> Also, I don't even use my ISPs name servers, I use Cloudflare or Google, so I don't think it's that unless the ISP is somehow munging the packet in transit, which I suppose is possible.

That's exactly the problem. You send out two v4 DNS UDP packets one after another (one for A, another for AAAA), both go via your ISPs CGNAT, the CGNAT gets confused, one of the packets gets dropped. I've seen this exact behavior when talking to 8.8.8.8 on Orange in Poland (and they do DS-Lite). It didn't occur with the ISP's DNS, because a) they were also on v6 b) they weren't getting CGNATed.

> But my point is, I shouldn't have to be a network engineer to make v6 work. I should be able to turn on my computer and just have it work.

By disabling IPv6 you're letting shit ISPs get away with this. Your ability to debug this and to figure out it's the ISP's issue should be used to voice your concerns, and not just let this slide.

You should know from your own background that AT&T doesn't quite have a history of excellence in the Internet realm.

Oh yes of course what I meant was they are a huge ISP and have lots of customers, and if it doesn't "just work" for them, then what chance does anyone have of v6 taking off?

Counterpoint: if they fix their shit, the chance of global v6 adoption increases dramatically.

I've never had any issues with IPv6 on Comcast, and I've been using it almost since day one. AT&T is not a company you should ever hold up as some kind of good example of network engineering.

I was never able to get ipv6 to work when I was on AT&T, couldn't even get addresses assigned. When I got on Cox at my new house it worked out of the box. So some ISPs get it right.