Cameron,
Jari states IPv6-only works well for mobile. I believe one of his docs (Google preso) states it approaches 100% functionality.
Yes, though for the record, draft-arkko-ipv6-only-experience does say that under certain assumptions (the ability to pick the right user devices) you reach 100% functionality.
I do believe that IPv6-only (with nat64) is an excellent approach for mobile networks, and I hope we will soon see this in widespread commercial usage. Perhaps in your network :-)
That being said, I do not want to give the impression that its a piece of cake either. The assumption above is a tough requirement, and for sure there'll be plenty of other work ahead as well. I do stand by my recommendation that *at this time* even for mobile operators, dual stack should be the default strategy even if some networks are going for IPv6-only. I am aware of the additional complexities on the dual stack side as well. In any case, I believe that a few years down the road the recommendation will be different, i.e., enough software and user equipment have matured so that IPv6-only can be considered as the default strategy. Of course, some networks have to be at the forefront. They will be the leaders, but will also get the initial pains from hitting various issues first.
Anyway, what are we really discussing here about? The document *does* recommend all four different models. I think we do have deployment evidence that speaks in favor of dual stack being the most problem free (at the moment), so this is why it is discussed first and described as the default strategy. I would be happy to use more accurate or different wording, but I don't think that you are arguing either that some of the models should go away. Would you like us to add something more specific about types of networks (e.g., mobile) where translation-based model is recommended?
Jari