In ArangoDB labels used by vertices and edges must be known in order to create a graph. As a result this information should be available at startup, i.e via configuration. For example shouldReadWriteModernToFileWithHelpers test expects a different schema from the standard graph than from the readGraph, the latter attempts to create vertices in a (default) vertex collection instead of using the people or software ones that were used for the former.
It would also help, for graph technologies that require schemas, to have a document that describe the schemas used/expected in the tests. I am collecting this information atm, but having to run failing tests to capture what label/edges I am missing for each test is a PITA.