When a network is only partially synchronous the biggest pain is not theory it is timing… messages arrive late some arrive twice some do not arrive at all and suddenly agreement feels fragile.
And this is where most PoS designs get exposed because they assume the network will behave nicely most of the time… but in real life latency spikes happen congestion happens nodes go offline and the chain starts producing weird outcomes like delays and competing blocks.
So the problem I watch for is simple… how do you keep finality fast and predictable when the network is not perfectly reliable and not perfectly synchronous
@Dusk answers this by building its consensus as committee based PoS from day one … the succinct attestation protocol runs with stakers called provisioners and instead of everyone voting forever it selects a single block generator and separate voting committees each round using deterministic sortition.
The flow is built to survive bad network timing… proposal happens first then validation then ratification and each step has timeouts so the protocol can move forward even when a chosen generator is missing or votes are delayed… and if a step cannot reach quorum it does not pretend it did it simply continues into another iteration to try again.
Committee voting is also shaped around partial synchrony because Dusk makes votes compact and verifiable… votes are signed with BLS and aggregated so the network can pass around a single proof of quorum instead of dragging a heavy list of signatures through a congested network.
But the real honesty is that Dusk expects forks when things go asynchronous … it literally says forks can be produced because messages can be delayed or lost and then it uses a fallback rule that prefers the lowest iteration block and reverts higher iteration ones if a better candidate arrives later.
To make that livable it adds rolling finality… blocks move through states like accepted and attested and confirmed and final so the chain can express certainty gradually instead of lying about instant finality… and the rules depend on attestations and successors so you get a structured way to measure stability when the network is jittery.
And when partial synchrony turns into a real outage mood where many provisioners are offline… Dusk has emergency mode after repeated failed iterations… timeouts get disabled and open iterations can run to increase the chance of reaching quorum and if needed an emergency block can be produced on explicit request backed by majority stake so the chain does not freeze forever.
All of this is powered by the token logic in a direct way… $DUSK is what you lock to become a provisioner there is a minimum stake and stake weight influences selection into committees through the sortition process… then rewards and penalties are applied based on attestations so the protocol pushes people to stay online and vote when selected.
So for me the review is pretty clear… when the network is only partially synchronous Dusk does not fight reality with wishful thinking… it uses committee based PoS with iterative steps fallback and rolling finality and emergency mode… then it uses $DUSK staking rewards and slashing to keep the human side of the system honest under stress.
