The graph exists as a list of edges (relationships) in the shared datastore. Interpretation of that graph to make permissions decisions is done by each individual SpiceDB node, and any node can resolve any graph traversal problem or sub-problem.