what about a new type of pex that was “flattened”/unhydrated; you’d build this super quick with a tiny payload, quickly ship it across the wire to the datacenter, run a hydration operation in the DC (think: parallel fetch of all wheels from science-binaries by URL manifest) and then run. some experiments for us show this would drastically improve iteration speed, since we routinely pay the cost to fetch bins during resolve, pack ~gigs of wheels into files, then xfer them to a DC to do full integration testing.