How do you choose between cache tags and path revalidation after a mutation?

I document the consistency promise per workflow: the editor may require immediate read-your-own-writes, while public lists can tolerate stale-while-revalidate. Multi-step mutations publish invalidation from one committed application boundary rather than scattering calls through components. I monitor cache age, invalidation frequency, refresh latency, and downstream load because an overly broad purge can create a thundering herd. Tests prove that the changed record updates, unrelated cached content remains reusable, and failure before commit emits no invalidation.