NEAR Whiteboard Series with NEAR | Technical Deep Dive Sessions, 8 Oct 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Whiteboard Series with NEAR Protocol | Ep: 48 Tarun Chitra founder of Gauntlet Network

Description

This Whiteboard Series episode dives into MEV (maximal extractable value). NEAR Co-Founder Illia Polosukhin (twitter.com/ilblackdragon) talks to Tarun Chitra (twitter.com/tarunchitra), founder & CEO of Gauntlet Network (gauntlet.network). Gauntlet is a financial modeling platform focused on risk management for decentralized finance.

Illia and Tarun unpack MEV––a term describing the maximum value that can be extracted from block production in excess of the standard block reward and gas fees by including, excluding, and changing the order of transactions in a block. The video breaks down how MEV works on PoS Ethereum, where validators are now responsible for transaction ordering/inclusion, and how the MEV environment is evolving across chains today. They also explore where NEAR’s unique protocol design differs from other blockchains and the opportunities this offers to NEAR developers and traders.

~~~~~

For the latest from Gauntlet, check out:

Website: https://gauntlet.network
Twitter: https://twitter.com/gauntletnetwork
Blog: https://medium.com/gauntlet-networks
Research: https://gauntlet.network/research/

~~~~~

Join NEAR's community:
Website: https://near.org/
Reddit: https://www.reddit.com/r/nearprotocol/
Discord: https://near.chat/
Medium: https://medium.com/nearprotocol
Blog: https://near.org/blog/
Twitter: https://twitter.com/NEARProtocol
GitHub: https://github.com/near https://github.com/nearprotocol
Dev Docs: https://docs.near.org/
Create a wallet: https://wallet.near.org/create
Apps on NEAR: https://awesomenear.com/
Learn to Build On NEAR: https://www.near.university/
Grants & Funding: https://near.org/grants/
#Blockchain #FutureisNEAR #NEAR #nearprotocol
#nft #dao

A

Hello, everyone uh Amelia coming back with whiteboard series here from New York I have tarun from Gun Hill Network and we're gonna go in to the Mev yeah.

B

So I think if you followed crypto Twitter at all nowadays, you know there's a lot of fighting over what qualifies as Mev, um but first of all, let's write out what the acronym is. The acronym has changed. It actually stands for maximal extractable Value, not.

A

Because no more miners right, because a few weeks.

B

Ago, yeah exactly so this uh this! This is what we're talking about, and so you know what you know. The problem is one hard to Define.

B

Two, it has the um the quality of porn, which is you know it when you see it.

B

This is from a famous Supreme Court case, where that.

A

Was a definition nice.

B

um But you know you know it when you see it unique to blockchains.

B

Because in tradify you don't really see it, you, you see your version of it, but the exchange doesn't tell you really what everyone else is doing. Yeah.

C

B

It's hard to see if, if someone is extracting something from you, but.

A

Yeah, it's like.

B

Porn because you literally go on block Explorer.

A

Yes, but but you have like Robin Hood, for example, which was kind of doing that right, like you, send all your trades and they would match them and kind of do.

B

It, but you don't actually see how much money.

A

C

Of this exactly you.

B

Don't really see how much Citadel made out or or jump or whoever made out of it right. So the interesting thing is that here in blockchains, you kind of there's this porn aspect of it in that you can go, run the transaction and see how much it should have cost or taken, and you can go around the you can go look at what happened actually happened and you see a difference so clearly, there's there's, there's some sense in which uh uh difference between expected cost.

B

And realized cost.

B

So you know this is I, send five tokens.

B

I'm supposed to get five dollars.

B

But I get 450.

A

But then you also said you're fine with this when you were sending here. Otherwise this would have failed.

B

Correct correct, correct, correct, but the thing is in a lot of applications right, an application will guarantee you something you know. It'll say like spend five dollars on Robinhood and you'll, get five shares and.

C

B

You say yes, and then you see that you got four 4.5 shares will be tests yeah. So this is the you know it when you see it aspect right now, there's this big question of, like is everything Mev, so um you know I think the The Flash buff people who who do Mev auctions on ethereum they may I think they take a lot of Liberty with saying even like our Dex arbitrages is Mev, but in some ways it's not really it it's.

B

It is just intrinsic to the system that there's some uncertainty over a block and whoever gets uh kind of to to pick what the block ordering is kind of has some power over the users.

B

um So instead I think the a sort of more kind of economic way of defining Mev is sort of. We have a sequence of transactions that are sort of sitting in the mempool. We have a sort of value function which takes in a ordered sequence of transactions.

A

You want to find such permutation and.

B

You, you want to say like what's the max permutation, so this is the permutation.

A

That's probably going to be the most formulas in whiteboard series, those those far right.

B

Or sorry, rather, you should say this is the max value m. This is the expected value e expected value. Permutation.

A

So but here's here's a question for you: yeah: why? Why don't blog producers by protocol not forced to just randomize this order.

B

Yeah great question: um one reason is that they cannot come to consensus on the actual transactions they have. So, for instance, if we have a network of validators I'll just draw them. As you know, this network graph, these are the validators and then there's the network of users. Actually you mind if I use the.

C

B

So we'll make the red as validators.

B

Blue is users, and so you know the user is sitting all over the internet and they submit transactions.

B

And, as you can see, they submit them to different validators. Maybe they submit them twice, just to get make sure their transaction gets in, and the problem is coming to consensus on the set of transactions you have so on the the set yeah tx1 a adds, a bunch of latency but B. If, if you know you basically have to ensure that people are not dropping some of these transactions, um but also you basically have to do what's sort of like a ranked vote.

B

So if, let's say, let's say even if we agree that these are the transactions the set of transactions, the ones we're going to include, then we have to say: okay well, I saw I got this transaction first and I got this transaction last, but the problem is: there's no way that everyone gets the same, ordering sure yeah right and uh effectively people have to vote on what the ordering is. They could vote on the randomness, so they could vote on like.

B

But the thing is that the randomness only chooses the permutation the the way each person is storing the set is actually up to them.

A

No but I mean you can say, like publish a set, first, get your consensus and then, through that also vote on the randomness like, for example, the randomness is Source from.

