The paper's concept store holds each concept node with four fields: a name, a textual description, a hierarchy level, and a set of prerequisite edges to other concept nodes. The hierarchy level locates the concept in the prerequisite DAG, and the prerequisite edges define the graph structure that downstream traversal and diffusion policies operate on. Encoding inputs (query texts and concept texts) are built from the name + textual description fields, while the hierarchy level and prerequisite edges are consumed by the graph-aware retrieval step rather than by the dense encoder.