Grats @dankrad @karalabe.eth you got me to interrupt my work on EIPs and make a twitter response that turned into an unscheduled poast 😀

Grats @dankrad @karalabe.eth you got me to interrupt my work on EIPs and make a twitter response that turned into an unscheduled poast 😀

https://vitalik.eth.limo/general/2024/05/17/decentralization.html

Thanks. In exchange let meg interrupt my shit-posting to write an EIP (brain-dump) :)

Thanks. In exchange let meg interrupt my shit-posting to write an EIP (brain-dump) :)

https://x.com/peter_szilagyi/status/1791589527452336341

It's an interesting proposal! I think the biggest challenge with this kind of approach is intentional "edge attacks" to try to split the network. So, suppose you have X+1 proposer slots in a row. You send a tx right on a slot boundary, and don't include it X+1 times. Half the network thinks your last block is censoring

BTW I do think that "ILs are not AA-friendly" is a misconception. They're actually not that hard to make compatible, and if some account's validation conditions are too crazy, ILs would fail in the direction of not giving them censorship resistance, and not in the direction of splitting the network.

If we want to do "anti-censorship fork choice", then there are ways to mitigate edge attacks; they end up looking either like this https://vitalik.eth.limo/general/2018/08/07/99_fault_tolerant.html or like random timers (eg. VDF-based schemes)

And tries to fork away from it, the other half thinks you haven't crossed the line, and keeps building on it. And so you get two halves of the network (even if temporarily) going in different directions.

OK I thought about it more, and I think you're right; if it's a fork choice penalty, and not a validity rule, and there's an eventual convergence guarantee, then it seems like it could work. The main risk I see is accidental mempool splitting; the mempool contains lots of objects and if you disagree even on one...

I'm unsure this is an attack though. If I am a malicious validator that try to mess with my own block, I could try and time the propagation to split the network, but the effect is that my block might get reorged out f 50%+ don't attest. If 50%+ does attest, the block remains. But I don't see a split here. 1/2