C

A

Signature on the Block sure, then the execution happens already.

B

So we need to agree in advance on something.

A

B

What the set is and like what like lexicographic ordering is like what alphabetical ordering is, so that sure, and then we draw.

A

C

Permutation right.

B

Yeah, uh a higher latency B. You still have the set problem of like how do you guarantee that people are including uh what they say because, like well.

A

People, well you mean that, like this is where the like tips and transaction fees are coming from. You actually like get included yeah exactly so as soon as you do that you kind of violating the and.

B

B that there isn't some side Channel agreement where a minor is accelerating certain transactions versus others right, which is what flashbots effectively you know does like. If there's an auction, you pay them. If you win the auction, you get to go ahead of everyone. So a it's bad B increases latency, C sort of like increases communication that the validators have to do makes them more susceptible to get ddosed. There's like lots of sort of things that make it harder.

B

So, instead you know one, you know the way things work on ethereum right now is there's effectively a centralized coordinator, called flashbots and, um let's say, I'm a I'm, a strategic user, so I'm going to call the Strategic users like gonna, give it.

C

B

With some glasses uh and the Strategic users, like hey, flashbots, here's, my transaction and I, want it to be first and I'm willing to pay five dollars more than than what.

C

The transaction costs.

B

And let's say that wins then flashbots broadcast this transaction.

A

Do all those all.

B

The validators and who subscribe to it and they effectively agree like a cartel to execute the transaction such that that occurs, and they only get the payment. If, like the the.

A

Things this thing passes.

B

Yeah- um and so you might say- hey I'm getting robbed here, because this guy's putting his transaction first so that, when I put in the five tokens, I only get 450 because he put in his two tokens before and took out my 50 cents, um not great. But arguably it is more efficient in the sense that people who are strategic, who, if, if this wasn't there, they would try to spam the validators with many many transactions. And you would actually get worse performance in.

C

A

B

Right so there's this trade-off between the economics of incentivizing, the Strategic users to to pay more versus the spam on the network, and that's sort of this intrinsic trade-off in Mev. You know: there's people who spend their lives trying to make these things called Fair ordering protocols that do really complicated like you've kind of the things we talked about. Like everyone keeps voting on the sets of transactions first, then they vote on sort of the randomness and then they execute so far. None of them have worked without either some amount of centralization.

B

They have sort of some fallback mechanism um or they just take.

A

Forever take forever yeah right, so so some consensus, algorithms, the.

B

Bigger the big problem is no one has actually proven any formal results about this. That say something like hey there's, always this trade-off of like you, either get the highest information process through your blockchain.

C

B

You have to slow it down to like guarantee some sort of fairness and there's sort of some well.

A

Raid off there's a laden super difference like everything is.

B

A

B

It's sort of latency personality.

A

B

For um sort of cost and sort of fairness, measured in some notion of like how how much you're, respecting like the the intent of the user, because, like one way of thinking about it, is we have blockchain block one block, two block three: the user sees the state of block three and it sees that it's like five dollars for coin.

C

B

And they're, like oh I, want to buy five dollar coin and so I'm, like yeah I'm gonna, send my transaction here. Take my money for five hundred dollars. But then you know of.

A

Course somebody comes in, and somebody comes.

B

In and they they they instead of getting 100 coins, they only get 90., and so um you know the the idea here is that fairness would say: hey actually, if there's 10 people trying to get the 500 thing instead of having one.

C

B

It 100 another person get 95. Another person get 90 split it evenly. That's the fairness. Access the latency access is like how fast is my transaction process? Yeah.

A

Just aggregate all transactions wait for a while exactly everybody chill and then settle and fairly.

B

Exactly and cost is, cost is a little bit complicated. Actually so um one thing that's interesting is like suppose we have two exchanges, so we have one exchange that shows five dollar coin and the other exchange that shows 450 coin.

B

What's the real price.

A

Well depends on the much you're buying and the exactly the pools exactly.

B

How much liquidity there is? What you're buying it's possible that the five dollar ship coin is horrible for you, because there's no liquidity and you push the price to ten dollars or it's possible that the 450 is actually worse. And so what is the natural incentive for people to do is to synchronize these because technically they have the same assets. So you know Buy Low sell High.

B

So in some sense your cost goes down by allowing people who are doing Arbitrage to go in front of you, because they're actually ensuring that the prices are sort of synchronized, and so it's as close to five dollars as when you saw as possible, and so that's sort of there's some sort of like surface in this 3D I'm about at drawing 3D things.

C

There's some sort of trainoff.

B

Surface between fairness, latency and costs, and there's this idea that, like hey, if I maximize Fair the the people who believe in Fair ordering they think that, if you maximize fairness, you're not trading off these other two but I. Think that, like one of the big thing, unsolved sort of mysteries is proving that you can never have all three of these I.

A

Think I mean the question is also because, for example, you build a blockchain where there is the swap is uh core part of protocol, and you can have like you aggregate also transactions, and then you execute swaps at the end of the block. Right like you, can have a more like the the transaction ordering, for example, does not matter in a block at all. Yes,.

B

So there are things like penumbra, which is a ZK blockchain, where they take all your transactions and they do a batch auction.

A

B

Batch auction gives everyone the same price, so it maximizes the fairness side, but everyone actually only sends encrypted transactions, so liquidity providers have to play this game or before the block they have to get, they have to basically submit how much of their liquidity they want to keep for that block and then the order flow gets executed. Privately against that. So in that that's another example of someone going way on the fairness axis and but.

A

B

Latency access is also quite low, they're, like they're, like they're, high fairness, low latency, but high cost, because the batch auction worst case is high cost right. But.

A

But let's say you don't do it private, you do it in public. You say we aggregate all transactions and instead of executing them in order, we we bad shows a holds the orders, but.

B

Now your consensus rule has to interact with an application for the.

A

Order but then let's say we build.

B

A

You know the fair low, latency low-cost swap chain.

B

Right but my point is you can't drive all of them to zero or, like maybe.

A

B

Some sort of like limit to how well you can do without making your consensus suddenly.

A

Well, I think I think that what will happen in this case is that the cost is lower on ethereum, for example, because there's so much more.

C

Lots of stuff yeah, like yeah.

