paper_type.json (250B)
1 { 2 "paper_type": "empirical", 3 "reason": "Introduces a method (Contrastive Rubric Generation) and validates it experimentally across 8 reward-modeling benchmarks, reporting quantitative improvements (8.4pp) and downstream task performance gains." 4 }