unimplemented notes:
- regarding git histories, at least one of guix itself !!
- /build/ is ratelimited, so the narinfos or derivation paths of the first joblists would ee quite useful. enumerating /build/details seems only helpful if the derivation cannot be found anywhere 
- on python, diskcache-rs advertises transparent compression 
Oh and the response metadata is diagnostically and forensically incredible but unnecessary and using a lot of space and time

my cachedb is over 4.6GB of metadata and still growing. I randomly sorted the builds and ran a few processes in parallel so the list should still be quite useful. "quick naive script" solution to this needs a lot of time*space

note "quick naive" might apply to almost all source i have seen since 2014, notably pytorch :s 

but guix15.py should be written in guile to reuse existing work and have separate threads for build and nar* data

It is hard to think about doing this right because the right solution is a p2p computer, which became politically controversial. We got the parts we just gotta overcome what we've been through and put them together, although now they're all quick naive solutions too

the last time I did this I sorted the output by use which was very usef