Hashing It Out
Hashing It Out

Episode 60 · 2 years ago

Hashing It Out #60 – Balance Research – Arthur Gervais & Dominik Harz

ABOUT THIS EPISODE

This episode Corey and Collin bring on the academics Arthur Gervais and Dominik Harz from Imperial College London. They discuss a new protocol that attempts to reduce the required financial deposits for various mechanisms in a way that does not reduce security. In other words, it is an attempt to help with over-collateralization, which could save investors a lot of money! As always, we get into the weeds to try and figure out exactly how it works, and where it is best used. Come listen and enjoy!

Links:

Donate to Hashing It Out!

Entering. Welcome to hashing it out, a podcast where we talked to the tech innovators behind blocked in infrastructure and decentralized networks. We dive into the weeds to get at why and how people build this technology the problems they face along the way. Come listen and learn from the best in the business so you can join their ranks. Welcome back to another week of hashing it out. As always, I'm your host, Dr Corey Petty, with my Co host, Colin Cuche. Say Hello, everybody calling. Hello, everybody calling. It's a little early force today, so Colin is a bit Karaggy. That's so energetic to the I'm fine, I'm full of PEP and vigor. I'm not. I'm not suckling on my coffee like heat of a mother. Like I'm full of it. I am ready to go. All Right, today's episode he brought some academics on. They have Arthur Dominic from a pure pyro queen's College, correct Pyal College in PABS right. Yeah, sorry, college, sorry about that. I've welcome to the show. I want to I this is a this is kind of like a we you were introduced us from Jason from last episode and we kind of like the research we all are doing, especially kind of the context of how it fits in with what he's doing. Why don't we start by each of you introducing yourself and then talking about how you get into the space and then we'll go into what balance research is. Well, start with your arther. Awesome. Thank you. Thank you. I'm I did my I started my PhD. So I'M A computer scientist. So I did the master and computer security. So I do like hacking and I started in two thousand and twelve my PhD in a Computer Science Department at e teaches Eric in Switzerland, where we're basically my main focus was on, yeah, checking the security of distributed ledger system such as Bitcoin, and I was really fascinated. But is this security property of pushing the trust from like a central note towards the decentralized network and and having different entities that interact in a trustless manner? I think this was the main motivation at the time. I finished my PhD into the sixteen. Then, a few months of post talk, launched to start up. So one is liquidity network, which is a scaling solution, and chain security, which does security consulting services for my contracts and after this, I mean in parallel, I got a assistant professor position at Imperial College in London will be doing research. So we're going to soon be seven peahd students, mostly focused on distributed ledger security, privacy scalability. All right, how about you, Domic Hey? Thanks for having us. My Name's doomach. I'm a PhD student at the Emperi College and I got into the boxing space around two thousand sixteen with ethereum and the development of Dow and I kind of read about it and I thought it was super exciting. And Call Okay crashing down, and that was even more exciting to me. So I just start with it, and that was the time. I did my masters on distributed systems, and then I started my phd in two thousand and Seventeen and I'm working on things like CRYPTOC coming protocols and cross chain interpobability. Yeah, and I think it's pretty exciting that nowadays we actually...

