The architecture isn't neuro-symbolic because nothing in it symbolically reasons. . What this article describes instead is a neural system with deterministic guardrails bolted to its inputs and outputs: a graph query strips overruled precedents before retrieval, Pydantic schemas typecheck the extraction, and a hardcoded function does calendar arithmetic for limitation periods. Every one of those is a filter or a calculator, not a reasoner. The actual legal determination — whether a term is a condition or a warranty, whether an exclusion clause is reasonable under UCTA, is still produced by the LLM, and no symbolic layer ever evaluates it, because there are no rules encoded to evaluate it against. The knowledge graph stores metadata about authorities (status, citation relationships); it does not represent what those authorities require. So the "zero-hallucination" claim holds only for the thin slice delegated to the calculators, while the reasoning path remains fully probabilistic.