But also, Dual EC was suspected of being backdoored from day one, was slower than existing CSPRNGs, and was therefore avoided like the plague. Whereas the premise is that if you put all the world's secrets behind one set of keys, there doesn't exist a level of defense that can withstand the level of attacks that will attract. Which doesn't apply when it isn't widely used.
On top of that, the attackers would be the likes of foreign intelligence agencies, and then them not getting it and the public not hearing about them getting it are two different things.
Exactly. They built a backdoor that "only they" could get into and then somebody else slipped into it anyway.
The backdoor is a vulnerability even if you don't have the keys because it requires the trappings of third party access. If you try to get something in the shape of a backdoor through code review, you should get knocked back. But if something in the shape of a backdoor is required then a change in who has the keys to the lock is much smaller, more subtle and easier to sneak in.
No, that's exactly what didn't happen here. The attackers in this case got and maintained for years the ability to slip code into Juniper/Netscreen releases. That the backdoor they chose happened to replace NSA's NOBUS backdoor is just a funny detail.
I don't think it's actually irrelevant; there's a reason they did it that way. Getting commit access and being the only one who can even read the code are two very different things. Even if you can modify the code, the less obvious it is that the change is adding a backdoor the less likely someone else is to catch you.