This paper presents an ontology-based graph RAG framework to address the challenges faced by Retrieval-Augmented Generation (RAG) systems in the legal domain. To address the problem that existing simple text retrieval methods fail to account for the hierarchical, temporal, and causal structures of law, resulting in anachronistic and unreliable answers, we build a knowledge graph based on a formal model inspired by the LRMoo model. We distinguish between abstract legal documents (Works) and their versions (Expressions), efficiently aggregate temporal states, and reuse unchanged versions of components. Furthermore, we explicitly designate legislative events as primary action nodes, making causal relationships explicit and queryable. This structural foundation allows us to apply a planner-driven, integrated query strategy to deterministically address complex requests such as (i) point-in-time retrieval, (ii) hierarchical impact analysis, and (iii) auditable provenance reconstruction. A case study on the Brazilian Constitution demonstrates that the proposed approach provides a verifiable and temporally accurate foundation for LLMs, enabling high-dimensional analysis capabilities and significantly reducing the risk of factual errors. Ultimately, it provides a practical framework for building more trustworthy and explainable legal AI systems.