A

And so as soon as you start optimizing just for swaps, a specific chain, you lose all the other functionality. And now actually your cost will be higher, because you don't have like flash like correct, flash loans and stuff like this to like actually or.

B

You don't have um routing where you have many dexes and I. Can you know I want to trade a to d, but I can I can split my trade.

C

B

And actually this gets to a paradox and a paper we wrote recently. That's actually pretty easy to explain that you can explain mainly without math by just some pictures.

A

Explaining something without math what.

B

A

B

I do my best, um so this is a sort of paradox and from you know, basically, in this paper we found this Paradox that the inverse of this Paradox that exists in sort of traffic modeling like when you're designing cities or designing roads between cities uh for amm. So the way it works is imagine we have four tokens.

B

um Unfortunately, there's not I, don't know in other languages, otherwise I would name them each a different form of, um but uh the we'll start start with the traffic version of this. So we have four cities on each Edge, so you want to go from a to d. So you know a b d e c d on each Edge is uh what What's known as sort of uh the latency. So let's say, there's a hundred cars that need to travel this.

B

If the the latency says X, then and 50 of the cars take this route, then the time it takes is one half and if it says one, it's it's like uh infinite. It's like a really large Highway it'll, just take everyone the same amount of time always, and so we have this thing where we have. You know kind of this really large Highway, really narrow road, really narrow road, really a large Highway okay. So you know, if you look at the equilibrium of this imagine everyone was like Hey like what's the fastest route to take.

B

Well, they would see okay! Well, my you know it takes me the same amount on each route and you would expect traffic to.

C

Sort of 50 50 right.

B

So you get 50 50 this way and 50 this way.

B

Okay, now imagine counter-intuitively that you add a road here that has latency zero.

A

Just imagine there's teleportation anyways.

B

Right and so actually, by the way when they're split, 50 50 the the time it takes. You is one half plus one which.

C

B

Halves right because half goes here so one half and then takes one okay now here, if everyone is selfishly optimizing, the fastest path is actually this.

A

B

Because if they, if they, if they took any of the other paths, it could be higher right, but here it's actually technically, you know x and x, whereas this is always two and X is less than two. So it's so basically what happens is but the. But then you get this weird paradoxical thing: if everyone takes this route, then there the time it takes them. Even though I added another road, there should be more throughput in the network. The network should be more efficient, takes them longer because, like everyone took this road, never.

A

Took this, so it takes time too yeah right.

B

So that's bad, however, you know what happens if I at now, let's suppose I add a 0.1 yeah a little Epsilon times x. It's not that's the only symbol. You're getting so now, as you make this like cost congestion cost go up. Then you'll see that people will revert to Distributing, 50 50 and like how much does congestion costs controls like how how people distribute themselves I'm. Sorry amms are the same way if I have four tokens- um and you know I have these edges that can be built.

B

It turns out that these people doing this Mev there's a thing called sandwich attacks which we can explain. I don't know if.

A

That's worth explaining I, don't know we can. We can yeah.

B

It's it's sort of this idea that I put in five dollar five tokens I'm supposed to get five dollars back at 450. and and the way it works is basically any trade you you put in you say: okay I'm, putting in five tokens, but I don't want to move the price more than ten percent and I say the percentage, but, let's suppose someone whose strategic notices that actually, if they put me first in the block, I'll, only move the price five percent, then they add another transaction.

B

So this is say attack they call it a sandwich attack. It's not really. I I find it hard to call these things. Attacks, they're, they're sort of like it's.

A

A sandwich trade.

B

It's like you're, a bad dumb user like you, should have been bidding in the auction for the same spot, but like you're willing. You know that that's what you get for being unsophisticated.

A

um And just saying order, books are better.

B

I'm, not I'm, not Anatoly, so right, so this guy adds in a trade that pushes the the cost at five percent we put in my trade, which pushes the cost up five percent, and then the attacker sells. The trade sells back uh so attack ourselves back their their amount that they put in, and so this sequence of Trades can be profitable, um so they they pushed the price up in front of me.

B

I get a worse price, I get less tokens out, and then they sell right after me, and as long as they can guarantee these are executed in that order. They can take advantage of me now. The interesting thing about this example is the sandwich.

B

Attacker is adding this congestion Epsilon, yes in some sense, and what actually is kind of a very non-intuitive thing is that without the sandwich attacker there's some scenarios where, based on the different relative liquidities of this, where actually you have a worse cost for executing end trades over this network than you do. If you have the sandwicher, so the sandwich are actually they cause congestion, but they kind of in the in. In the same way, they make people's routing go back more optimal.

B

So there's this trade-off between like yes, the users who take that edge are getting sandwich getting a worse price, but the overall flow across the network can be better and there's a trade-off, be congestion and routing costs.

A

So but but usually, this kind of trade is is uh a taunt Atomic and so like, even if it is like, if it sounds like, let's say my trade is, you know, A to B, to D, plus a to c to D in one transaction, yeah and so you're, saying that like they've done this on like b a b and on AC, for example, yeah and then sold it.

A

That's still in.

B

Some cases that it's not always true yeah, it's just that there are examples where, if this thing is actually sort of low liquidity- but these are higher.

C

B

You and there's some ratio between how much liquidity is here you you can basically get the scenario where the sandwich actually helps avoid people taking the low liquidity path, because the sandwiches profit is highest on the low liquidity path. So people's routing changes around I see I, see but.

A

This is like, presumably when, when somebody does doing the search, they should probably try to predict how much they're going to be I.

B

Mean instructors someone's doing the search if they were, if there were a way to do this in ethereum? You can't do this, but I think if, in anything where you have multi-transaction flash loans yeah, you could actually just flash borrow B and C. Add liquidity here and then actually make the path cheaper, but you'd have to do inject. The flashlight inside the atomic.

C

B

It would be annoying so you have to do this. Multi-Transaction Mev of like flash loan, execute trades, unflash loan right and it's hard to compress those atomically. But if you could do that.

A

Well, if you had resources, for example, that are multi-blocks.

B

With some guarantee yeah, then you then you could actually just never have any of these types of problems and just assuming there's like a flash loan of old place.

