That would be one way to implement the system in a DB-centric way. However we believe that intermediate application code is pretty critical to any GraphQL implementation.

We model mutations as a function that can do arbitrary business logic followed by a client-specified query where the client can query data it knows must be updated in order for the view to reflection the state changes the mutation caused.