The Interpretive Graph Overlay: Modeling Case Law as Normative Events for Deterministic Legal Retrieval

Legal retrieval systems may recover the text of a statute as it stood on a given date while missing the judicial interpretation that then governed its application. This paper introduces the Interpretive Graph Overlay, an extension of the Structure-Aware Temporal Graph RAG (SAT-Graph) framework that represents adjudication as an event acting on a normative target. Judicial decisions remain versioned documentary objects, while curated adjudicative Actions produce temporally scoped Interpretive Versions carrying the relevant ratio, holding, authority, and scope.The overlay reuses four application-level primitives—Item, Version, Action, and TextUnit—through a shared event topology rather than introducing a separate case-law ontology. Where a statutory provision exists, its textual Version may coexist with one or more interpretive overlays; in judge-made domains, successive Interpretive Versions may instead represent authoritative articulations of a doctrine or the authority state of a scoped proposition. Retrieval is deterministic only in the bounded sense adopted here: once canonical identifiers, temporal parameters, and curated graph assertions are fixed, structural and temporal traversals are reproducible; legal applicability remains the result of an explicit jurisdiction-specific authority-resolution policy.Four manually encoded case studies exercise interpretive mutation, constitutional dialogue, concurrent panel interpretations and institutional unification, and the evolution of the English duty-recognition methodology from Donoghue v. Stevenson to Robinson. The public artifact includes the JSON encodings, a browser viewer driven by a shared temporal-resolution engine, structural validation, and executable behavioral tests for fixtures, interval boundaries, input normalization, and equivalent timestamp representations.Expanded author version of a two-page demonstration paper accepted at the 21st International Conference on Artificial Intelligence and Law (ICAIL 2026). The ACM Version of Record, once published, should be cited by DOI.