005,0.006536602,0.003739403,0.0256544,0.008660882,0.00834165,0.005173991],"category_scores_gemma":[0.7120219,0.001738899,0.002399988,0.009797222,0.03329949,0.02914498,0.01573902,0.02005999,0.002570867],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01014484,"about_ca_system_score_gemma":0.01724264,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01141704,"about_ca_topic_score_gemma":0.004470924,"domain_scores_codex":[0.5772017,0.294138,0.02283281,0.02244843,0.08060777,0.002771234],"domain_scores_gemma":[0.1108718,0.7254237,0.02404773,0.08097336,0.05375158,0.004931756],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0008281323,0.0002890649,0.02984825,0.003238572,0.001186297,0.0002419938,0.003304286,0.01063016,0.00141033,0.3071983,0.08858178,0.5532428],"study_design_scores_gemma":[0.0002357011,0.0004190794,0.006533588,0.00178713,0.0001940445,0.0003913617,0.001425432,0.04384301,0.00183278,0.8797889,0.06337265,0.0001762334],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01787135,0.04738326,0.5098339,0.3895094,0.005346678,0.0007221657,0.002334238,0.00356174,0.02343708],"genre_scores_gemma":[0.5030618,0.0178712,0.404368,0.05028671,0.01267439,0.001343924,0.002777225,0.002384947,0.005231882],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.403006,"threshold_uncertainty_score":0.7361999,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6182808189797534,"score_gpt":0.5581897740451959,"score_spread":0.06009104493455752,"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."}}