...have more academics working on this. When I started out into my master's thesis, I was the only strange blockchain guy in the lab, surrounded by people who were working on AI and well, now it's nice to be surrounded by other academics also work on these protocols. There can definitely say, coming from the academic background, that back when you all started, it was weird to do this stuff. Now not so much. It's, I still call it relatively niche and the a coultive exetting, but it's it's like okay, that makes sense, which is awesome to hear. So think the the I guess main topic of today is is this this paper that's put out balance, Dash Research, the see it dot ch at the end of that. That talks about, I guess, dealing with the deposits for different protocols and how you can maybe minimize how much required to deposit well maintaining the same level of security. And it looks like it's basically a reputation system. Is that it is that a like a bird's eye view and how it works, but yeah, I think that's correct. So essentially the idea is that you assigned like a short term reputation. So when you think about other Internet reputation systems like on redded or Amazon, you kind of have this persistent reputation, whereas here you really don't want to long term trust any agent. Then the system. So the reputation is only valid for a set number of rounds or like a period of time measured in either absolute time or blocks. And since we deal with a pseudonymous system, we kind of reset the reputation if you're inactive or you do something that's considered bad. All Right, what sexual? Backing up a little bit, I want to big in those details, but first let's talk about what the issue is and why it's an issue and why there's like a solution like this. Would even need to be we would need to exist. Right. So yeah, dominic or, I'll go ahead if you want to. If you look at like protocols such as stable coins, such as die, a cross chain protocols such as exclaim, there's a certain collateral involved in order to, you know, in order to feather off cryptocurrency volatility and in in general, if you if you have such a protocol where they are certain unknowns, right, for example, as private information, so people don't really know what other parties, what kind of value other parties assigned to a particular asset. But as well, even dependency, so there might be some events and the price might change. So if you have a system where you have these two types of sources of and certainty. Then you typically need to overcollateralize. So overcollaterization means, instead of having a hundred percent collatorization, so in a payment should, for example, you a hundred percent collateralized. Right lightning, your hundred percent collateralize, and then you can set and the amount of up to this on a person coulaitization to your counterparty, but in in coins or like stable coins such as die, you overcollactress. So you collacterize more ether in terms of yes to a lot of value. Then you produce either afterwards, and that's because of this volatility of ether that is behind there. So you don't need to so if either if the enterprise would go down, you don't necessarily need to top up directly your your die, so called CDP, before it gets liquidated. So it's kind of a safety margin, innocence for these uncertainties that you experience in the system and you ask what the problem...

...right. So over corect realization costs money because you if you look collateral, if you look capital, then while you could, you could invest the amount in in alternative investments that might yield higher returns, such as the stock market or other cryptocurrencies, do you see? It's a fundamental issue to like procurrencies in general. Like this volatility issue will basically always exist. I mean I believe it's it's an issue, but alsocial utunity, maybe dominic. You can share what you believe you yeah, I think it will sort of persist because we don't really have a stable currency, so we're forced to kind of use this volatility, usage volatile cryptocurrencies to ensure against risk, and I think going forward you will have that problem and nobody really knows exactly how much collateral you need. That's sort of the other side of the problem. So we have some data on like how much cryptocurrencies are fluctuating, but that's obviously only pasted it and we cannot say in the future how much insurance we actually need because we don't know the fluctuation in the future. And the other problem is that sometimes, you know, like agents the system or actors might have sort of motivations to do damage to a currency or have some hidden like bibing a text, so that you get these external motivators. So it's not just the fluctuation that's a problem, but also these sort of external motivations and for that. Even if you have a stable currency, bribing will still be OK. So for a certain set of protocols you will need to have over collateralization. Yeah, you say part of it. I guess part of your measure here is reducing the amount of guess coll vitalization will maintain the same level of security. When you say security, what is that? What does it mean? What do you what do you mean by security? Great Question. so that is against a economically rational actor. So we assume that you care about your loss that you are facing when you cheat and misbehave. So under the same level of security means that if Alice and Bob, if those are two actors, and Alice is like sort of in a higher collateral level, so let's say at at two hundred percent, and Bob is at a hundred eighty percent, for example, and assuming that you implemented the balance protocol, then the utility for Bob and Alice cheating should exactly be the same thing. So if BOB has a reduced collateral, Bob shouldn't get any additional payoff from misbehaving. And we use a specific trick for that, because we reduce collateral, but only to like a lower bound. That depends on the opportunity cost. So security in that sense, we say, is bound by the economically irrational actor. And the other property that we chief is, for like an unbounded adversary. So an adversary that has inflat funds. That adversary should actually cheat right away in the balance protocol. So while that adversary is still at like the highest level, if that adversary wants to cheat and misbehave in the protocol, that...

