This is a great point. Solutions like Graft which involve syncing data to the edge work poorly when the dataset size is too large and not partitioned enough to be consistency partial.

This is why Graft isn't just focused on client sync. By expanding the focus to serverless functions and the edge, Graft is able to run the exact same workload on the exact same snapshot (which it can validate trivially due to its consistency guarantees) anywhere. This means that a client's workload can be trivially moved to the edge where there may be more resources, a better network connection, or existing cached state.