Please note that this is an automatically generated transcript and will contain errors.

---

Anyway, Quintus asked me to give you guys an overview of the energy supply chain, particularly its geography. And basically just talk about how things look today. So if you want to give some really quick context, PBS or proposer builder separate, is an auction mechanism that connects proposers who own the rights to sequence a lot with lots sophisticated entities that specialize in ordering transactions, and then to sort of protect both parties.

In this arrangement, we put a relay in the middle with this relay auctioneer. And then the other term you like your lot stay is energy Boost, which is essentially the software that the proposer runs that interfaces with many relays. Validators are super proposers. A proposer is basically the validator selected for an individual block. And these validators are the backbone of the network.

So this is a map with data from a chain down the theorem foundation. It basically shows that validators are quite geographically distributed. The density in Africa in South America was somewhat low, but otherwise they have good coverage. What about the relays? At first glance it seems pretty decent for their number lots relays in Europe and on the east coast of the United States, and then some instances in Asia.

But in reality, there's sort of two classes of relays ultrasound and block group combined for about 85% market share, meaning that all the smaller businesses really don't have that much contact. For example, the relays in Asia likely account for only like 1% of all blocks. So back to the full map. If we now only look at your one relays, you'll see that they're only located in two locations.

Those happened to be the economic centers of charity and happen to be connected by subsea cables. But there have been some studies to address this. One of those is relay proxies. So this is the diagram from Blockstream. And basically the idea here is that they create mirrors of their information screens and push those to different nodes around the world.

These are just errors. They're not full instances. They just broadcast bits rather than accepting them. But the idea is that if you're a proposer in Switzerland, instead of connecting to the auction in North Virginia, instead of talk to the server in Frankfurt and only have five seconds of latency instead of 50 milliseconds. Now, of course, this doesn't actually reduce the total latency, but because the mirror gets a constant stream of blocks versus a single response, it means that the variance is less important and also it's much faster in the second stage because they cannot reveal scheme has multiple stages of message passing.

So this would be like signing the bids actually payload. Those are much faster because the instance is closer to people backstage. You want to relay that and then we add proxies. We see that once again there's better global coverage. And in theory, if you're elevator it throughout most of the world, you should have a proxy nearby. And it's cool.

So we can actually look at some latency statistics to see the impact. So here's some data from wider proxy who's basically just like spins out bunch of servers around the world and then pays them. If you look at the latency to the datacenter in North Virginia, we see that basically everyone outside of the United States has latency over 50 milliseconds.

Or if we look at the French relay, we see that it's basically just European validators that low latency connections to it. We can do a superposition and it looks a bit better, but we see that validators in Asia still suffer. But when we add in the feeds from all the proxies, all of a sudden we get good coverage basically everywhere, with the exceptions being Africa and South America and China.

But I would assume that that's probably their national firewall. So I think, you know, at that.
Okay. So proxies give proposers a low latency connection to big streams. But what does that mean for information flows? One thing that we can do is construct a simple model of the most latency sensitive water flow. That would be price signals from centralized exchanges, particularly from Binance in Tokyo. And what we see is that the total on the wire Regency is comparable.

When the signal goes through US relays or French relays should be about 150 milliseconds plus infrastructure through France versus 130 most sites for the US. But again, the US relay has that big 50 millisecond call between the relay and the bottom. So in this case, we had a proxy in France. And for a normal lengthy trade off, we get more reliable.

In the call between the proxy, the proposal and then everything on. But between the proxies, it's sort of like always a metric bullet. And we can sort of look at special cases for this. So one of those might be a proposal showing up in Japan. So in theory the price signal should only be 30 milliseconds wait for a proposal.

But because the auction stop and say Virginia, the signal has to go all the way there. And so so that's 150 milliseconds of latency. And it's still better than like not having a proxy. But if they're sort of competing with the relay in Tokyo, it just it can't it really can't compete. Or we can look at sort of a closer in Phoenix.

Here they have lower latency to the proxy on the West coast, but the proxy is actually further from the relay than the access. So even though I was there, can get a more reliable signal from the proxy. In theory, the proposer could actually have more up to date information on the proxy itself. So with this framing, proxies are kind of a valuable tool for proposers, but actually don't provide that much value to the searchers because they just distort the information.

Okay. The next thing we should do is look at the slide timeline that we have today. And sort of the key takeaway here is that the last eight seconds of the slide is really inefficient. It's basically just small message passing and then just running a four choice algorithm. In contrast there's a ton that happens in the first four seconds.

There's the entire commit reveal scheme for PBS. And then there's block propagation and execution. And because the only deadline here is at four seconds, we see different proposers and relays use the time differently. This leads to what we call timing the ends and sort of a blurry crossover between the two phases. And we actually have this data from end ups where we can see a very similar, so if we separation based the most intense timing games and from all the honest proposers, the profiles are kind of just night and day.

So it's just really pushing events just as far as they come. And even inside these groups there's a lot more nuance. So I've been going to other stuff further. And so for, you know, these other geography, we see that nodes in the United States, in Europe can handle films, Titan games, but the validators outside these regions struggle to adjust on tidal not likely to have some meaningful impact on the rewards.

But broadly, for honest proposers, there's enough room in the slot timeline that every out of every elevator otherwise sort of earns the same rewards. Ratifications. And then sort of the last thing that we should look at for geographic decentralization here is sort of what's happening inside the auctions themselves and how that changes the co-location incentives for block. So what we see is that over the last two years, auctions have gotten much faster.

Sort of back in 2023, the median ultrasound relay during the heart of the auction life for 84 milliseconds by the end of last year is down to just $20. Companies. And then in tandem, we can sort of look at changes in bidding patterns. And what we see is that the increments continue to shrink. And sort of when you put these two metrics together, these options start to just look a lot like a PTA or sort of like a shared private value auction for just trying to respond as fast as possible to still capture some of it and sort of waiting on these incentives to read it and respond as fast as possible together.

It shouldn't surprise anyone that it encourages co-location. So one cool thing we can do here is we can take dates from both the ultrasound and block screen and sort of compare the time stamps. And what we see is these two distinct groups of bidders. So we have a set on the right for bids from certain hotkeys to ultrasound much faster.

We have a group on the left where bids get to block three much faster. So this is sort of like introduction evidence of co-location or one by how much of an advantage do they get? Well, that's really depends on the network latency in the system. And sort of over the past two years, what we've seen is that this is declined a lot.

So everything from sort of like optimistic relaying, faster block building algorithms sort of amplifies the returns from co-location, basically like as you can cut down the infrastructure latency, sort of all that's left is the network latency. So it becomes the dominant factor. And that's sort of the dynamics that we find ourselves in today, where co-location matters a lot.
And you sort of have to do as a blog if you want to be top tier. I don't think I'll read these questions out loud. But these are sort of the topics that are on my mind. So if anyone wants to jam, please do find me. So a such clustering and I hope you guys have a great SPC.
So it's going to be ten minutes talk and then five minutes. Okay? Yes, I start. Yeah. Let's solve the problem. Let's see.