[**Danning Sui**: Execution Quality and Orderflow Landscape](https://www.youtube.com/live/uHdIfX3ZDE0?si=YGN1gyqEweAa4jiv)


---

**Summary:**

This talk reviews the evolution of DEX trading, from early orderbooks to today’s solver-based systems like CoWSwap, 1inch Fusion, and UniswapX. It examines how DEX frontends, aggregators, and wallets have increasingly privatized orderflow, with around 70% of DEX trades now routed through private channels.

The talk also highlights the increasing amount of spam on high-throughput L2s like Base where cheap gas enables bots to flood the chain with non-reverting transactions that consume large amounts of gas without producing useful outcomes.

---

**Transcript:**

I'm going to talk about the state of the market reviews of data on DEX, MEV and Orderflow. I want to lay out the context of what this looks like for the DeFi space. How it has evolved in terms of the DEX trading design and solutions over the years, and who has the most of the DEX orderflow or user adoption in the frontend as of today. Then we'll look at the trend of orderflow market privatization. And then, we'll look at a study that we did earlier this year with some other researchers to compare the execution quality across CoWSwap, 1inch and UniswapX, and then we’ll review some trends about who is using DEX trading and for what reasons these days. Lastly, I would like to highlight a study about L2 MEV.

This is a diagram of how DeFi evolved from 2016. Basically, in the very beginning of Ethereum, there were naive orderbook DEXes, where you have a taker and a maker, and a relayer to match the orders. And if there's a match, then this order is sent onchain. Initially, even the order listing was sent onchain, which is highly gas inefficient. And the matching is also inefficient. Later on, when Uniswap came out, they kind of solved the issue of liquidity by allowing anybody to just put an asset into the pool and the pool defined the price.

AMM became popular, and numerous forks of Uniswap emerged. And when there's more liquidity that comes around, people realize that liquidity is now kind of fragmented again and we need to aggregate them to have the best price. There could be good prices sitting across a pool and if I'm only using one pool, my price impact will be big. I would want to access the middle point of multiple pools.

And so aggregators came out around 2020. 0x shifted from an orderbook DEX to become an aggregator together with 1inch. They're starting to talk to professional market makers as well who were saying “Hey I can actually give you a better price than the AMM orders. So if you introduce our inventory through this thing called RFQ (Request For Quote), I can basically provide you a better price than what you can get for your user via AMM.”

More and more institutions became interested in DeFi trading and providing liquidity. Now they're everywhere. Wintermute is one of the biggest, SCP is another MEV trading firm. They both provide liquidity as market makers, and they also take arbitrage to clear the stale liquidity from AMM. They also build blocks sometimes and they're deeper and deeper in the stack.

Then CoWSwap came out and they said “If aggregation theory works, then why don't we just aggregate on top of it? We can also outsource this work to a set of people and introduce some auctions to make those people compete. So instead of us having a routing algorithm designed to check across AMM and trying to aggregate them, why don't we just let a bunch of people compete to give us the best routing?”
And so that's when the ‘solver-DEX’ design came out, 1inch Fusion and Uniswap X launched it right after CoWSwap started to see adoption. Another thing I want to highlight is MetaMask. They are often called an ’aggregator of aggregators’ or a meta aggregator, similar to CoWSwap, but the way they are meta-aggregating is really just by querying a bunch of aggregator APIs.

They just compare on top of fixed routing rather than having an actual auction. It's really interesting because the wallets basically just see that users are not really price sensitive. They’re basically saying “If we have control of the user entry point, why don't we keep them inside of our app and charge them fees?”. And users will actually still use them, even though it was a pretty high fee.

I think the trend this year is that more and more DEXs are trying to integrate down the supply chain and trying to sell their order flow to orderflow auctions, OFA. So basically when users' transactions are moving the liquidity in the pool, it's very likely causing an opportunity for arbitrage. If you expose this order information to a set of trusted searchers, they can likely submit a backrun on top of it, extract some value, and return some of this to the application or the user.

That’s what the DEX with OFA idea is all about. So MEV-Blocker and Flashbots MEV-share are some of those. We're also very excited about all the Uniswap v4 hooks that are coming up this year.

So a quick quiz here in the room. Who do you think has the most retail adoption today of all the DEX frontends?

Audience: CoWSwap

Okay. Why?

Audience: Because it’s integrated.

I thought you were going to say because it moos when you make a trade.But to clarify, the retail adoption is measured by a deduped version of how much volume was requested from the frontend.

It's not directly querying liquidity pool volume and it doesn't include any MEV bots that are using the routers or liquidity. So the answer for me is yes, it's CoWSwap. I want to caveat the data here though, is that we are undercounting 1inch Fusion. But CoWSwap is still bigger overall, even after including 1inch Fusion.

