The backend database representation (whether as JSON or Neo4j) relies on numeric node IDs to link text, equations, and parameters. As a result, loading the content into an LLM creates cognitive load and confusion when tracking numeric ides.
A simpler representation would be graphviz (with no numeric IDs)