is incomplete: a distributed consensus system (which is, at heart, a distributed database) can not have all three features... but there is no guarantee that a distributed database has any of those features.
As with everything else, execution matters.
If you're going to be recording history for financial transactions, you need to put immutability as your first goal. This is not compatible with unlimited space-time separation of trusted inputs, so the second thing you need is to decide how you're going to resolve inconsistent histories. Doing so always involves a centralized trusted system, even if it is fed from a distributed system: someone needs to decide what transactions really happened. You can claim that you have a distributed algorithm to do so (consensus) but that itself will always fall into the same distribution problem.
And that's what seems to have happened here.