Anti-Correlation Penalties

Great. Thanks a lot to everyone who showed up here. Um to quickly introduce myself, my name is Tony. I want to I'm doing research at the Firm Foundation, and today I want to talk about uh topic I perceive as very, very important. Um namely decentralization.

And to be more specific, I want to talk about the topic of anti-correlation penalties. To get into the topic, let me quickly describe the problem that we have today. On this chart, you can see the missed attestation rate um of different entities going from Kiln, Eth 2 DeFi, Lido, the very professional entities on the on the left-hand side, to solo stakers, Kraken, Rocket Pool um on the right-hand side. To be fair, Kraken is an outlier here, so I think when I did this chart, they just had some struggles, but in the end, the picture is quite clear. Solo stakers are struggling with um what the network demands from them.

And what this chart actually shows is the missed attestation rate compared to the expected missed attestation rate based on the market share. And if your bar is above zero, this means you miss more attestations than expected. And this is of course problematic because solo stakers um we might all agree are the most important entities in the Ethereum ecosystem. They are super important for various different properties, and we should make sure that solo stakers remain competitive compared to all the large pools. One thing that is also very evil and dangerous are economies of scale.

So, what's what big pools can do, they can just put many validators, for example, on one node, and this allows them to lower their costs. Solo stakers cannot do that. So, if you are a solo staker, you might just have one validator, um therefore, you have one node, you use one internet connection, and you have to manage it yourself. On the other hand, the pools, they can use economies of scale to make their operations more and more efficient, right? They can use one cloud provider, use one internet connection, and so on.

And in the end, what we might um go towards to is a very centralized validator set. And this is something that is very bad for Ethereum, and we should definitely try everything to avoid that. The cool thing is, so now getting into some positive vibes, the cool thing on the right-hand side is you cannot really leverage economies of scale without also creating correlations. For example, if you put fork 4,000, 5,000 keys on one machine, there is a correlation how those validators behave in the end. If your node goes offline, then all of your 5,000 validators will be offline.

If your cloud provider has a problem, then all of your keys might be affected. And the same applies to all the other things that affect economies of scale. So, what we see on the right-hand side, I would say it's a um correlations and economies of scale should be ver- um should have a very linear relation. And now, how can we actually leverage this knowledge, leverage the knowledge that economies of scale might scale also the correlations um to the good for Ethereum. And this brings me to an EIP that is still very young.

It's EIP um 7716, uh called anti-correlation attestation penalty. And as you can see here, um it's very young. So, the EIP is only a few months old. Um the original idea is from Vitalik, and Dapplion and me hacked on the implementation at the Eth Berlin hackathon. And now we are getting closer to final um not really finalizing but getting closer to finding ways how we can best introduce anti-correlation at the station penalties.

The high-level idea is the following. If many validators miss their attestations in the same slot, we just scale the penalties up and vice versa. If they if only a few validators miss an attestation, we can lower the penalties. So basically, if you're running all of your keys on one machine, the machine goes offline, then you might expect higher penalties and a solo staker that goes offline should receive lower penalties. And this is the goal of this EIP.

This is just an example how it would look like. You would have many validators on one machine, the machine goes offline, the correlation would be high and so would the penalty be. On the other hand side, you're a solo staker, you have one key on one machine, the machine goes offline and the uh it should then affect your penalties by just lowering them. Finally, let me quickly present some initial results. So this is from empirical data.

It's approximately um um 6 months of data and what we can see is that the comparison between the pink bar and the gray bar and you can already see that only the small-size clusters and the unidentified cluster um which both include many solo stakers, they would profit from the whole thing. So we can see solo stakers would have been penalized almost half what they were actually penalized. Whereas the big stakers, like we saw earlier, Lido, Kiln, Eth FYI and so on, they would receive higher penalties. And in the end, what we want to do is just create a leveling playing field for all these entities and allow solo stakers to finally also econo- economically compete with the big guys. That's all I have for this lightning talk.

Thanks a lot. If you have any questions, if you want to engage with the EIP, we are still in draft stage. So, happy for any input on the EIP itself. And yeah, feel free to reach out if you have any questions. All right.

Very unfair to be a solo staker, right? So. Uh could a withdrawal address be considered a correlation? So, like if a staking large staking pool used the same with contract address for a withdrawal address, could that be considered a correlation? Um that's an interesting question.

So, of course, we always have to check what can the protocol actually see, right? The protocol can see correlations because of missed attestations. And also, as you said, the protocol might also be able to see withdrawal addresses. I guess this is very easy to circumvent, right? We will basically create an incentive for the big guys to use multiple withdrawal addresses.

So, this is very easy to circumvent, whereas um correlations because that arise from running everything from a very centralized setup cannot really be circumvented. Uh Tony, thank you so much. Um can I kind of reduce my correlation by artificially kind of taking individual of my 10,000 validators offline selectively to make a pattern that is not correlated? Yeah, you can. And that's actually one of the goals of the EIP.

So, for example, if you're a big staking pool and you have thousands of keys running on a few machines, what you can do is basically split those keys up and put them on many different machines. The cool thing is you lose economies of scale. So, it costs money. You have to maybe manage more machines. You have to diversify your setup.

and in the end, you basically do exactly what we want you to do. You diversify your whole setup, and you um increase your fault tolerance. Okay. Yeah, I mean, if I'm a if I'm a non-staker, but I want to put ETH into a staking pool, you're saying that I will be slashed disproportionate to my ETH. Is that So, you're not in that you're not penalizing you're penalizing people below solo stakers, right?

People that don't that can't run the node. And I that's a that seems like a very large weakness here of uh possibly this proposal. Yeah, I see your point. Um I would say it's the following. So, why would someone stake with a pool?

Because it's because they offer a high enough um a high enough profits for you, right? As soon as we force big pools, as as soon as we lower the profits for big pools, either by increasing the penalties or by forcing them to diversify the setup, their um rates that they can offer, they will also be reduced, right? So, I would I would argue that this would create an incentive for someone who thinks about maybe staking with a pool to yeah, consider it again, and maybe don't stake with a pool just because the incentives are not high enough. Right. If they can't stake, then for example, what you could do is look into uh one of the entities like Rocket Pool that offers you to also stake with less than one ETH while still having a very anti-correlated setup.

Okay, guys, time's up. If you want to continue the the chat, you can you can do it afterwards. Thank you very much, Tony. Thank you.