Why You Should Be Optimistic About MEV - Shea Ketsdever

[Music] Welcome back everybody. It was truly an interesting fireside chat. Well, luckily we had a break in mind that fills in this buffer really well. So, we're going to just jump right into our next talk and then instead of having a break, we're going to finish the rest of the the afternoon sessions. So, without further ado, let's welcome our next speaker.

And for this one, I'd like to bring on Shay from Flashbots to tell you why you should be optimistic about MEV. So, please give her a big round of applause. Um, hi everybody. Um, let's see. I'm Shay from Flashbots and today I want to talk about why I am optimistic about the future of MEV and I think that you should be too.

Um, if you've been building in Ethereum lately, you have probably heard a lot in the discourse about things like scaling, interop, making the L1 more competitive. But I think what you might not know is that all of these things uh kind of depend on an actor that many people don't spend a lot of time thinking about. Uh, which is what we call a block builder. And that is really what this talk is going to be about. I want to tell you what block building is, how it's evolved on Ethereum over time, some recent breakthroughs to upgrade upgrade this part of the stack, and a bunch of exciting implications that I think this has for Ethereum.

So what is block building? Uh we will get to the cool stuff in a bit. Uh but the very simple answer is that each block in a blockchain contains an ordered list of transactions and block building is how we decide what goes next in the next list. So, what gets included, what gets excluded, and kind of what order things go in. And the default Ethereum client has a sort of built-in way to do this.

It just orders all the transactions in the public meool by price. And this works pretty well for simple things like transfers. Uh, but it gets more complicated when you want to do DeFi. So a lot of DeFi protocols will offer permissionless incentives for bots to come and do useful tasks uh like liquidating bad debt in lending protocols or um you know bringing the prices between two exchanges into line and we call this the incentives maximum extractable value or MEV. And sometimes also users like you and me who are transacting in DeFi will create me.

For example, if they create a price movement with their trade that actually just creates an arbitrage opportunity for a bot. Um now, uh DeFi transactions that generate or compete for me are more complex than simple transfers. Uh they need privacy because if you publicize the fact that you're going to make a trade, uh someone else could just frontr run you and give you a worse price. Um and they also cause a lot of spam. So because there's a lot of competition for these juicy me rewards, uh only a few of them will actually win and you end up with a lot of failed transactions.

Uh this caused some big problems for Ethereum kind of five years ago uh where MEV uh was causing all the spam that was cluttering up the chain and users were getting wrecked by front running in the public mem pool. Uh so in 2020 Flashbots introduced a sidecar called MevG that augmented the standard client with off-chain auctions to sort of resolve the competition for MEV. And it created the first way to privately submit transactions to miners to avoid frontr running. And it also introduced what we call revert protection uh which prevented failed transactions from landing on chain. Um and this was great for me bots um because they didn't have to kind of uh pay if they didn't win an opportunity.

It also cleaned up a lot of spam on Ethereum from those failed transactions and overall it actually increased the value of blocks for miners uh because bots were not pricing in the risk of failed transactions. So you know okay great mission accomplished we solved block building. I can leave. Uh, no, there was still a problem which is that users didn't have any real privacy or commitments from miners. So even if MevG kept your transactions private and didn't frontr run you, miners could still just peek at everything and reorder you to steal me.

And in addition to these issues for users, this dynamic also created a huge risk of centralization in the protocol. Uh, because extracting ME and optimizing transaction ordering is actually really, really sophisticated task. So more sophisticated miners would likely, you know, do better, capture more me rewards and then that would just centralize the network more around a few sophisticated parties. Um, so in 2022 during the merge, Ethereum introduced a sort of new mechanism called proposer builder separation. And as the name implies, PBS separated the job of broadcasting or sorry of sequencing transactions, i.

e. block building uh from the job of broadcasting blocks to the network uh previously done by miners now by validators. Um and this quarantined the core protocol from the centralizing influence of MEV because all proposers can regardless of how sophisticated they are they could equally and freely access the next block. Flashbots uh helped implement PBS with a protocol called MEV boost. And a key part of this implementation is a new actor called the relay which provided the the privacy and commitments that untrusting builders and proposers needed to coordinate safely.

So the relay kept blocks private so proposers couldn't steal me and then it validated submissions from builders so that the proposers didn't have to trust them. And this kickstarted a market of block builders. So because now anyone could permissionlessly compete to build better blocks, we got a lot of people um in this market which was cool. That also drove more innovation and competition in block building. So blocks built with MEV boost in the last year uh generated something like I think four four times more execution layer rewards than blocks without it which is very cool.

Um and it also largely helped curtail centralization in the validator set. Uh Vitalik gave a really nice talk actually that touched on this at ECC last year. You should check it out. Uh so again mission accomplished. Uh no PBS did not solve centralization in ME.

