Season 4 | Bonus Episode
In today's Bell Curve, we are joined by Stephane Gosselin, the founder of Flashbots and Frontier Research. We discuss why we need to reframe MEV, MEV burn, app-specific order flow auctions, user intents and EIP-4337, why centralized sequencers are fine, enshrined PBS and more! Stephane is a legend - you don't want to miss his unique insights. 
- -
Timestamps:
(00:00) Introduction
(01:58) Reframing MEV: From Allocation to Wealth Creation 
(09:36) MEV Burn
(12:37) The Evolution of DEX Architecture and MEV
(22:44) Order Flow Auctions
(27:30) The Case for Centralized Sequencers
(38:56) Share Sequencers and Cross-Domain MEV
(43:37) User Intents and Limitations of EIP-4337
(49:21) Block Building's Race to the Bottom
(54:45) Optimistic Relays and Enshrined PBS
(1:07:05) Frontier Research
- -
Follow Stephane: https://twitter.com/thegostep
Follow Mike: https://twitter.com/MikeIppolito_

Subscribe on YouTube: https://bit.ly/3R1D1D9
Subscribe on Apple: https://apple.co/3pQTfmD
Subscribe on Spotify: https://spoti.fi/3cpKZXH
- -
Resources:
Frontier Research
https://frontier.tech/

MEV Utopia vs Dystopia by Stephane
https://flashbots.mirror.xyz/bqCakwfQZkMsq63b50vib-nibo5eKai0QuK7m-Dsxpo

Don’t Overload Ethereum’s Consensus by Vitalik
https://vitalik.ca/general/2023/05/21/dont_overload.html
- -
Disclaimer: Nothing said on Bell Curve is a recommendation to buy or sell securities or tokens. This podcast is for informational purposes only, and any views expressed by anyone on the show are solely our opinions, not financial advice. Mike, Jason, Michael, Vance and our guests may hold positions in the companies, funds, or projects discussed.

Welcome back to another episode of bell curve.

Before we jump in quick disclaimer, the views expressed by my co-hosts today are their personal views, and they do not represent the views of any organization with which the co-hosts are associated with nothing in the episode is construed or relied upon as financial, technical, tax, legal, or other advice.

Before we get into it today, just want to give a quick shout out to this season's sponsor, Rook.

Close to a billion dollars worth of MEV has been taken out of users' pockets, and that's just on Ethereum, and that number is only getting larger, unfortunately.

Rook thinks that it's time for a change, and they've built a solution which is going to automatically redirect that MEV back to where it belongs into your, the user's pocket.

So you're going to be hearing all about them later in the show.

I'm huge fan of this team and what they're building.

This is a little bit of bonus content from the MEV season today.

Today, he is the founder of Frontier Research, which is putting out some of the best research in MEV.

In a past life, he was a founder at Flashbots and the chief architect over there.

You might know him as the Ghoststep on Twitter.

I think that might be the best intro I've ever done.

Maybe I'll save the silent clip and just send it to people.

This will be a little side hustle for myself going forward.

But we've got a lot of ground to cover today.

I want to give you a little shout out here.

Frontier puts out, far and away, the best content in terms of MEV.

Whenever a new piece drops, all of our analysts are always talking about it.

It sets off some lively debate for at least a week.

And I kind of want to use this as a chance to kind of go through some of the topics that you've outlined.

So maybe we can kind of start at a high level and talk about the transaction supply chain of MEV and just how it stands today, how it's changed over the past year since you were at Flashbots and wherever you want to take it.

Well, first of all, thank you so much for having me on.

I haven't done really media since I left Flashbots, and so I'm excited to sort of get back into the open and start talking about these ideas that I enjoy talking about a lot.

So the MEV supply chain or the transaction supply chain, I published initially this post, it was in April of last year, around the time of the DevConnect conference in Amsterdam.

We hosted, or I should say, Flashbots hosted a conference there called Mev Day.

And I sort of opened the conference with a presentation that was titled, you know, MEV Utopia or Dystopia?

And in this presentation, I outlined sort of these new sets of roles that I think are becoming very common in the vocabulary and the language that people use in blockchain protocols now, and especially on Ethereum, by the time warrants, you know, very popular at all.

So the MED supply chain, I think a lot of people recognize now is there's users, they have intents, they communicate those intents to wallets or applications.

Those help them put them into transactions.

Those transactions are put into some kind of transaction system, whether it's an order flow auction, a transaction pool.

Searchers see these transactions and they augment them with their own transactions.

