Patrick Hochstenbach, Jos De Roo, Wout Slabbinck, Beatriz Esteves, and Pieter Colpaert: "Building Thoroughly Tested Semantic Rule Engines in the Age of GenAI", SAGE 2026: The International Workshop on Semantic Architectures for Governance and Explainability, co-located with SEMANTiCS 2026 (2026).

Abstract

Data sovereignty starts before data leaves a user's device. As data processing moves toward the edge, future Web applications will need to evaluate rules, policies, and explanations locally, rather than delegating every decision to server-side infrastructure. At the same time, data will increasingly flow across organizations, applications, and regulatory contexts. We argue that such flows must be interoperable across ecosystems, legally compliant with applicable policies, and trustworthy to end-users. This requires a rule language that independent engines can implement consistently. In this paper, we introduce 1k+ Notation3 compliance tests for validating implementations against the Notation3 specification, developed in parallel with Eyeling: a JavaScript, browser-deployable reference implementation for interoperable in-browser trust engines. Applied to cwm, EYE, and jen3, the suite reveals conformance gaps with the published specification, while Eyeling reaches full conformance and shows that LLM-driven synthesis can bootstrap spec-compliant rule engines from a specification and conformance suite. These results suggest that in-browser semantic rule engines can provide a practical foundation for evaluating interoperability, legal compliance, and trust where data is accessed and used.