> self-replicating
More like something similar to how Freenet does it, you install a client and files propagate to different nodes across the network as they're inserted/requested. The biggest problem with Freenet has always that the algorithm it uses tends to favor keeping small but often-visited things like someone's "lel I'm using Freenet" test page while dropping less visited but potentially more useful files (for a few reasons, partly the popularity algorithm itself which might or might not be improvable, partly because access tends to be from an index page and anything linked to on said index page gets more hits than anything else. This system also has the issue that you have to trust the index maintainers in what they may decide to show/bury; Freenet's tendency to prioritize trust systems over building something that works from a "no one is trustable" stance is the other big issue I have with it, but ultimately they may be on to something with that one, any sort of connection to another computer requires some degree of cooperation that could be maliciously manipulated by a bad actor). Ideally, of course, everything posted would be retained forever, but limits of available disk space make some sort of prioritizing a practical necessity. If that's too virus-like for your taste, enjoy your server takedowns I guess.
> intended to be impossible to stop
That should be a no-brainer as necessary to the definition of censorship-resistant. If it would be easy to stop, it would be easy to censor, now wouldn't it?
https://github.com/freenet/wiki/wiki/FAQ gives a pretty good overview of why they made things the way they did. Most of the ideas are good (and generally better than other systems I've seen), I just think there are a few flaws that keep it from being a really viable for general use.
I'll also add the disclaimer that it's been a few years since I've used this thing and it may well be that stuff's been improved in the interim.