It just sort of moved this risk from proposers to an external market of block builders. And while anyone could, you know, technically compete in this market, that doesn't mean that the distribution of power ended up being equal. Um, two years after Mev Boost was introduced, 90% of Ethereum blocks were built by just two teams. Uh, and it's really interesting to look at why. Uh, you might think that the winning builders are the ones with 100x better software.

Uh but actually we've kind of benchmarked these and there's not a huge difference between a builder winning 1% of blocks or a builder winning 50% of blocks. Uh the real reason is order flow. So over the last 2 years u monetizing order flow has been one of the biggest metas on Ethereum. Wallets and apps generate tons of value in the form of the priority fees that their users pay to land transactions on chain. And the apps and wallets want to monetize this.

So they started making deals where they would exclusively send their order flow to specific block builder in exchange for sort of preferential treatment and refunds. And a great example, classic example here is sort of the deal between Banana Banana Gun who is a top Telegram bot and and Titan Builder. Um but the the sort of specific way that these orderflow wars played out, the exclusive deals was really bad for the market. Uh it became really hard for new teams to join and innovate and compete. It sucked time away from existing teams that wanted to actually just innovate on new features, not do business development.

And because exclusive orderflow deals were so opaque and discoordinated, uh builders often competed against each other in these really efficient ways. So, uh there's a very funny cycle here where builders would overpay to land more blocks so they could kind of qualify for access to more order flow and boost their rankings in these leaderboards so that orderflow providers would pay attention to them. Uh but then in subsidizing their bids, the builders just actually drove the cost of inclusion higher for users. Um ultimately I think that the sort of underlying problem here is that we're bottlenecked on trust. The only way right now for users, apps, wallets to get the privacy and commitments they want like refunds or just not being frontr run is to make social agreements with the teams running the infrastructure.

And the problem with social agreements and trust is that it just doesn't scale. you know it limits the number of participants in these markets which reduces competition gives people worse outcomes. So I want to talk about how we move forward um how do we remove the barriers of trust and exclusivity that really limit block building today. This is a question we've been working on for years at flashbots and I kind of wanted to share a few key breakthroughs that we've had recently. Uh ultimately we think that minimizing trust in block building starts with better technology.

uh we really need infrastructure that keeps data private uh make sure that builders are upholding their commitments and is really really fast. Uh we think there is a type of hardware which can meet these constraints. a sort of new type of machine called trusted or not a new type of machine although we're using it in ME in a new way a type of machine called trusted execution environments where TE I think some people actually were talking about them earlier but basically TEES gave you this machine attestation uh that specifies exactly what software someone is running right so they prevent that person from you know peeking at sensitive data sent to it as well and this basically lets you run verifiable confidential off-chain infrastructure like block builders on completely untrusted machines uh you don't have to you know trust the operator or make special deals. You can just remotely verify that people are following the rules. Um and we've had recently in the last year a lot of success at applying these for production me use cases like block building.

Uh we've created a network of TEES that is optimized for building Ethereum blocks. Uh we call it builderet. It's verifiable, private and very fast. Um, today Builderet is building uh 35% of Ethereum L1 blocks. I was actually showing this number a week ago.

It was 25%. So, it's growing very quickly. Um, and we're really excited about this and also about the potential for Builderet to upgrade block building on Ethereum. And, and there are the few things that I think just make it really special. Uh, first of all, you will notice there are multiple teams running nodes inside of BuilderNet.

Um unlike legacy block builders that are operated by a single team, we can safely let many different people who maybe in the limit don't trust each other uh run tees in builderet and just have them basically attest they're running the correct code. Uh in practice, builderet has three operators today. Flashbots, nethermind, and be forbu. Uh and this multioperator model has a few advantages. So with multiple operators, builderet is much more robust against faults.

If nemine's node goes down, you know, builderet will still build blocks. Our ultimate aim here is actually to make it fully permissionless for anyone to operate a node. And Builderet is also more resilient to censorship. So nodes can customize their compliance policy. Even if one operator is filtering out certain transactions, another one can still include them.

And part of what makes this all work is something that we call orderflow sharing. Um all the nodes in Builderet share the transactions they see with all other nodes in the network. Uh this actually wasn't possible to do safely before um because sharing order flow with untrusted parties exposes users to frontr running. Uh but TE solves this uh and it kind of kickstarts a virtuous cycle where sharing order flow reduces the startup costs to running a builder because you don't have to make deals anymore and the limit you just spin up a node. Um and that kind of increases participation in block building improving resilience and competition on the core product rather than you know spending all your time making these deals.