We have this website called orderflow.art, trying to show the landscape of how different layers are integrating with each other.

One interesting structure I want to highlight, which we mentioned earlier, is that the OFA integrations are getting bigger and bigger. The solvers are getting the order and they're not sending to the public mempool, they're sending to some private services like MEV-blocker.
There are many other OFAs to send user transactions privately rather than in the mempool and then to block builders. Sometimes they directly work with block builders.

If we measure the trend over time you could see that over a two and a half year horizon, the number of transactions in private mempool on Ethereum has been growing and growing. As of today, around 40% of transactions are in the private mempool. Meaning you see it onchain, but it was never seen in public by any of the nodes.

If you break it down, we can see that some OFAs are taking a pretty small fractions of market share. The bigger share sends it towards block builders because they want to get exclusive orderflow to get an edge in their blockbuilding competition.

Another way to look at this is to look at the subset of the most valuable orderflow, which are retail DEX trades. There has been obvious growth around last year from people down the supply chain who are talking to all these frontends to get their flow. So today, around 70% of all the transaction volume on DEXes is going through a private mempool.

If you look at the priority gas fees people are paying across all these transactions. We kind of call them MEV value. You’re going to see that the mempools is dry already. Less than 12% of the MEV value is floating around. All the valuable transactions are already getting privatized or commoditized. That's one of the trends I wanted to highlight.

Next, I want to share an execution quality study which I did earlier with several researchers. One thing we see all the time is teams fighting each other on Twitter. CoWSwap is constantly saying “We have better execution!” And then the Uniswap data scientists will be like, #That's not true, because of this and that”. So we wanted to draw a conclusion of which one is executing better.

Another reason we want to do this is that users today still don't realize that the quoted price they see in the front end is not what they get in execution. The price you see listed here is not going to be the amount of USDC that will land in your wallet. Using this mental model from 0x’s blog post a few years ago, there’s a whole process happening when you access the liquidity. There are LP fees you have to pay to the pool, and there’s a price impact that increases with the trade size.

Additionally, when you send the transaction to the mempool, it’s possible that there is just a simple collision of trades. Other people are trading the same thing in the same direction, paying a higher gas fee. Then they get placed earlier than you, they get the liquidity first, which means you’re going to have slippage. Or, the MEV searchers saw a transaction in the mempool and they are extracting some value out of it and you're getting slippage. That's why we really want to look into this data.

One other thing to highlight is that even though all these like solver DEXes have similar designs of solver competition auctions, they have very different market structures.

Uniswap X has two market makers creating 90% of the volume. They are heavily dominated by RFQ inventory. I would think Uniswap X is pretty much one approach for the team, the companies to try to introduce more and more professional market makers inventory.

1inch Fusion has a slightly less concentrated proportion, with three market makers making 80% of the volume. And CoWSwap is a lot more decentralized, has a lot more participants and solvers are creating around 10% of volume each. The top market maker is only making 25% of the volume. And we know that's Wintermute.

So there's this running joke, right? Like Wintemute is everywhere, they hold everyone's token, and they dump everybody's tokens. Whenever the price goes down, they just tag their CEO on Twitter and be like, what's going on? What are you doing? People are pinning every problem on Wintermute. One silver lining is that on CoWSwap, the solver Barter is now making more volume than Wintermute in the last 30 days. We’ve got to believe in something, it’s not all just market makers and tradfi coming to DeFi all over again.

I want to highlight a few fundamentals as well when we look into the data. Even though market makers are so powerful and have competitive pricing, they do have a preference to avoid risk. If you look at this chart breaking down volume that gets filled by different type of liquidity. You can see that CoWSwap, 1inch Fusion and Uniswap X have very different distributions of what type of liquidity is winning orders on their platforms.

More particularly, this is showing that for long-tail tokens, are mostly filled by AMM. When we go down the chart to the ETH-BTC major pair, they’re getting fueled by market makers more and more. Market makers would not want to hold super random memecoins, which can go to zero in the next second. They will likely not fill that as a solver or market maker.

Essentially, we took six months of data and compared it was a very naive routing of a single pool, which I think is a limitation of the study, to compare it with executed trades, to see how the benchmark number compares against the Uniswap v2 and v3 pool look like for the solver DEX. So does these solver DEXes actually improve pricing compared to a naive routing in AMM?

We compared two pairs for the major USDC token pairs. We saw that all of them had a positive improvement compared to v2, which is the blue bar. The orange bar is compared to Uniswap v3, which is way deeper in liquidity and harder to compete with for a solver DEX.

This data might be a bit misleading if you just look at it and say “Oh, CowSwap’s better than 1inch and better than Uniswap X”, but you really need to control the distribution of the trade sizes and look at the trend over time.

