So I just got a response back on this from someone who doesn't want to be identified that changed my thinking.
"So you are going to give me the ability to assign up to 2^66 identifiers locally that are guaranteed[*] to be globally unique in a 128 bit space. I can use them in my QR codes"
Which makes perfect sense. If I was going to stick a QR code on every bag of stuff I put in the freezer, I would want those to be globally unique. Then I can avoid thinking about the cases where this occurs by accident.
And that got me thinking that the mapping should go both ways. EPC codes are 96 bit codes used in RFID tags. We should reserve a chunk of IPv6 address space for EPC codes. We have a whole /8 for registered ULAs, this would burn one out of 16 million code points.
What is a non-routable IP address? It is a name. It is a signifier with the characteristic thirdness. The signifier bears no intrinsic relationship to the signified. It is a purely conventional relationship.
So lets see how this might be applied in a real life situation:
A can of beans is produced in a factory, it has a unique EPC code X printed on the side of the tin.
Alice places an online shopping order for some beans, the can with unique code X is delivered.
It is discovered there was a defective batch of cans and the beans have botulism. There is a recall. The online shopping company comes round with the replacement and collects the precise can of beans that was defective.
That might not look like routing, but it is a form of routing. Its just not a form of IP routing. There are many variants: tracking stolen goods, counterfeit drugs, etc. etc.
Why would Alice need to mint her own codes? Well she probably has a digital camera or six and they produce digital assets by the thousand.
So I am going to go ahead and write up a draft describing how I intend to allocate /65 chunks within a 128 bit address space which all begin with a /32 prefix in FC00::/8 . If IETF wants to avoid creating unnecessary ambiguity, they will accept my proposal to create an IANA registry and allocate me a number. Otherwise, I will create my own prefix[*].
If someone else comes along asking for an assignment, you just give them the next prefix. If my registry fails, there is no real consequence. If it succeeds its because the IETF mission has advanced a little bit.
[*] People who propose schemes for ubiquitous use of end to end encryption after 30 years or periodic harassment by various intelligence agencies concerned at the loss of their SIGINT capabilities are not likely to be the sort of people who take notice of officials saying 'no'. People who have spent two years developing such a scheme at their own personal expense are even less likely to.