A

Capability, yeah yeah: this is actually something we we wanted to do, because our stuff is always multi-block anyway right and so we've been thinking how to have like pretty much resources or walls where.

B

You actually walk me through how the design of the, how that, how? How do you guarantee kind of atomicity we.

A

B

A

These each each contract is almost its own chain or its own roll up or its own sub chain, or its own power chain or its own. Like.

B

We'll name it even within the chain. What I'm saying is.

A

But wasn't it well, that's the point we don't have like each con. So let's say you have um like uni swap.

A

So, on near uni swap has all the pairs in one contract. Okay,.

C

A

So then, what's called refinance, and so then all the trades are just linearized around that. But if you want to do anything else like, for example, actually send tokens to the user that is done in the next block. So the token itself, like you, know, usdt dot near or whatever it's called is did like. You actually send you deposit money in one block, then you do a trade and then you withdraw. You know I mean in a way, but.

B

You can't do that all in the same block.

A

No okay, because the each cross country call happens in the uh the.

B

Cross contract column is the token.transfer call yeah.

A

B

Usd2 yeah exactly yeah.

A

So it's all based on callbacks you're, always calling the uh like. It's.

B

Almost it's message: passing am I guaranteed. My callback doesn't revert like what happens on reversion.

A

uh You get a call back, you.

B

Get a cup, but then you have to wait another block right for that callback.

A

B

C

B

The exception ever take more than a block, no space. You can guarantee.

A

You guarantee that, well, they have gas limits, so I mean you can write some code that will run out of the.

C

A

Yeah but yeah, so you kind of need to guarantees uh like.

B

If someone can't make an exception, that throws an exception, call back and like has this chain of like if they wanted to grief you because you're paying the gas for that.

A

uh Well, it says the user Pace like the whole transaction.

B

Is paid by the user right right, the user would, but the user could trigger this chain of exceptions that it keeps calling back in no.

A

No there's a limit for.

B

A

There's a fold the whole thing transactions.

B

Okay, yeah yeah.

B

So how would you do flash loans in near you.

A

Don't okay, so I mean you would have a uni swap having a flashlight capacity, or this is what we're thinking to have volts. So volt would be something that travels like that's on the runtime level and it travels between contracts and it can, and it has guarantee similar callback guarantees pretty much as uh so. You would like program a vault like you say: it's either settles in this way or gets returned right, and so you package, whatever resources you want in the world and.

B

If it gets returned, it has a per: do you have like a per block gas spend on The Vault, like as it goes through each.

A

Yeah you pretty much will like program, but, like you, pretty much vote will be like a piece of programmable data that travels around right. So right now, data travels around without like with with just a global but like what, if you actually can program it so that it can execute so.

B

So actually maybe this is. This is a good analysis question so like if we were to do a sandwich attack on near how what's the number of what's like if I drew, like you know, block height, h, block height.

C

B

Plus one Blockhead h, plus two well.

A

So right now I mean depends on how, like so.

B

Can I like batch the transfers like yeah.

A

You can deposit everything, trade, everything and then withdraw everything, because.

B

Like say, the attacker.

A

B

Does their trade they say, transfer a to the unisoft contract, this one says: transfer B the users yeah.

A

So you can do that in.

B

A

You can do that in one block: yeah, okay,.

B

A

Like an attacker would just already have all the all the deposits on on the contract, because you can have your internal balance on this contract. You don't need to.

B

Oh I see I, see I, see so.

A

B

Have just already transferred.

A

Yeah, it just put all the money in and then and then just when you see somebody executing you just try to execute around them.

B

You execute in the same block, then.

A

Yeah yeah, that's linearized, okay,.

B

A

Yeah, so within this you can have many many like call executions, pretty much happening within one block.

B

um Are those that but I forced some serialization around them right? Yes,.

A

And so this is, and this is where the fun starts so so each contract is its own quote-unquote like chain roll up whatever, but then they get batched into like actual shards physical shards, which are actually I, mean semi-deterministic and then and the cross-contract calls are getting batched and received from different shards and also like deterministic, but uh kind of determined by a previous block.

B

How do you guarantee that the validator executes the Callback correctly on the next block like are.

A

They that's right, yeah, that's just invalid State tradition, okay, so like pretty much the way it works. uh So this is a block right. So within the blogs, there's chunks, yeah.

C

A

Shards and then you have transactions here, and so this produces. We call them receipts, but this is just uh cross-contract calls and so they're getting included in the beginning of the next. So these things will get.

B

Included all right so there's like actually two parts. There's transactions with the block header of the previous block includes the set of things that must be called.

A

B

A

Something yeah, and so like.

B

How can you is it possible that the number of these things have to get called here? Take up the entire next block.

A

It yeah so there's like some there's, always some limit to include more transactions but yeah. So there's a queue, pretty much been built and it may take some time to like undo it. Yeah like I, mean.

B

The user doesn't actually have that. So, like Suppose, there was like a order book yeah.

A

B

I had a liquidation from a lending protocol that.

A

B

Out the whole order book, it's possible that if I wanted to like trade on that order, book, I submit a transaction I'll be behind. You know like: uh is there a bound on that time before or like? Is there a bound on the space per block? That's taken by the previous callback.

A

Yes, yeah yeah, so there's some space always for new transactions to come in, but.

B

Okay, so that's where you have some nav. So if I, if I know the the the the the the new transaction space yeah.

C

B

Say I know: there's a liquidation, let's say I'm the Liquidator, so I caused the liquidation happen and then I want my orders to get filled.

C

B

Just like add a new new kind of like order, so the order book such that oh these, these some of these, have to get executed in this block right, yeah non-zero number like like if I look at. Let's say this is like box based as like.

C

B

Some percentage that is callbacks from last block.

A

Yeah or like receipts yeah.

B

Before and how much and some new transaction.

A

B

What's the ratio.

A

I mean so there's a limit on the whole block, yep and then there's a uh minimum percentage for this I. Don't remember what it is but like then, then, usually it's you know less than, and so it fills in with the rest.

B

Yeah, this is an interesting way of manipulating a t-rop Oracle. So another form of Mev is like. Let's say you have a unisoft transaction and there's a you know someone's using it as an oracle price like they're, marking a loan on another contract.