...guy should do that right away without reducing his his or her collateral first. Yeah, I would like to add the the irrationality assumption is something very common within blockchain security papers in general. It's questionable whether it's a legitimate assumption because people are not always rational, as you you know. So we could also try to assume like maybe just a Byzantine adversary, but it gets much harder to to prove properties if you if you typically relaxed these assumption. So that's why I think rationality makes sense from mathematical point of view, but it's true that in the real world this might not always apply. Yet most of the blockchain security papers, like also those that that discuss, for example, Layo one security, like the security of of a of a transaction that is not being foked away, for example. Typically all rely on on the rationality assumption. I think that's something that's kind of definitely is your beckwardous across most of the assumptions we make when doing when I'm trying to align and sentives or doing mechanism design. We have to admit not just limited the cryptocurrency. I mean that's it. Yeah, economic problem and free been talked a lot about it. Like he's he's very proponent of doing rational actors, but he was like this is why we have to this is the only option we really have and if we did anything else it would just be absolutely chaos. So I mean it's a fair assumption, I guess. Would you? Could you make the assumption and then try and do make the mechanism work in such a way in which you mitigate as much risk to the non rational actor? Right? It's if you make it overwhelmingly Arad inefficient to behave irrationally than people will tend towards the acting rationally. You can simulate non rational behavior but you can't model it. It's very difficult to model it and so when you build the models around what you're doing? You build it around the ideal system and then see how it withstands this irrational attack science. Yeah, well, let's say I see in balance and balance. What we do in our assumption of the adversary? We actually make an assumption that if we have an irrational adversary that doesn't care about his economic damage, that adversary doesn't gain anything if balanced is in place, because for such an an adversary it doesn't make sense to reduce a collateral in the first place. What's let's actually walk through the protocol on how it works. Like, let's say like this is this is brought up because true, but it was interested in what you guys are doing. How would drew bit implement what you're talking about? Let's to it's like step by step in terms of like what would they do, and then how would it how ould to work? HMM, okay, so step by step. The the paper has sort of an update on that and still have that in the show notes as well so people can read it's also the short I recommend anyone who's listening to us to go read it. It takes ten minutes at most. Yeah, so the website is relatively short. And then if you look at the the main paper, like in section seven, there's like a step by step thing. So what we are or what we need in truebt, we first need to understand what the roles, what agents do we have? So and in true but you essentially you have the person or the agent who is requesting to solve a computation and then you have the solvers in the ideal case where we don't have a dispute. And how it would work is basically we would say that, applying balance, we would need to you first determine a couple...

...of external factors to calculate how much we can reduce the collateral. So first we make an asumption that there's an opportunity cost for locking up to deposit and that opportunity cost is an external parameter that you somehow have to determine. So the potential interest that you could earn somewhere else. If you look at something, I think there's a website call loan compare or something that looks at defy protocols on how much interest you can earn. I think they have for like dialaning. They have around thirteen percent a year. So that could be an indicative for this return right, and then you need to determine a discount factor. So any kind of future value that you're going to earn as an agent, it's going to be discounted, because future money is always sort of worth less than what you have today. And you need to make an assumption about what that discount factor, Delta is, and in the paper in I think we're using standard zero point nine. So you discount by ten percent for the next steps. And then we want to sort of determine actually what actions do we want to reward. So what are positive things? And in the Trubert protocol post the things would be solvers actually providing correct solutions that are unchallenged. We would need to I would need thinking around how it would work in the dispute case, because we want to reward, for sure, disputes that are triggered if the solution is actually incorrect. Let's and let's just keep it simple and assume that there's basically two different behaviors, a good one and a bad one. Yeah, so if it's a good one and a bad one, it's essentially just providing a correct solution is rewarded and providing a false solution would be discouraged and I think in the suport protocol you already balance. What balance does is essentially it would be integrated in the trubut functions. So if you think about it in a low level way, if your trubut function evaluates to true, then that means it's like positive, and if your trubut function evalus to false, that means it's negative, and you would just sort of forward that results to the balance smart contract and the Balance Smart Contract with and say, Oh yeah, this was a good action and this was a bad action. When a good action happens, the agent is rewarded with a score and if the score is high enough then the agent can progress to the next layer. And each layer represents a collateral value. So you have to think about it almost like a sportly kind of system, right. So you start at like the lowest layer with the highest amount of collateral and as you make your way through the layers you can reduce your collateral over time until a certain upper bound. So or well, low abound in that case. So at the highest layer you will have the lowest collateral. So every time a truebert like a Sovera, provides a crack solution, you could, for example, progress to the next layer and then thereby reducing your your collateral. You would also need to specify in which time frame this happens. And since balance is fairly new and also attribute is new in the whole spaces, can in you it? I guess it takes a bit of tweaking to determine how long that period is, because a shorter you set the period, technically, the faster the agents can purpose through the layers. But you in the tribut case, you want to make sure that agents actually have a chance to perform an action in around. So you need to make sure that there's actually computation problems available in...

