In the auditable strict-parity evaluation of prerequisite-graph retrieval for RAG, the QASC benchmark is rebuilt as a directed science-fact graph with 16,444 nodes and 25,590 edges. This provides a much larger and structurally different substrate than curated prerequisite DAGs. Because the official QASC test split withholds evidence, retrieval results are reported only on the 926-question validation split. Graph descriptors include a mean depth of 2.3, 79.7% of nodes with depth $\geq 1$, a mean degree of 3.1, and a mean fact-text length of 8.9 tokens.