- Am I right in saying that the question that you are ultimately answering with the FBA approach is how we onboard new verifiers?
Yeah, FBA makes onboarding new participants (buyers and proposers not just verifiers) pretty trivial while keeping social consensus load low (Much lower than a POA system since it’s a lot more decentralized). It also provides pretty clean tools to effectively respond to poor performance rather than only having the binary tools that address really only malicious behavior since nodes can downgrade trust levels etc. Then it’s also really lightweight and cheap from a cost of capital perspective. @tdep may have more to add here
I’m not very familiar with FBA, but the idea that a malicious verifier or proposer can influence the network topology and consequently which chips are tested by which verifiers is a little scary to me
Theoretically FBA handles this fairly well since a buyer/proposer could “weight” their trust threshold to accept a verifiers commitment. So if say node A onboards 20 malicious nodes that don’t have additional trust intersections beyond the one with node A - node B (who presumably trusts node A) could still avoid interacting with a malicious node as long as they require more than 1 trust intersection with a “peer”. This makes it quite difficult to sybil the system. Additionally - such behavior by node A probably leads to them being downgraded by the peers currently trusting them. The SCP whitepaper explains this much better and is worth a read, it’s a very clean mechanism https://johnpconley.com/wp-content/uploads/2021/01/stellar-consensus-protocol.pdf
I would want to speak to people at the labs who we expect to be doing this kind of work. Perhaps this is not a natural additional task for them and relying on economic or (other reputational) tests make more sense.
Yeah would be great to talk to these people about what their requirements are - I’d imagine that trust relationships are much more natural for a university than having them put up a stake (i’d figure this is a nonstarter) - but we should actually check that