That in turn creates better markets that more people want to use which drives more order flow into the system. And finally, there's a one other special property of builderet which is that it enshrines a permissionless refund rule. So everyone who sends blocks to builderet automatically gets paid back for the value they contribute to a block. So instead of these sort of approximate fragmented uh handshake deals, we've created a protocol level rule uh that neutralizes the exclusive order flow deals and and we've refunded about a million and a half dollars uh to date. Pretty cool.

Uh altogether, I think this is a big step forward. You know, if you look at the problems from earlier, builderet neutralizes exclusive order flow wars through orderflow sharing and permissionless refunds. And it reduces single points of failure by making it easier for more parties to participate in the block building process. Uh trust doesn't scale but builderet can. Um a coordinated refund mechanism can offer better execution than exclusive deals and a diverse set of operators can be much more resilient than a single team.

Uh and this is also really just the tip of the iceberg. If you remember how we defined block building at the start, which was this um there are actually many many many parties under the hood who are determined in how trans who are involved in determining how transactions get included in blocks today. So I think block building is increasingly not this monolithic entity we're used to thinking of but a process a market that includes you know searchers solvers fillers orderflow auctions sequencers builderet's first application has been to address the sort of immediate market structure issues in L1 block building um but we intend to use these same rails to upgrade a bunch of other parts of the stack so you know te searchers t solvers te sequencers as a bit of a foreshadowing we're actually running a lot of these things already. Uh so what comes next? Um I think that builderet is secretly the key to accelerating a huge amount of the things we want on Ethereum right now.

Uh I am just going to throw a bunch of ideas at you. Uh each of which probably could be a whole talk. Uh but hopefully this grabs your interest enough to make you come collaborate with us and all the other great people working on these things. So okay, let's look back at the discourse. the sort of big questions, three big questions in particular that I think a lot of people are asking today.

Uh question number one, how do we beat Salana? Uh I think it's not a huge secret that Ethereum feels competitive with other L1s like Salana. Uh no shade to Salana. I actually think this is very productive, makes both of us better. Uh but builderet is one thing that can help Ethereum differentiate.

Um if you have been paying attention to Salana lately, uh you know that the validators are doing tons of sandwiching. Uh Ethereum on the other hand has long been ahead of the curve on this by developing things like PBS and now with TE based block building in builderet we can actually take this a step further and get hardware level guarantees that users are not being sandwiched. Uh we can also use builderet to upgrade the entire Ethereum gas market uh to be second price. Uh which if you don't know basically means it will be a lot easier and cheaper to use. And this is actually a really big deal.

No other chain I know of has this. Uh finally, one other thing that I think Ethereum is really betting on is being the platform that other people want to build on. Um like other institutions or apps. Uh earlier this week we heard, you know, Robin Hood is launching an Ethereum rollup. And here's the thing, apps really care how their users get executed.

Um they don't just want to write smart contracts. They actually really want to control or have a say in the infrastructure that it depends on. And I think Ethereum has two unique paths to doing this. uh both of which we can actually make even better with builderet. Okay, option number one is basically to launch your own rollup like Robin Hood did with purpose-built block space for your app.

Uh take uni chain, you know, customuilt block space for trading or world chain block space that is prioritizing, you know, transactions from verified humans. Even outside of Ethereum, you know, this is a very clear industry trend. Apps like Hyperlquid have these purpose-built uh blockbased chains that prioritize things like cancels and Ethereum's rollup ccentric roadmap is uniquely you know situated for this uh compared to more monolithic chains like Salana. But a common concern is that these apps rely on a centralized sequencer or a very restricted set of parties doing the sequencing often because that's actually important to making the app work. You know, it's it's hard to uh do really fast cancels and trading if you're making a bunch of requests all around the world.

Uh but I think that still begs the question of how we can at least improve the guarantees for users. Uh well one big and easy improvement is to just run the sequencer in a TE. Uh then you can confirm that they are verifiably doing what the purpose-built chain was supposed to do. Um and it's so if Ethereum is kind of the best home for purpose-built block space then builderet is how we make it credible. Uh okay, option number two, I said there were two, is to basically customize the block building process on a shared chain like Ethereum L1.

And unlike Solana, Ethereum actually has a very programmable block building process. We saw that crazy uh sort of chart, the Sani diagram of all the people involved in it earlier. And the reason for this is because we've outsourced the whole thing, the entire block building process to an external market through PBS. And apps are already taking advantage of this. In the last few years, just as an example, Cow Swap and Uniswap X introduced off-chain auctions where solvers or fillers will kind of compete to fill trades and give users better prices.