Those make their way to block builders or sequencers, those put them into blocks and then propose them to the validators and to the consensus of the system on which they operate.

So the last year, since I sort of published this article, I mean, we've seen a lot of excitement and innovation happen at every layer of the supply chain, which was really the goal of framing sort of blockchains in this way.

I think for me, it's also been sort of a personal journey, like going from trying to sort of define the way that these systems work in a fairly simple and linear way.

And 1 of the things that I had in there was the question of how will value flow?

Is the value going to flow sort of towards the inside, so towards the searchers and the block builders, and they'll sort of aggregate a lot of the value?

Or is the value going to sort of flow towards the outsides, towards users and validators?

I think for me, what I've sort of come to realize is that that was actually the wrong framing of the question.

Like this question of the MEV supply chain, how is value going to be redistributed amongst the supply chain?

Is it going to tend towards utopia or dystopia?

And the correct 1 is actually like, what can we create using sort of this model of the industry, what is the additional value that this allows us to generate?

And so it's sort of reframing it from a wealth redistribution, which is sort of more 0 sum.

And I would associate more with a lot of the questions that we have around MEV, to 1 of wealth creation, which is actually like, how do we deliver systems that are better for users, that allow us to generate sort of genuine value for the technology that is cryptocurrencies.

And I would sort of frame that 1 more as transaction supply chains.

So how are we sort of enabling these systems to settle the kinds of user activity that is most beneficial.

So that's sort of my introduction of my journey and exploration of the supply chain over the last year.

I'd love to sort of dive into any of the components of it and see sort of where has the energy been and where they're going.

Yeah, let's dive into that concept a little bit.

The first thought that I had when I was hearing you describe that is sort of this trend in MEB of returning MEV to users, right?

From the intents, which we'll get into, that they sort of create and preferences that they want to express onto a chain.

Right now, the vast majority of that value flows all the way down the transaction supply chain to the validators.

Right now, there's this big trend of returning that to users.

Is that kind of what you're talking about?

Or expand on that concept a little bit of moving away from sort of this allocation mechanism to a wealth creation mechanism.

So there's sort of some very simple mechanisms that we can think about that like manipulate this value distribution, right?

Like, oh, we're going to add in an order flow auction, which allows us to sort of claw back a lot of the fees that are being paid through transactions to validators and instead, like, send them back towards the user.

So, you know, there's this whole area of research and I published an article about it, you know, the design of order flow auctions, how do you design them in a way that's incentive compatible and sort of helps return value to the users.

I think the other place where a lot of this discussion is happening is for like Uniswap LPs, right?

Or, you know, you could say Dex design as a whole.

How do we design better decentralized exchanges that don't leak as much value, either from users swapping and being targets of information leakage through sort of sandwich attack, or LPs being vulnerable to sort of slow price updates and uninformed liquidity provision, which leads to LVR and sort of large arbitrage opportunities between between the decentralized exchanges and centralized exchanges.

And a lot of the conversation there again is how do we redistribute the value away from these arbitrage players who are bidding a lot on gas to be able to pay for the block space and more towards these liquidity providers.

Maybe like a final 1 is the recent conversations around MEV burn, right?

How do we sort of redistribute value away from these like large spikes that are sort of random to individual validators and more towards the network and sort of spread out the amount of that value.

I think a lot of these conversations completely miss the point that we're trying to solve with all of these blockchains.

We're trying to create systems not just that move money around between different actors, but rather are genuinely more useful and are going to be able to attract sort of the next billion users.

And developing mechanisms that allow for those better user experiences isn't just about redistribution, it's much more about expressibility and like, what can you actually do with this infrastructure?

And I really think that's where the limitations of our space is more so than any sort of, you know, protection or extraction or any other framing of it.

So can we, I would love to get your opinion, you know, on each of those things that you just mentioned, sort of in reverse order, love to just get your thoughts on MEB burn and you know, that sort of dichotomy between different stakeholders.

The way that I think of MevBurn and sort of value that it brings, well, I think it fits within like a fairly complicated question, which is like, what is the future of proposal builder separation on Ethereum?

What is the future of roll up adoption and sort of MEV activity on layer 1 versus layer 2?

MEV burn is kind of I think a desirable property when you have a system like we have today where all the activities on layer 1, all the economic activity is on layer 1, there's a significant amount of arbitrage and like sandwiching activity that is highly contested.

So you have high contention rates, or you could call it contention rates between, at every block for having sort of a specific state update.

And what that causes is basically all of those opportunities are being paid out to the validator through block builders and MEV boost.

