{"id":"W7106795407","doi":"10.48448/yr05-m123","title":"Hard Negatives, Hard Lessons: Revisiting Training Data Quality for Robust Information Retrieval with LLMs","year":2025,"lang":"","type":"other","venue":"Open MIND","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Annotation; False positive paradox; Focus (optics); Reliability (semiconductor); Set (abstract data type); Training set; Quality (philosophy); Code (set theory)","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":[0.01384118,0.003138915,0.002069867,0.002663965,0.001640703,0.002807219,0.004265947,0.002479093,0.003445934],"category_scores_gemma":[0.06426677,0.001051459,0.001198763,0.001506841,0.002548018,0.007389784,0.004086344,0.00466177,0.002464556],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001975082,"about_ca_system_score_gemma":0.00261106,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009616189,"about_ca_topic_score_gemma":0.01632077,"domain_scores_codex":[0.9926697,0.003776416,0.0005256505,0.001602943,0.0009956751,0.0004296577],"domain_scores_gemma":[0.9758466,0.01216551,0.001056565,0.006816136,0.003598371,0.0005168338],"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":[0.002356434,0.0009660692,0.02356692,0.001236974,0.0003855161,0.000359335,0.001102337,0.1287893,0.0394381,0.005253359,0.04567204,0.7508736],"study_design_scores_gemma":[0.0003363162,0.0007802336,0.004141024,0.0001987327,0.0001789869,0.0002651699,0.000440328,0.9286812,0.03926956,0.01451633,0.0110662,0.0001259961],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.2598452,0.005516581,0.6439851,0.005080765,0.0009337479,0.001085077,0.003019178,0.07453328,0.006001023],"genre_scores_gemma":[0.7042503,0.0004194687,0.2830186,0.001394139,0.0002604221,0.0004933854,0.005136386,0.001818142,0.003209199],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01384118,"threshold_uncertainty_score":0.07320005,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3651665725194361,"score_gpt":0.4249903887196824,"score_spread":0.05982381620024635,"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."}}