I tried to answer the big questions you had clearly and concisely. I hope this helps!
but in the registry of DID methods in development Atala Prism or IOHK is nowhere to be found
Indeed, the Atala PRISM DID method is not published. Public registration is only required when Mainnet switches on - which is coming soon. I can assure you that DIDs and VCs are being issued on TestNet by Atala PRISM Pioneers and multiple companies.
What is the benefit? Really?
Privacy, security, data portability, convenience, time and cost saving, and agency are just a few.
What measures are there in place that I don’t just register three DIDs and issue myself a “verifiable” credential that I am Jeff Bezos and hold a PhD in quantum mechanics?
This question is challenging to answer in a forum setting. In essence, what you are saying is possible. However, as a verifier, I would only accept trusted or verifiable issuers of VCs. I have an article coming soon about trust on the Atala PRISM blog.
The DIDs are on the chain! It has to be true!
Interestingly enough, Atala PRISM DIDs don’t have to be on-chain. Only the DIDs that issue credentials have to be on-chain.
How decentralised and interoperable is it?
The only centralized portions that I can spot are in governance frameworks and the trust registries they create. Again, this touches on the trust issue that is quite complex to explain in a forum setting.
Interoperable? Having a universal resolver (being able to resolve other DID methods) is a high priority.
If that really wants to live up to the idea of “decentralisation”, it has to be interoperable!
The Atala PRISM team agrees.
What happens if the blockchain I foolishly chose to put my DID on doesn’t survive the next crypto crash?
What happens if the blockchain my government, school, or employer foolishly chose to put their DID on seizes operation?
This problem is resolvable by creating a new DID on another platform and performing a key rotation from the defunct DID method to the new one, which also covers issuers. However, with full transparency, the issued credentials would still be verifiable in a limited capacity, depending on the assurance required. There may be a solution to this, but I’ll admit, I’m not technical enough to explain it thoroughly.