That's somewhat destabilizing to the consensus of the chain and also centralizing to the sort of distribution of the nodes.

The reasons are that you have a strong incentive to pool your stake to be able to have the yield that comes from all this MEV activity that is highly variable.

So my chance of getting a hundred ETH block as a validator is very small because those don't come very often but I can have the same expected value if I join a validator pool that's able to spread it out.

Exactly the same economics that we've seen on Proof-of-Work Ethereum and are sort of playing out here as well.

Burning in this case sort of creates less of an incentive to join into larger pools because there isn't sort of this high variance value that you're missing out on as a solo staker.

It also means that, you know, now this MEV value is getting burned, which is, you know, great for, for, for ETH stakers.

There's some economics here that's sort of above my head and like understanding.

I'm not going to, I'm not going to LARP as like knowing what the economics of Ethereum are.

But anyways, that's the argument as I understand it.

There's a great, if you want to really nerd out on this and get super in the weeds on it, Justin Drake gave a good talk on this on Mevcanomics, which you can just look up on YouTube and sort of get his take on it.

I also want to get your, before we move sort of up the stack, you mentioned sort of the future of rollups on Ethereum and how Mev's going to look up there.

But before we do, I want to talk about, because yeah, you've written about sort of DEX microstructure, you've talked about order flow auctions, and that's just pretty cool, interesting content.

So maybe could you talk a little bit about sort of your thoughts on the evolution of DEX architecture, and maybe touch on, take something like Uni-V3 or something like that.

What are the sort of leakage points of MEV and how could we design DeXs in a way that will be better for users?

So This is an active area of research and exploration for us.

I should say, I should give a shout out to my research collaborator here, Ankit, who's been doing a lot of the research on this particular vertical as well.

He sort of helped classify a framing of MEV, which is called EV ordering and EV signal.

And it tries to sort of provide the basis for analyzing how a lot of these MEV opportunities are being targeted in various different mechanisms.

And so it's quite a useful framework for doing this kind of analysis.

Now, in the case of Uniswap, the way that we would think about this and the value leakage is, let's say you have a user that sends a transaction through the transaction pool, through the mempool, and they have sort of a slippage tolerance that allows them to get sandwiched.

Well, EV ordering says, you know, any bot operator, any searcher can basically take the transaction, sandwich it, and extract some value internal only to the system that is Uniswap.

Another example of this would be sort of doing arbitrage between Uniswap and some other exchange on-chain.

So that's certainly 1 piece of the puzzle and where sort of the value comes from.

There's a second piece of the puzzle, which is defined as EV signal, which is actually the value that comes from this activity happening on within the environment of Ethereum, but also taking into consideration the sort of external state to Ethereum and its mempool.

It looks at the state of centralized exchanges like finance.

And so here you have a much more rich sort of field where you have arbitrage between these chains, you have arbitrage between the centralized exchanges and the decentralized exchange, and you can also measure as part of this sort of arbitrage between the LPs of Uniswap and Binance, which is the second biggest source of leakage for the protocol.

It's what's sort of described when people talk about loss versus rebalancing.

It's the same sort of value that's being exploited here.

So to your question about how do we do better than Uniswap v3, well to do better we need to solve those 2 things, those 2 leakage, the ordering that's based off the slippage and the leaked value through Signal.

The design space for here is still being explored.

I think 1 of the insights that's quite clear is that it's really difficult to have price discovery on a system that is not very fast and has high fees.

And so we're sort of in this situation now where the price discovery for a lot of these assets is on Binance.

And this leaves all the LPs to be sort of targeted by toxic flow, basically arbitrage traders that are more informed and are able to get those price updates reflected on-chain.

It's like, you could think of it as like a very inefficient decentralized oracle where you have a lot of LPs that just put money and like constantly get bad execution and offer this service of having a price on-chain to anyone else who wants to trade against it.

There's been sort of a lot of innovation, a layer up the stack on like RFQ systems, right?

So looking at, okay, well, clearly, the on-chain price isn't the best 1, right?

It's a delayed price that can be up to 12 seconds stale depending on the situation.

Is there an opportunity to actually use this price as the best bid and offer, have it as sort of a backup price, but provide the opportunity for market makers to improve upon this price.

And that's essentially how a lot of these aggregators work today, right?

So they'll say, okay, well, we see that the current price of this swap on Uniswap is X, let's put it out there.

And if anyone can provide a quote that sort of beats this execution, then we will settle with them and provide some level of price improvements to the users.