B

um If you bought up, if you basically changed the price on use up, so you paid to move the price up for a certain amount of time you you will somehow keep adding trades here that are callbacks and you'll have to add some new transactions.

C

B

A

So yeah you can pretty much like try to push up the price, and while this gets.

B

I guess there's.

A

Yeah more stuff scheduled to execute yeah right.

B

Right yeah, that's kind of interesting and this. This callback, Mev I, don't think anyone's ever talked.

A

B

It clearly exists that like so, how does this ratio get chosen? Is that governance, or is.

A

That yeah, it's a configuration.

B

So values have to vote on change to.

A

B

Have you seen people taking advantage of this like in this.

A

Way like not just right now, I think the yeah there's some sandwiching. That happens, but it's also hard because, let's.

B

Say well, the sandwiching is capital inefficient here.

A

B

You can't flash like you, can't really flash on your sandwich.

A

Well, you can on a specific protocol Implement like uh application implementation that has internal balances. Okay,.

B

So so you're saying the Utah contract I.

A

Could put it has yeah well, so unislope contract is almost like it's its own chain where you can deposit money, trade and then withdraw money right, yeah like that's. Why, like they are extra chains right, yeah and you can do it in one transaction, but obviously it's less efficient than if you deposit do a bunch of Trades and withdraw.

B

Right but can I, can I flash and borrow as my deposit and then do the trades and then flash close.

A

Assuming it's profitable, not there's no implementation in the current, but in in.

B

Theory, yeah interior could okay that that, should someone should do that that actually I feel like that, would just make this process more efficient.

A

Yeah yeah uh yeah interior because, like this contract, actually has a lot of liquidity itself, so it actually can borrow like well. Actually it doesn't, it doesn't even need the real money. Yet you can actually literally print the money because it will be within one one. Yeah.

B

All fits execution yeah exactly exactly it should it should. You should be able. So this is an interesting question in a protocol like near where I can think of as like each app is almost effectively its own chain.

B

Although there's some sequencing, there's.

A

Still there's yeah there's, so there is a a pretty much synchronization that happens right. That's that's! Really what it's like! Our point was if it's completely async right, which was like a lot of original sharding designs, it becomes like completely unmanageable right right because, like as a user, you now have no idea. What's going.

B

On I mean you already still do have the.

A

Little you have some yeah, you have something like the user.

B

You need to like, like the the wallet needs to basically like resend, if it doesn't make it into the. If.

C

B

You're on high priority: let's say you get into the Callback, but you have no clue whether you'll be in.

A

B

Block or five blocks away, you have to keep layering new orders and try to get into this. um Yeah I mean it's actually kind of interesting to think about what the.

A

Because, in a way this is this is a more optimized version of a multi-chain world right like if you think of it like this, is.

C

A

Each thing is mode chain world, it's just. You have kind of a single right Machinery to do with it. But, like you have this problem across like say: you have Cosmos with IBC. You have the same type of thing right. You have like things, send you know sending money between. You can trade you can withdraw and so like a multi-chain mov, that's what neroes already have.

B

Right, so what? uh uh What kind of so so one interesting thing about multichannemuv another another kind of funny: little allegory, this one, this one again I'm trying to do all the ones that don't need an email. um This one doesn't really.

A

Actually, just I was thinking about this yeah. So actually is this: if this contract is refinance, adds A Primitive, specifically for sandwich attacks.

C

A

Where you can lit you literally say: I'm gonna borrow X trade like a to b or whatever some set of executions, and then, whenever somebody else trades, this I'll return. It.

B

This changes, the the the dichotomy of this unisol pool right. It's like, let's just say we have you, know R1 R2, which is like the reserves. Well, what are you borrowing when you're flash borrowing well you're, borrowing one side of this so you're, actually changing the price slightly yeah.

A

You literally just say: I'm gonna drop the, but then, but then this contract can just do it himself. They can just give everyone the minimal price, the minimal, like whatever the price that user said, the minimum they will be fine with it, can just give it to that and take all the money to itself right.

B

A

B

That's a great question: you should ask them why their token doesn't no.

A

No, but like whatever unit swap just did that like it will always give them whatever minimum price or or fail.

B

Sure so that's what the like penumbra type of thing does. When they do batch auctions yeah effectively, you could do it on the single transaction level. They just do it on a mini transaction level.

A

um But if we're talking about like generally yeah.

B

Yeah yeah yeah having the.

A

Contracts themselves, of course, that yeah, because.

B

A

Will like actually optimize this because right now, it said like.

B

The five percent- the problem is we're ignoring one user here, which is the liquidity provider and they are losing out on this potentially.

A

But no, but the contract can make money for liquidity. Provider Services like they can split the profit.

B

Right right, they.

A

Have to say in a way in a way like right now you have this, like you know you get.

A

I was like you know, there's there's a price curve right like this and you're saying you know if you if you're buying, then the price goes up if you're selling and the price goes up down, but the reality is you can just say this is like you will you'll be going from here instead, and this is your.

B

Yeah, the problem is the language that you design trades in unisoft. Doesn't give you a price. You just say the language is trade, size and slippage limits.

A

B

Doesn't give you a price that you pin to yeah um if you added the price, you could sort of do this, provided that, like it is true, this is sort of sort of like adding like stop loss orders exactly.

A

Well, yeah! Well, it's it's more like when, when you trade, you just say like I, want I trade A to B and I want at least acts of B.

B

Yeah yeah well, which is how slippage units are defined.

A

B

Is this is how and.

A

So you just get all these: the X yeah.

B

Yeah and so so the problem is the user. Might their transaction might fail many times because.

A

Like someone but but this is true right now too, if when somebody's front running.

B

Hopefully, yeah yeah, as long as you can enforce.

A

It so if, if there's efficient value, extractors like flashbot, they, you will always get X.

C

A

Otherwise, they're not efficient, and they can extract more so so in reality, you're already getting this except the money are go to. Whoever runs Searchers instead of right, LPS.

C

A

Why don't swap protocols just give it to a piece and their token holders? They have.

