Does this include async updates as well? I'd be interested in getting something like core.async / reactor in place to stream updates back to the client

If you feel like writing that, go for it. Just getting the infrastructure in place to accept updates to a task is the first step, though.

What should we call these entries?

Should these task updates be indexed like claims?

Should that index be handled by the client or the server?

Should they be last-write-wins for a given index?

Oh, looks like I forgot to write this. Yeah, the term is "log"; there should be one log per claim, and the log for that claim should be a vector, merged pairwise, expanding to the size of the larger vector. Client API should be (update! task-id claim-idx log-idx message-byte-array); the client keeps track of an auto-incrementing index for the claim, starting at 0.

Thanks. That was just the guidance I was looking for.