And this whole auction again runs before the trade lands on chain um as part of the broader block building process. But but again, this kind of comes with these trade-offs. So now you have to trust this new off-chain actor which might introduce new permissions or privacy leaks, right? Uh builderet can help with this too. Um, it can give apps verifiable confidential rails to run highquality off-chain auctions in parallel with or I actually think in the future sort of embedded in the block building process itself.

And this can really help reduce dependencies for apps that don't want to operate extra infrastructure could maybe also improve the competition in their auctions. Um, so imagine if unis swap x, you know, uh, use tees to let more people compete in their permissioned RFQ. Uh, TLDDR, Ethereum is for apps and institutions. builderet helps them run better. Okay.

Uh big question number two, how do we scale Ethereum? Uh this is obviously top of mind for a lot of people. And well, one of the biggest limits to scaling today is actually ME. Uh I gave there's a longer talk that touches on this earlier in the week, but really quick TLDDR. Basically, spam from me bots right now are eating half the gas on major Ethereum roll-ups.

uh because fees are cheap, the mele is private, and so bots are just hammering the chain with tons of speculative transactions hoping that maybe one of them will succeed. And there are two essential parts of solving this. Uh first, we need an off-chain ME auction. Uh if you remember, we had the same spam problem or as a very similar one on L1 many years ago, and we addressed it by moving the competition for ME to an off-chain auction, sort of the predecessor to today's block builder. So an efficient block building process like the one we've been developing in builderet is actually key to reducing spam.

Um second we also just need to make searching more efficient in the first place. Uh so that the bots send less spam. And if you look at it actually the best way to reduce spam is to give people more information so that they don't have to guess and they can just send fewer transactions overall. Uh but this kind of begs the question how do you give more information to the bots without exposing your users to front running? And great news TE's also solved this.

So we figured out a way to let bots run completely private code on completely private user data without either of the two of them or the operator of that T leaking data. Um and there's way more if you're interested in this, but basically the first T searchers have been live for many months. They've actually already captured a portion of the atomic ARB market on L1 and they're really key to improving the efficiency um and reducing uh spam on rollups too. So, TLDDR, the path to scaling Ethereum is through T block building primitives. Uh, all of which we are already developing in builderet.

Pretty cool. Okay, I said there were three questions. Third and final question. How do we improve the user experience on Ethereum? Uh, I think a lot of people really want Ethereum to feel fast and interoperable.

Well, one of the main ideas to make Ethereum feel faster is to give users pre-confirmations before their transactions are included. But how do you know the confirmation is valid? Right now the best answer is you trust someone centralized sequencer maybe use reputation maybe you have people stake a bunch of money well what if we just use tees instead uh then you could just verify the software that your sequencer or builder is running you wouldn't need a lot of collateral or trust and we've actually built this it's called flash blocks it's been used by several roll-ups today already um we think it maybe could be used to help power L1 pre-MPS in the future um a cool implication here by the way is these TD nodes are interchangeable you can just kind of swap them out doesn't matter matter if it's a Flashbots T, a Uniswap T, your T, your T, right? Um, so long as they have valid attestations. Um, so it's actually pretty straightforward to add a rotating leader election to a T pre-MP solution.

Fun fact. Um, and on top of making Ethereum feel faster, uh, builderet could also be one tool to help improve interop between them. So, uh, just as an example really quick, if you look at op style interop, it basically has these two steps. Uh, first transaction lands on source chain. second that ends up initiating another transaction on the destination chain about one to two blocks later.

Um but unfortunately this means you know you have to wait for the second transaction that's like two to four seconds. Uh well if you have a TE block builder uh that you could use flash blocks to give fast pre-confirmations of the state updates from the source chain to the destination chain. So now the second chain could start including your transaction in like a couple hundred milliseconds instead of seconds. Okay. Another interesting thing here is that the destination chain uh needs to get updates about the latest confirmed state on the source chain so it can accept interrupt transactions.

And today the basically the solution is you run a full node for everybody you're interoperating with. Uh but you might not need to do this if the chains are running in TEES because then you can just verify the TE proof way less work. Um those are some of the lowhanging fruits. I also think there's a lot more we can do with a really well optimized TE block builder to improve interop because if you think about it in the limit synchronous composability between chains is actually just a question of how you can coordinate their block building process. So basically as we keep pushing the envelope in tight interrupt between chains I think it will start to resemble a lot of the problems uh that we've already had to address in high performance TE block building.

you know, how do you parallelize parts of the process, resolve conflicts, handle paranoral reverts? TLDDR, it's all about block building. Um, and I am optimistic about this because I think Ethereum has consistently been ahead of the curve with things like PBS and now builderet. Um, so we have the tools to address some of these big challenges ahead. Uh, but we're just getting started.

So, I hope you will come and join our collective of researchers and developers working on these things. Thank you. [Music]