B

That they could I guess the the weird thing about this borrow this, like hey, I'm, I'm gonna, take some of the reserves and like use it to do. The sandwich. Taco sandwich is that it doesn't work at very low liquidity, so, like it long tail, assets are, um but then it also like doesn't work for really large trades because they will like wipe out it'll have to borrow so much of this. That will push the price like there. Is this trade-off between between, like how much you're willing to borrow to do the sandwich versus.

C

B

The trade size and it's unprofitable in some cases, but you're right that you could do it just on the trades it's guaranteed profitable for LPS.

C

A

um So it just solved the MV is that.

B

No, no there's other parts.

A

No I mean not this; this is the sandwich. We.

C

Just saw sandwich.

A

Text s pretty much all all deck. All amms should just always return whatever minimum user said and make money for LPS.

B

Yeah, that's definitely true: they should just sandwich themselves. I mean people have made. These I mean these uh uh MMS have done that, but for some reason they've never been able to get liquidity because there's there's this like liquidity, beginning liquidity effect of like people's wallets are already routing to you. If, like.

A

Yeah I'm not I mean there's Network effects, there's Network effects, yeah.

B

But like I think somehow so.

A

So so you actually talked about like originally and I know you wanted to uh flip another topic, but the you know defining kind of it's through maximum permutation yeah, but like things like twap and other stuff, that takes multiple blocks. Actually, not that yes,.

B

I guess so, so it's really the the the the the the the the thing here is Max or permutations Max over transactions. You can add Max or transactions. You uh uh um such.

A

B

You know the sum of all the transactions is less than the block limit.

A

But there's also a context of like you've been, can be doing this for multiple blocks. Yeah.

B

Yeah you you again, that's true, then you need to make sure this block limit thing takes advantage of that of value of mempool yeah, um so yeah, so it is, it is expensive. I do think these um the amount people have.

B

So one question you might say is: like suppose, I gave you a flash loan that, let's say flashbots offered a service where they lend you collateral over a block.

C

B

Let's say: you're a Searcher, so a Searcher is a strategic user who who does a sandwich or does a liquidation? Whatever flashpot says: hey, we'll lend you 1 million. You can do a bunch of transactions we just but we'll only it'll only be valid if we get back at least one million plus plus some interest reward yeah right, uh but but you can buy arbitrary amounts in a block. Now you could imagine just doing that repeatedly and the cost of this t-wop attack just went down dramatically right.

C

So the question is.

B

This is actually incentive compatible for the auctioneer, because it's actually better for them to be lending you money right, because they don't they actually don't earn any yield on right now. Yeah.

A

Because I mean yeah.

B

Right and for them this is free interest. So the question is: at what point does it become? Does all the value go from the Oracle LPS to the auction and my suspicion is in the long run, the validators always hold the keys and they can all literally, and they can literally always do this and there's no reason for them not to, and if you're accelerationist validators should be selling these types of services.

A

But shouldn't vow: they just be the oracles then, because.

B

I hold the keys anyway, yeah yeah.

A

Exactly may as well just they.

B

Should actually provide the data yeah yeah, so uh eigenlayer, which is like this thing for eth2 for restaking. So what it is is like, let's say: I'm an E2 validator I've staked, 100 East and someone is like: hey I'll pay you five percent a year. If you take your 100 Ethan these two and you restaked it in my Oracle protocol. So what that means is I if I draw it. It's like I have my 100 East.

B

I deposited into staking contract and now have a hundred steak teeth. I'll cut SC. This is separated from STS because that's slightly different and what um what restaking does? Instead, it says: hey, put your 100 Heath in the Oracle contract.

A

That will actually take it underneath, yes,.

B

And the Oracle deposits it there and then when you so so so the the withdrawal address that the main protocol sees he's Oracle.

C

B

The Oracle, but the Oracle is a smart contract that guarantees that, when you call withdrawal it atomically sends you back yeah. So now you basically get this pass-through where you can earn extra interest as the validator by basically delegating your stake to a contract. That does good part, put it in the main network, but then also earns fees from you know, whatever people calling the oracle so there's a sense in which that's the way, I think that all of these services and validators could offer like like lending.

B

You could basically make a lending protocol like this um that, where, where the amount that's allowed to be lent effectively is like the the the protocol say it takes so, let's, let's suppose we had a lending protocol version of lend they take 95 each and restake, and they take five eth and put in the lending pool, and the idea is maybe this is getting you. You know, seven percent fixed and the loans are getting. You like five to ten percent variable right.

C

B

You can choose how much you want to do that yeah and then you would have. Then the validator would have the capital to like do this to that Glenn yeah. But then the t-bop Oracle completely is ruined.

A

But but like what I mean, but you know if they go further, we just go to add this in on the protocol level, where.

B

This I mean you could add these as Services directly yeah.

A

B

Can which I think is any amount of anything right.

A

You just borrow any amount, like infinite amount of anything. If you return more.

B

Of it, hey, hey you win, 64 has a max button.

A

Well, you 256 to.

B

A

uh You can borrow as much as the limit of the.

C

Of the integers, yes,.

B

um But like I think this is actually what will end up kind of defeating a lot of like on-chain trading as oracles like I. Think, there's there's a lot a lot of people trying to build lending protocols they use Hue, operacles and the moment validators start doing and like eth has already gone. This way, proposer Builder separation, sort of screws over a lot of these um things that want to use t-ref oracles.

A

B

You might as well just shove it into the validator.

A

Yeah there was no attacks yet on all the stats that depends on two up. I. Think right.

B

A

Validators can do whatever right, like.

B

There have been a couple, and most of them actually just have to do with people's using bad t-rep settings. Okay, because, like obviously they make the window Too Short, it becomes really cheap.

A

They're really easy yeah.

B

um And so like yeah, I, I think uh another another kind of interesting liquidation game to consider is: uh oh yeah. Sorry, oh yeah, cross chain um cross chain, our cross chain auction design. So this is something I'm working on now and it's formalized formally. But like there's one example in there, that's kind of interesting. So suppose we have two chains chain, X chain y and there's a amm here that is trading X versus synthetic y.

B

And there's something here: trading synth, X y okay. So, let's suppose the price here is P0, the price here is P1. P0 is less than P1.

B

Let's suppose that there's validators no I should let me draw them as a little stick. Figures.

