You can scale writes with parallelism. Each machine manages a selection of writes.
but there are surely other ways of parallelising writes, that I want ideas for.
How you get them back to the source is an exercise that I want ideas for.
No child categories.