MHGB: A Multi-Hop Knowledge Graph Benchmark for Legal Reasoning Evaluation

Initial public release accompanying the KDD 2027 Datasets & Benchmarks submission. Typed statutory knowledge graph: 5,519 article nodes, 8,760 edges of five relation types across ten Russian federal codes Statutory corpus: 5,570 parsed articles with validity metadata Public task subset: 120 instances (60 fabulas × 2 modes), stratified 10 per cell of the 4×3 matrix Per-model scores for all 12 evaluated models over the full 300-fabula benchmark Full evaluation pipeline, judge prompts, and Streamlit graph explorer Code under Apache-2.0, derived data under CC BY-4.0, statutory text in the public domain.