And so moving up outside of sort of the chain execution into this environment has some benefits.

It does seem to provide better execution for users, but it also means that all this now it's sort of private market makers that have all of the liquidity and that's being kept off-chain as opposed to being placed on-chain.

I think there's still a lot of innovation happening.

The biggest question is, is it going to be sort of a decentralized exchange on layer 1 that's going to be the place where price discovery happens?

Or is there some designs for exchanges on layer twos that will allow to actually have price discovery happen within the decentralized context and bring that liquidity back onto the chain and not necessarily have it out of the chain.

Yeah, I think it's all, you know, I think the point that you make about basically LPs overpaying the arbitrageurs is going to be a pretty interesting 1 in the coming years.

In some sense, that's sort of a political argument as well.

The other thing I would point listeners to as well is sort of FBA sort of designs for exchanges, like sort of the cow swaps of the world, where very similar to kind of an RFQ based system, There's sort of a baseline, right, of execution that traders are willing to pay.

And then I forget, there's a, I think solvers is the word for this particular actor, but they sort of go out into the world and figure out a way if they can execute at a cheaper price and that surplus is actually measured for traders on the exchange.

And that was we'll get into this later, but I think that's kind of 1 of the first experiments in terms of intents, you know, an exchange utilizing intents.

I mean, there's there's so many interesting components that go into this.

1 of them is you don't necessarily need to only source liquidity from a single chain, right, once you have this.

The intent is to source liquidity from everywhere.

I think 1 parallel that's not explored enough is actually how frequent batch auctions at the transaction layer are fairly equivalent to payment forwarder flow systems.

They're doing very, very similar activity in terms of auctioning off the sort of ability to solve it or to settle a set of transactions, but it's looking at it from a different time perspective.

And there's some sort of auction theory around this that means that you might actually have sort of more collusion in a system that does frequent batch auctions than a system that does payment for order flow.

This is not by any means sort of a question that's fully settled.

That's 1 of the questions that we look at when we analyze these systems.

So it all has to do with the type of mechanism that's used.

So this is the same case for order flow auctions, that sort of auction off individual opportunities.

You have, in the case of EV ordering or EV signal, a fixed amount of opportunity, but with some probabilistic payoff.

So you're not actually sure if you're going to get the execution that you want.

And so when you're providing a bid, you know that, okay, the max that you can make is x, but you have some risk factor on your execution.

And because it's a single winner auction, there's only the person that bids the most that actually gets settled.

And they always end up like winning because they overpay for the opportunity that they have.

The end results of this is that you have sort of an incentive for all the participants in this system to collude and underbid and say, like, we will all bid with like 20% discount to what like our real preferences as a way to make sure that we don't, you know, lose money.

You know, 1 can argue it's not like a stable equilibrium and you'll have like bidders that overbid, but the thing is that those bidders will eventually run out of money.

And so if you have infrastructure that you can run, and that your cost of capital is sort of within the range of your opportunity, the amount of opportunity that you capture, You can run sort of a discounted bidding profitably for a very long time.

Well, it will be interesting to explore, honestly.

Very interested in the FBA sort of design space.

And I think Calswap, which by the way stands for coincidence of wants, which I didn't understand, didn't know until recently.

We've also talked to Henry Devalens this season, and he is building Penumbra on Cosmos.

Similar FBA sort of design with privacy in mind first, which is super cool.

I want to transition here to your thoughts on order flow auctions, because you put together a great piece sort of exploring the design space there.

And you just started to say 1 of the highlighted 2 of the differences between payment for order flow and order flow auctions, which is sort of the time, the time component, and then basically the individual sort of transactions that's possible to do in an order flow auction.

But maybe I'll just turn it over to you, Carte Blanche, here.

And can you sort of explain what the design space looks like if you had to categorize the different types of order flow auctions and maybe highlight how it differs from payment for order flow?

I think, you know, the top line takeaway for that piece is a lot of these systems tried to solve 2 things at once.

They tried to solve for price discovery and for settlement sort of together.

And it sort of confuses 1 thing for the other.

That ends up being expressed in like a lot of challenges and how these auctions are constructed around trading off value, like basically best price, you know, return for the user, best price, best execution for user and reliable inclusion.

And so based off of this insight, I sort of propose that there's actually 4 different types of OFA designs.

And they can be sort of put on a 4x4 grid between if they're application specific, so only for sort of a swap use case or generic for any transaction.

And if they settle based off of individual transactions or based off of a group of transactions.

