Similarly to the size partitioning store, there's cases where you'd like to change the backend store dependant on the operation direction. For example, a dev AC and prod AC, you can flow prod to Dev but not vice versa. Another example is worker nodes, currently they store their results in the local CAS for no reason, the local filesystem store should only be populated on read operations, not write.

This would be a trivial store which would act like a diode when paired with a noop store.