Or, I guess, "someone" could apply for a custom gTLD and link it. Of course, that "someone" would need the $200K needed to review the application and all that stuff :P

Eh, you realize that this very work on the GNU Name System prompted IETF to create the ".alt" zone for this purpose already, minus the $200k fee? Registration is open at https://gana.gnunet.org/dot-alt/dot_alt.html

I'd settle for global and not human readable, if it means I get a domain I can use as a CNAME on a nicer domain.

Dynamic dns services are nice and all, but needing to pre-register gets kind of annoying.

It can't possibly work, if you want your nicer domain to be globally unique.

If you don't, it depends on how local your domain needs to be; maybe all you need is a record in /etc/hosts on your home router.

If I understand they want a (globally unique, secure) GNS name and a (globally unique, human friendly) traditional DNS name which acts as an alias for the GNS name via CNAME.

This can work, and sounds like a good compromise in that it lets machines and people who care deeply about security use your secure name (which is more portable than an IP address), while providing a human friendly name for people who don't care and just want things to work.

But what's the point? Federation, I suppose?

Because if globally unique, human-readable DNS still works, I see no point in migrating off it. If the point is smoother migration, then we should start forgetting about human-readability, because it's going to disappear anyway.

TLS certs without having to buy a domain? Create a GNS domain, set a LEHO record with the necessary Host name, and make your cert based on that? Obviously you'll need a CA that's willing to issue certs for GNS LEHO names, but that way you can use the current TLS CA system to a domain without having to actually spend money on one. Alternatively, have the CA issue wildcard certs for zTLDs and then you can manage your own zTLD without issue.

Letsencrypt allows to produce TLS certificates for free, under a reasonable, globally trusted CA.

If we wanted to go away from centralization here, that would require a serious breakthrough, the magnitude of Bitcoin.

Sure but that's contingent on owning or being able to host something under a domain.

These are all valid deployment questions, which we tried to address in Appendix A.

In a nutshell, we expect that resolvers would ship with a (large) set of default "suffix-to-zone" mappings, that can be overridden by the user to provide a usable and convenient out-of-the box experience. Not that "we expect" means that this would be the ideal scenario, not something to expect when installing our reference implementation right now.

Would <UUID>.example.com not work? With first to register getting priority. Or <pubkey>.example.com with the corresponding private key needed to do updates.

The "nicer" domain I am referring to would be a normal domain from a registrar.

Indeed that would work. In theory. Especially since we thought of that use case (delegation into DNS) with the GNS2DNS record type.

There is a BUT: You need an initial label for ICANN zone to resolve the names. Unless you have a resolver implementation that "hides" the zkey of ICANN in the UI. But technically, under the hood, a name for this ICANN zone would look like:

www.example.com.THEICANNZKEY...

ICANN could also publish the TLDs individually as zones, however, and you could have an "ICANN Start Zone" (see Start Zone in the RFC) consisting of the TLD/zone key mappings.

Since the TLDs are multiplying with no end in sight, using a zone key seems smart.

Zooko's triangle https://en.wikipedia.org/wiki/Zooko%27s_triangle You can't have it all.