Deterministic Legal Retrieval: An Action API for Querying the SAT-Graph RAG
Abstract
The Structure-Aware Temporal Graph RAG (SAT-Graph RAG) addresses core limitations of standard Retrieval-Augmented Generation in the legal domain by providing a verifiable knowledge graph that models hierarchical structure, temporal evolution, and causal events of legal norms. However, a critical gap remains: how to reliably query this structured knowledge without sacrificing its deterministic properties. This paper introduces the SAT-Graph API, a formal query execution layer centered on canonical actions-atomic, composable, and auditable primitives that isolate probabilistic discovery from deterministic retrieval. These actions enable: (i) high-precision hybrid search; (ii) robust reference resolution; (iii) point-in-time version retrieval; and (iv) auditable causal tracing. We demonstrate how planner-guided agents can decompose complex queries into Directed Acyclic Graphs (DAGs) of these actions. This two-layer architecture transforms retrieval from an opaque black box to a transparent, auditable process, directly addressing Explainable AI (XAI) requirements for high-stakes domains.