And we can put basically all the different designs of order flow auctions, whether it's an RFQ, sort of a cow swap, a PFOF model, a sort of a MEV blocker style system, or more of a MEV share type system inside of this, you know, 2x2 grid.

Each of them have their trade-offs, each of them optimized for something slightly different.

And I think crucially, I don't necessarily think that any of them solve the problem that like users actually want to have solved, which is like, you know, being able to trade in a venue that, that is, they're gonna like high confidence is going to give them a good price and timely execution.

I would actually be that that 2 by 2 matrix is, I think, a really good way of thinking about it.

I'd be curious just because if you have thoughts between sort of generalized order flow auctions versus maybe more specific, like maybe in the future, like Dapp run order flow auctions.

If you imagine something like a Uniswap trying to run their own order flow auction, you can certainly see the incentives for them to do something like that.

And I would just be sort of curious how you would assess sort of the pros and cons of those 2 different approaches.

I am a strong believer in application specific order flow auctions.

I think that's likely the way that you optimize for user experience.

I think there's like some arguments you can make that you can have like a specialized order flow auction built on top of a generalized 1, but I don't see a way to really do this that doesn't introduce a lot of trade-offs.

And so My intuition says that certainly the ultimate best user experience for users who want to do the use case of swapping or providing liquidity trading in some kind of way, which is 1 of the few proven use cases, I guess, of Ethereum today, those will end up in a specialized order flow auction model.

I guess a separate question is like, is there still demand and, you know, value added to long tail order flow auctions or generalized order flow auctions, maybe.

But I see them more as block space aggregation platforms and sort of discovery of contention on state access, then I see them about returning value to the users or providing better price.

So I wouldn't see in those system, it to be very successful to build an order flow auction that returns value to users.

I do think that it's likely that those systems are more used to minimize the block space fees paid by individuals trying to interact with the chain.

So speaking of fees, I would love to kind of get your thoughts on sort of moving up the stack, you know, to the rollup layer and how MEV works there today.

You know, I'd be, I'd be curious right now, obviously not a secret, right?

That the, all the rollups use 1 centralized sequencer, which has big implications for MEV.

And a lot of these rollups today sort of view MEV as a source of revenue, right, for their rollup.

You know, Hasu's argued for something like PBS that exists on Ethereum main chain today in sort of the form of MEV boost.

Eventually it's going to be enshrined, but something like that needs to exist on sort of layer 2.

I'd be curious just to get your sense of how you're thinking about rollups these days.

And I would love to sort of get your kind of high level thoughts on how MEV is going to work up there.

You said every rollup except for 1 uses centralized sequencing?

I think all of them use centralized sequencers.

I'm still not convinced that that's a bad thing.

I still, like, I don't think it's a done deal to say that actually first in first out sequencers on a layer 2 are a bad thing.

The argument that we often hear when this is mentioned is well, then it creates a latency game and causes centralization around like a specific region.

And therefore the rollup is susceptible to the censorship rules and regulatory rules of that, of the region in which the sequencer is deployed.

But still, the question is like, is that actually bad?

Because I think it's like a bit of a controversial or like,

I'm very curious to see where you go with this.

This is something I've been trying to disprove.

So so you know, if people want to add me on Twitter and argue with it, I would love to.

So if you have like a maximally decentralized layer 1, that probably means like a layer 1 on which there is very little economic activity, like it only settles, you know, data, and there isn't sort of contention rates, right, there isn't this sort of demand for settling with a, you know, on a specific position, you only have congestion rates.

That's sort of how we, we, we, we, Ethereum has been designed.

When you start to think about economic activity, then you have all the issues that comes with MEV and the latency games that are sort of ensued.

If you have an architecture where the layer 1 is only settling sort of data blobs, and there isn't contention, and you have all the activity inside of layer twos, it reduces the the centralization pressure on the layer 1 significantly.

Now, okay, how is it any better for the layer twos to sort of become centralized?

Well, if you have some way to have cross-chain messaging between these layer twos that is quote-unquote censorship resistant, right, You have some way to like push messages out from the layer 2 back into the layer 1 or maybe for it to be interpreted by some other spin-up of that layer 2 somewhere else.

You might be able to, when I say like interpreted by another layer, layer 2 somewhere else, I mean like a fork of that layer 2, then you might be able to be completely okay with having a centralized rollup that's targeted by regulations.

Because it remains censorship resistant, it remains, you could say, non-custodial.

See all Bell Curve transcripts on Youtube

BONUS EP: Stephane Gosselin on MEV, OFAs, and Frontier Research