#dusk $DUSK Dusk is often summarized by a single phrase, “deterministic finality.” When I first understood it, I thought: once a transaction enters a block, it will never change again. Only after reading the definitions of consensus and the node-side state did I realize that finality here is a progression process—there is still a window in the middle that can be replaced. #dusk$SNDKB
The way Succinct Attestation works is: in each round, a committee is drawn from the stakers, and they vote on candidate blocks. Once the vote threshold is met, the block is confirmed. But on the node, it’s not simply a binary choice between “valid/invalid.” There are three different treatments: a block that was just accepted, a block that has collected enough signatures, and a block that has been pushed by subsequent blocks to the point of no return. The previous section is what’s often called rolling finality—here, a fork can still replace the record you just saw.$SPCXB
The meaning of this window varies greatly for different roles.
For ordinary transfers, differences of a few dozen seconds can be ignored; the payer may have to wait a bit longer at most. But for custodians and exchanges, it’s completely different: if the accounting logic only checks whether “the transaction has been written on-chain,” then you’ve effectively taken the rollback risk onto your own books. This is also why integration guidelines emphasize waiting for finalization, rather than just giving a fixed number of confirmations—confirmation counts are an empirical value, while finalization is about state.
Institutions performing asset settlement need to go one level deeper. Securities settlement is about being irrevocable: once a clearing instruction triggers the subsequent actions before finalization, if the chain rolls back once, the off-chain side has to handle an entire set of reversal procedures. What these systems truly need is not “fast confirmations,” but “when it can be determined that there’s no need to look back again.”
So when I think about DUSK’s finality, I break it into two numbers to record: how long it takes from broadcasting to inclusion in a block, and how long it takes from inclusion in a block to being irreversible. A block-production latency on the order of 10 seconds only answers the first question—the second question determines whether institutions will actually connect the settlement workflow. Deterministic finality is a real advantage, but it’s the endpoint along a line, not the moment the transaction first gets on-chain.@Dusk