> For the most simple use case of an client auth state; you want to be able to revoke auth straight away if an account is compromised. This means you have to check the auth database for every request anyway, and you probably could have got whatever else was in the claim there quickly.
I fail to see the relevance of your scenarios regarding JWTs. I mean, I get your frustration. However, none of it is related t JWTs. Take a moment to read what you wrote: if your account is compromised, the attacker started abusing credentials the moment he got them. The moment the attacker got a hold of valid credentials is not the moment you discovered the attack, let alone the moment you forced the compromised account to go through a global sign-off. This means that your scenario does not prevent abuse. You are revoking a token when it was already being abused.
Also, as someone who implemented JWT-based access controls in resource servers, checking revocation lists is a basic scenario. It's very often implemented as a very basic and very fast endpoint that provides a list of JWT IDs. The resource server polls this endpoint to check for changes, and checks the list on every call as part of the JWT check. The time window between revoking a token and rejecting said token in a request is dictated by how frequent you poll the endpoint. Do you think, say, 1 second is too long?
> Same with roles; if you downgrade an admin user to a lower 'class' of user then you don't want it to take minutes to take effect.
It's the exact same scenario: you force a client to refresh it's access tokens, and you revoke which tokens were issued. Again, is 1 second too long?
Also, nothing forces you to include roles in a JWT. OAuth2 doesn't. Nothing prevents your resource server from just using the jti to fetch roles from another service. Nevertheless, are you sure that service would be updated as fast or faster than a token revocation?
> So then all you are left with is a unified client id format, which is somewhat useful, but not really the 'promise' of JWTs (I feel?).
OAuth2 is just that. What's wrong with OAuth?
Also, it seems you are completely missing the point of JWTs. Their whole shtick is that they allow resource servers do verify access tokens locally without being forced to consume external services. Token revocation and global sign-offs are often reported as gotchas, but given how infrequent these scenarios take place and how trivial they are to implement (periodically polling an endpoint hardly changes that.