A lot of the negative points here for v2, v3, benchmarks and the Binance mark out are all around super small trades. That's likely very noisy due to all kinds of gas estimation that we're doing in the study, which is another limitation I want to highlight. But you do see it across a trend of bigger trade size, the blue line goes up for all of the projects.

There's definitely a significant improvement of price execution when the trade size goes up if you're using a solver DEX compared to if you're using an older classic router on Uniswap. It's much less obvious if you look at the orange dots, which is a comparison against the v3, which is harder to compete with and improve.

Another pair we looked at is PEPE-WETH which is a representative of long-tail assets. But it's also big enough to have enough of a data sample. The results look super different because it's way harder to compete with v2, but it's way easier to compete against v3.

It all comes down to liquidity depth. For PEPE, most of the liquidity is actually in the v2 pool and that's why they're actually a lot better in terms of pricing. While USDC has most of the liquidity in v3.

The conclusion is, it's highly dependent on the asset's liquidity condition, and the execution performance can vary widely against the different benchmarks.

Overall, solver DEXes can improve the pricing better than naive routing. I personally think this study can be improved if you can simulate a very complex routing or you can simulate the 0x aggregator or 1inch aggregator. It's not just better than naive routing, but actually even better than modern aggregators. I'm looking forward to other people working on this as well, as trade size grows, performance actually becomes more and more prominent

What are the new use cases growing on DEXes?

When we aggregate the data we have and look at different front ends? A year ago I was looking at this chart and people were like “Oh, the new alpha is Telegram bots are taking volume”. But that was a very small part and it dies down really quickly, especially on this Ethereum dataset.

So did they really take off? Not really in terms of scale of the volume, because a lot of the Telegram bots are long-tail and small trades. I think most of the people who are educated enough about self-custody and wouldn’t really put more than $100 in Telegram bots where they hold your private key. But, I wouldn’t say Telegram bots are not a trend or are not happening any more, because that was mostly on Ethereum.

This activity and meme trading is definitely still happening, mostly on Solana. We do see that there were two giant spikes earlier this year because of the $TRUMP coin and $LIBRA coin. Now it's kind of dying down, but even this number is still making somewhere around like 100 million USD in annual revenue projection for just for the top Telegram bot, Trojan bot. It's still a pretty crazy profitable business for the bots.

Another trend I'm seeing is that there are more and more use cases and adoption for cross-chain bridging and swapping. All these bridge protocols have a pretty decent amount of volume. This one chart is the LayerZero chart, which is kind of sad, because after the airdrop snapshot the volume is just gone. We need to question the number here, how much is organic and how much is farming?

When I zoom in on this data though, say look at 1inch cross-chain swaps particularly, you realize that cross-chain swaps are still kind of just cross-chain bridging, because you see the token they're swapping is just USDC from one chain to another chain. It's more accurate to call them bridging rather than swapping at the moment.

One last thing I want to mention is that we're also looking into whether there is MEV happening on L2s, and whether Flashbots needs to do something to provide any solutions for those problems.

You would think there are a lot of studies showing that when you have a faster blockchain, there will be less MEV, which I think is proven to be true theoretically. There was this paper that's showing that your LVR amount is proportional to the square root of your block time, and MEV in terms of arbitrage opportunity will probably follow that rule. But there is another problem which is spamming. Recently, we published a study highlighting data, particularly for Base, that indicated spamming is becoming the bottleneck problem for L2 scaling.

If you look at this chart, it shows that the Base gas throughput is scaling up is around 35 million gas per second. But is that true? If you look at actual transactions on Base, then you realize a lot of the MEV bots are doing onchain searching because the gas is so cheap.

They're basically checking, using onchain math, using gas, checking if this pool and that pool have a price difference. Checking if they can extract an arbitrage, keep checking multiple pools, and iff there's no arbitrage just finish the transaction. These transactions do nothing, in the end they were just reading the state of the contracts and they’re just costing millions of gas.

They're also really good at hiding it. When you look at a transaction, it doesn't revert. You want to think it's an MEV bot. But if you actually simulate the traces, you realize they're doing a lot of searching onchain. And if we identify those transactions as spam and categorize how much gas was spent by them, this is what it looks like.

Half of the gas that Base is scaling is used by spam bots. So are you really onboarding 10 million users? I think this is a good question to highlight for them. So we basically converted this chart to be a chart called Effective Gas Throughput, which comes down to 20 million gas only, when it's saying it's 35.

It's not a problem just for Base, it's a problem for all the prominent high-throughput L2s, so that's what we're seeing for those. And this was a GIF, but you can see that for one block this is how the spam transactions are just coming in as a red, around later position of the block paying lower gas as well.

MEV and the Limits of Scaling is a blog post we posted recently, highly recommended to give it a read. And if you are looking for a job, we're looking for people who are interested in this space.