Ok, we'll revisit this as soon as I have this part in the draft or when it's finished. In the mean time...Iljitsch, the big hole in your -00 draft is that is doesn't at all discuss the magic needed to distribute mapping state and keep it fresh.
And as Jim said, there are stateless solutions, but that means adding bits to the packet.And as I replied to Jim, I don't feel what he mentioned is truly stateless. Sure, it avoids state in the core of the network and at the receiver. Nobody was suggesting we should have state in the core, so the problem must be with mapping back at the receiving end. I don't see how this can be considered problematic as doing this for the destination address is entirely trivial (just replace PA prefix A or B with PI prefix X) and for the source address it may either be unnecessary if we transmit the source address unmodified or the state necessary for this must be present in order to be able to send packets back anyway. Similar state must be present in order to add the routing header that Jim mentioned.