B

There's validators.

A

Same validators of people.

B

They might not be the same, they might be the same. Who knows, and they can do one of three auctions. One sell the Arbitrage so, like you know, tradex for since y go to Y redeem, it trade y for syntax go across back, so you can price the auction price. The bundle in x, price bundle in y.

A

I'm prize both.

B

Or sell the legs separately, yeah and depending on the liquidity and the gas cost model. When you do this trade here, if you'd bid on them separately, you're actually bidding on how much you're going to change your gas cost here, because you're actually pushing down the price.

B

Why so, if there's, this price difference is really big, then you actually can, as the auctioneer earn higher Revenue by pricing separately versus pricing in one or the other, and this is the type of stuff I think that like makes this cross chain Mev stuff, much more complicated, it's like actually, sometimes the auctioneer should be offering you very different sets of bundles bidding languages. Well,.

A

I mean this is where you like. Ideally, you have some options that you can buy to.

C

B

Yeah yeah yeah yeah sure sure I mean like Azure latency I'm, saying like that. There's certainly hopefully that but I for a lot of coins. That's not true right! Well,.

A

I mean so this is I. Think one of the interesting things in general from for any multi-chain stuff is: can you because, like there's a latency, you know L that exists here, and so what you want? You want some options for, like 2L plus you know some Epsilon that will guarantee you, whatever you're doing here, yeah and so having like a facility that gives you such options like efficiently priced again.

B

This is another thing that validator should be selling you yeah.

A

Especially if others are trying to cross all the chains right so.

B

So exactly so the difference in value so in a sort of no Arbitrage World, which is obviously somewhat idealistic. But it's not that idealistic, because the validators, if they're issuing the options, it should be no Arbitrage like like by almost otherwise they up that.

C

B

That they're, they wrote their code wrong, uh then, in that case, if there are options, these three should all be priced the same, because these two will include options.

A

B

But but my point is right now that if.

C

B

Look at the dexes that there's the the way muv is extracted is kind of um is is actually the the way people are designing. How you bid on these things, changes their revenue and I, don't think anyone's very much solved. That problem.

A

Yeah I mean I think that there's an interesting like if we're talking about validator right like right now, they're just running like a regular, node, yeah and so already with flashbots. That is, like you run something that, like uh you, know, proposes blocks. Yeah sidecar proposes block. So the question is like: should this be kind of a standardized facility to allow like, for kind of customization, of the almost like a block like block logic right like, for you know, flesh mints and kind of.

B

All these things and I think that's that's inevitable. This is this. Is this? Is the direction I think all the cosmos, Mev stuff is moving this direction and I'm gonna guess near Mev will move in this direction. This is where it's different than these, because eth is doing first proposer, Builder separation, which kind of complicates doing these sidecar things and because the block, so the proposer is the person who it gets drawn from the stake distribution.

B

The Builder is the person who proposes a block, and so you have this auction. Where the proposer is given a bunch of blocks, they can't see what the transactions are, but they're bidding on. You know a particular block, so it's like no.

A

They're, just picking whatever who pays them more pretty much and they they.

B

Just they should be, but I mean there's some weirdness here in that and I. Don't know why flashback people don't believe this, but like the proposer and Builder could easily collude by being the same person.

A

So there's like Builders, who propose the number of.

B

Builders propose a block. It.

A

Was a block, which is why it's confusing yeah.

B

A

The six proposers to propose a block, correct and then, and then the the validator. Let's just call them is the one who actually picks and so, like naturally, I mean. Ideally they should pick whatever is a pays them the highest.

C

A

But they can, if they are the builder, then they just.

B

There's something they're just lazy: well, they might want to choose the cheaper one because they have a bunch of their own. Well,.

A

There's a yeah there's like extracted value here, which is higher than this price, and so this.

B

Again gets at this fact of like Fair ordering is stupid, like this thing will never be compatible with this idea that there exists some Global like Kenneth Arrow, won the Nobel Prize for proving this impossibility theorem for ranked voting and like somehow. We still want to like shove, our shove ourselves into like pretending that that's not just like a fact of reality.

A

Since there's no fairness.

B

A

Destroying your world there's no.

B

Difference, there's a trade-off of fairness, yeah there's a big problem with machine learning too right now, big problem in both machine learning and crypto. So like I, actually think that's that's where the intersection of research is I mean that was like the fairness thing of like how much fairness do you, trade off for accuracy, slash, cost and latency and like yeah, very hard problem, but.

A

Well, that's yeah! That's why there's but.

B

I do think people will make a node where there's sort of like a DSL or that like lets, you like, say, hey I'm, going to use this mainnet contract as my restaking mechanism and like you, deploy these contracts like Oracle lending, whatever that basically guarantee the execution for the validator.

A

Yeah but well I, don't think you need the restaking contract. You just literally have some you just say like there's a facility of like pre-block plus block logic, yeah that everybody just the whole blockchain just votes that we're gonna accept yeah this piece of code to run and like in Cosmos. It's like everybody does that all.

B

The time well I think the restaking thing's kind of important, because you yeah yeah I, know the pre-block post block header thing. There is like a little bit that makes sense, but.

A

I think like in Cosmos. They just have that like they literally have that two pieces.

B

Of I think the interesting thing here is that the Lend contract may actually want to Source liquidity, not from the validators right. The Lend contract might just be like hey it's the normal Ave pool, yeah and you're, just using the.

A

Obvious, if you don't have just flesh mints, if you have flashes.

B

C

That's what I'm saying.

A

Like if you have just like facility to do whatever yeah I mean it's that that's where like well, it's at cloud and stuff like this but yeah, but like generally saying right, I mean this is where like. If we have, if we had resource programmable resources, then you can have a flesh mint in the resource where, like in x-blocks, it needs to come back, but.

B

I guess my question with the design you're saying earlier of, like I, have to wait. Another I have to wait each block.

A

B

The like, how does Flash mitt work because, like in Flashman but then the Flashman is stuck.

A

Oh well yeah, so this thing that's what I'm saying is programmable. There's like a piece of logic there that says in X blocks. This thing needs to come back and and call this function that will like flash pretty much minted I mean burn it, and so what it doesn't matter where this resource.

