I can see that. Your roadmap makes sense to me. My [[wife|k0sh3k]] (without having seen your roadmap) picked out many of the same points as reasons she must wait to use it for her library staff. I appreciate how you must triage with the resources you have. Most of what I have to say isn't going to be as important to the non-poweruserbase you are trying to capture (and rightfully so!), so please take what I have to say with a grain of salt. You obviously know what you are doing. Reading through your account, I think you already know what I have to say here.
> Would desktop-only i2p support be useful to you?
Yes, and not just because I think it's important to contribute back to the network as a regular user. It would be especially nice with command-line tooling, and you can pickup UDP support. I've encountered plenty of failures on Tor (and you have too), and there are many I've encountered who I believe are rational in their desire not to be so reliant upon that network. I'd also be happy to use i2p on a mobile device just in case I had somewhat convenient control over when that occurred. If you provide logical accounts (where my decentralized identity can be wielded by multiple cooperating devices), then i2p becomes even better. Retroshare's use of both Tor and i2p have demonstrated the value of diversity here to my eyes as well. I'll add that something like multiplexing over multiple Tor or i2p tunnels could significantly improve throughput performance (a difficult place where you might be able deliver in some cases where no one has).
However, I think your focus on mobile is more important (and I say that with a deep-seated grudge against phones). You serve the poorest on the planet by targeting mobile first. That isn't to say that you don't have any use for desktop users to assist on this network, but I can see why i2p, among many of my suggestions, should take a backseat. `/nod`.
> It would be awesome to have a p2p github alternative that could ship builds over BitTorrent or IPFS
I understand you've been around the block, so please ignore my gibberish. To my eyes, you're also building a serious competitor to [[P2P Matrix|2021.11.17 - HN Log: Arathorn & P2P Matrix]], which is still under development. I'll mention that Radicle may be worth your consideration. The functionality of [[DarkMX]] is worth your time as well (it is closed source, but its creator is one of the few with the right to ask us to trust him [even if I stand in disagreement]). I know some, like [[Ian Clarke]], will claim it is better not to "focus too much on competitors, better to focus on the needs of users," however, given the specialization of this tool, I think there's something to be said for having tasted the alternatives. [[Aether]] may be a direction you could eventually go, but I understand that isn't your immediate target; I don't know if protocol design decisions now might affect the future of such a thing.
> Beyond participating in Tor, Quiet does not allow you to help any network you're not a member of
`/nod`. I wonder if it is feasible for Quiet to have something like the Encrypted Key option in Resilio Sync, enabling one to seed in swarms without knowing the contents. There are some cases where it's useful, though there are tradeoffs.
There may come a time where you've decentralized to the point where it hurts too much, and you'll have to centralize until it works. I wonder if there are cases in which randos must hold at least metadata for others or act as proxies (I understand that Tor solves many problems for you). [[Tox|Toxcore]] is an exemplary tool in part of this space you're solving, and even they haven't solved some of these problems. Seamless multidevice and offline messaging are dealbreakers for many people.
> If you have access to a large number of machines, rigorous performance testing with hundreds of nodes could possibly be helpful?... manual and automated testing on many diverse Android devices; that would be helpful!
I'll be thinking about that then. NixOS with solid orchestration on a beefy machine (I'm not sure if zswap would assist here, but memory seems a serious bottleneck) may be able to push hard. There's probably no way around using a lot of machines unless a much thinner client were feasible. The nice part is that donations might allow you to control the machines remotely while dogfooding.