Graph-aware retrieval beats flat semantic search for relational facts

Santiago (@svpino) argues on X that flat semantic-similarity retrieval fails when a query depends on relationships between facts.

He proposes graph-aware retrieval instead: storing facts as memory records, linking them with labeled relations such as "supersedes," "supports" or "contradicts," and expanding search results through those links. The report gives no benchmark results or other evidence for the claim that this approach performs better.