C

A

Up well, it's that's! That's where.

B

The guarantee by the runtime I see I, see I, see, okay, so so so in.

A

Runtime there's a new there's, a new pack like a new place where it records all the resources that are there and iterates over there and says. Like is any of the resources need to.

B

Come, but it also says, basically takes a lock on all these resources.

A

Exactly yeah this is this is this, is this is a mutex of resource that you can travel around? This is this is how we were solving the train and hotel problem. Yeah.

C

A

We have not implemented this because it is more complex, but actually move is doing this. This is where yeah it gets interesting, because move is actually programming, language that has the resources already yeah, which are like half conditions.

B

Right I do think the problem with this stuff is like the uniswap style, like Global State. Everyone is in the same pool one day yeah, you know, you're executing, doesn't work very well here, because you have to have many of these running concurrently for each user in some sense, which.

C

B

Be like a pain in the ass, so you need to build a stuff to like manage, like all the users like State, tripsing and.

A

B

When there's you know like.

A

The same block.

B

Like you have some rules.

A

But this is where, like usually something in unisol will just be one chain contract account.

C

A

Where you just synchronize, everyone, trade, it and yeah synchronize, but.

B

I just mean for developers, that's quite a bit of overhead if they have to think about that, they want to write JavaScript man. They.

A

Do that's why we have now JavaScript well in JavaScript, you do have callbacks, and you know if you're writing. This is actually. This is literally how microservices work yeah yeah. This is like you have each each contract on here is a microservice was a database and it receives incoming.

B

Calls and it has callbacks, but the question to me is actually more like the lock on the state. That's touched.

A

Like based on how you want it, it's a developer. Okay, that's why we want to ask we wanted to add some facility like a standardized log facilities, but we haven't is.

B

There, a way to like wait can I control whether whether I have a view only resource first mutable resource of like.

A

You as a developer can Implement whatever you want. You can literally resolve your own like blocks and everything so, okay.

B

Okay, because that yeah, that makes this seems like there's more Mev here honestly than there isn't even like I I know all the move. People are like. Oh, we have less mov like this actually seems like there's way more of like for.

A

The resources, the locks.

B

The sequencing, like suppose, I, have like five resources. Let's suppose, there's State one state, two state three and like calculation, one calculation two and we have five people, and they that you know, let's suppose we, you know, we have those sort of like complete, bipartite graph on this, and each of them is taking some path like one. That's doing that, one is doing that. One is doing that and and they have some, they have some contention right, like they they'll reach the same state.

B

The resolution of those contentions is either random, but the randomness is not really random there because it's you know the.

C

B

You know the validator code. You can figure out what the they have to come to consensus on it. So there.

C

Must be some deterministic.

B

Source of that, so you could try to basically simulate ahead. The previous, the the.

A

Rules, unless you, unless you're pipeline it, then it's the next block. So this this is where, like in our case, we actually have. We have a one bit of uh like validator, can only affect one bit of Randomness right by not producing a block which then actually I.

B

A

It perhaps their ability to do anything and- and we actually had a fully like fully non-biased vrf uh implemented but not launched.

B

Well, how come there's? No one thing I haven't understood, because there's so many benefits to application developers is why no one has a vrf the application Level, because, like like to some extent the thing you were describing earlier, where we're like hey, we will, you know we'll we'll let you borrow a little bit yeah before right. The contract.

B

Does it itself is a sort of form of like hey the contract is collecting its own Mev or, like there's application, specific Mev, but one of the things is like that's not quite easy to do right now, and one thing that would make it easier is actually having just like a vrf. You could call in your contract, that's tied to the consensus so.

A

So we have that because we have, because when you submit a transaction it gets included and then it gets executed. Okay, so by the time it gets executed, you already had something that was like a.

B

Consensus yeah I haven't made like numpy dot random for consensus. Randomness like I, want to sample gaussian I want to do the like. You.

A

Can do that on there.

B

I know, but someone should actually make the library on that because, like I feel like that, will actually give application developers a lot even more easier, easy control over like avoiding these. So.

A

You're, just saying like oh just give a random amount back pretty much within the range.

B

Right but with some distribution like yeah,.

A

It was it just like. Well, you have you, have a slippage and you just say: well, we're gonna, give you yes, random amount, yeah.

B

Yeah well so so we we wrote this paper proving how to get differential privacy, which is sort of like a way of you know you can show for these amms. You can't ever give pure privacy, even if you had a ZK proof blocking the reserves, because if you see the prices, you can always invert the prices, because the functions are determined yeah for how they change.

B

So the question is: how do you add noise such that you know if I add x, x, bits of entropy to your trade, which change your price by some amount of there's some bits of Entry to.

C

B

Price, how do I make sure that I I someone can't infer what your trade size was with more Precision than x, but then I also don't move your price more than y right. So there's, like you, know, there's like the trade size space and then there's the sort of you know price and you know I want to add Randomness so you're in here, but then I also don't want this. This thing to.

A

Be related right, yeah.

B

And so there's this trade-off and if you had different distributions, you can like add noise here with certain distributions to guarantee that this thing is bounded.

A

All right well, there's a project right there for someone to build I.

B

Think the library thing might be first numpy dot.

A

Random, well, that's what I mean yeah like I mean because there's a source of Randomness you just need to like then build distributions. Yeah yeah.

B

You can actually, you can just copy it. You.

A

Can actually run them on here.

B

Right, you could just copy that and just make sure it uses.

A

The correct to see code, so you can just run this.

B

You just need to make sure it gets well. The problem is your vrs 256.

B

My point- and you probably want 256-bit Randomness, so you might want to rewrite those.

C

B

Support u-256 but I think there's like lots of stuff like this unexplored territory, for people all right.

A

Well, lots to build, uh there's and loads to fix with existing protocols, so.

B

I mean look, he didn't think we'd have an hour of Mev content, all.

A

Right awesome thanks thanks tarun and uh please uh check out all the videos like it subscribe. You know the usual YouTube stuff that I don't know what to do, but thank you.

B

A

Mean I'm sure we can talk for.

B

A

Hours to be clear.

B

Like yeah I feel a lot of yeah no.

A