Trace-Indexed GraphRAG: Precedent Retrieval over the Decision Category
Standard retrieval-augmented generation indexes documents by topic. For enterprise architecture work, the relevant retrieval object is not the document but the decision trace: a typed walk across the context, causal, and knowledge layers. Indexing the trace category rather than associated documents produces a retrieval primitive that respects compositional structure. This article specifies four retrieval patterns over the trace category, sketches a property-graph implementation and argues that selective embedding — over node content, not trace structure — is the right hybridisation point.