Right, I agree that for the update aggregation e2+e32 is better than e2+e3 (but still lossy). And the thing that p2:p3 affects is the weighting in the average—so if e2=e3 then the ps don’t matter! (which is a possible answer to your question of “how much aggregation/disaggregation can you do?”)
But yeah if e2 is very different from e3 then I don’t think there’s any way around it, because the effective ei could be one or the other depending on what the pi are.
Right, I agree that for the update aggregation e2+e32 is better than e2+e3 (but still lossy). And the thing that p2:p3 affects is the weighting in the average—so if e2=e3 then the ps don’t matter! (which is a possible answer to your question of “how much aggregation/disaggregation can you do?”)
But yeah if e2 is very different from e3 then I don’t think there’s any way around it, because the effective ei could be one or the other depending on what the pi are.