...the tribute system that solves can actually solve, because without that almost like a balance of you need to do enough good behavior to benefit from the economics of misbehaving at a high tier. That's like it's a tradeoff because you, like would you said earlier, is that they the like the unbounded, the rational actor or or misbehaving actor is better off mis behaving right as he interest the protocols, opposed to gather in the reputation, getting through the layers, reducing this collateral and then acting poorly with it. Why is that the case? So, essentially, say you you need five rounds. Let's say, super simple system, right, even, okay, even more simple. We just have three layers, right. You started the lowest, then you make your way to next one and then in the third layer you reach the lowest level of colatural. So when you do these three rounds, you still have the collateral locked up, right. So in each round, even though you reduce it by a little bit, in each round, you you had colatural locked up for three rounds and now you could decide, basically, am I going to cheat now or not? But the adversary is going to decide. If I know in the first round that I actually I want to cheat, then I'm going to calculate. You know, how much cost does incur for me to lock up my collateral for three rounds, even if my total damage is less? And the trick that we do is basically saying, okay, if I need to lock up this collateral and I could use my collateral to earn some interest somewhere else, then we reduce the collateral only by this or by less than what you could earn somewhere else. So looking at being at like the lowest layer with the highest collacter row. You're going to make that decision at that point and if at that point you're actually behaving honest, it doesn't make sense to, you know, misbehave in the further rounds. And if you do want to cheat, the cost of going through the layers will be higher because of these opportunity costs then to actually cheat right away. Okay, then that you seem like. This work in action is a something that has been implemented somewhere. So I mean, I'm kind of curious as to how something were, if it's re implemented in it grows, what are the emergent properties that kind of come out of it? What could I what, basically, what can I data mind out of this? But what kind of information could I glean from a system working as intended, with the different numbers of layers of people, because you have a smart contract of people with different reputation and assuming some type of publicly attached to that reputation? Is there something that I could pull from that that may be beneficial to me outside of like being outside of the system or to use it to gain the system? Yeah, right, question. So it's a paper for now. It hasn't been implemented in the will and I'm super curious to see how it would actually behave in the wild. Because so we make some assumptions about like these opportunity cause to s gun factors and so on so forth. The the question is like, first, do these assumptions hold the reality of the world and the other cases, will people actually find a way to gain game the system? We have like formal proofs and the paper that this shouldn't be possible technically, but there's always this clash of theory and reality. So I be quite curious to see...

...an actual implementation in the wild and see how how sort of people would direct to it. What's the path to getting that implementation happening? Are You, I notice, you know, we were recommended. You were recommended us by Jason. truebit. Are they implementing your protocol? Who Else is getting involved in this? Are you thinking maybe throwing up your own product to prove this out? What is what is this thought processor? I mean, we haven't discussed anything concrete with Jason yet regarding that, but if there is true interest, I believe we could, we could discuss this potentially. I'm I haven't we haven't basically decided anything yet on that end. What it's trueal Oh, sorry, go ahead, I thought you were dead. Yeah, but it's true. Right, if you can really reduce collateral substantially, like an exclaim, we can reduce it up to ten percent this, this does lower cost significantly and yeah, you can. You can see it in many of these locked up protocols. They they do. So you can get more revenue, but you can maximize your revenue if this would work. But I agree, like with Dominic, we should be quite careful in general with reducing collateral nevertheless, because there might be some other types of vulnerabilities that are outside of the system even or like like things that I'm looking at today. Just talk to dominics today. If you look at like I mean dies being produced by collacterizing eive, then Eith. Then this die is being lent on compound and then who knows what happens with compounds. So what happens if a CDP actually gets liquidated to the to the sea die, for example, right there? They could be like a chain reaction of something will happen. So I engine root is more competent, I think. I mean for my said at least I find it very exciting to bring academic projects into the real world. So that's what I've done. For example, liquidity network, and you realize the many things that you need to take care of that maybe the paper hasn't thought of, which is very normal I think. But still you need this initial, initial research work excess. Feel like I've just one one consequence of for caressing through the tears of this system. You'd have to set the initial the initial layer, the the highest collateralized, right layer to be the almost most inefficient version of overcla of colateralization, and the last layer should be the most efficient, because you don't want to progress through a cheaper and cheaper colateralization layers to get to a point where it's no longer secure in a sense that, like your your your at risk of being liquidated because you have such a low overclorization based on the volatility of the underlying asset. Like if we look at how die is made right now, it's overclateralized because etherium is or ether is is relatively volatile. Let's say it implemented a system like this and I, slowly but surely, I gathered to the layers and my collateralization was like really, really easy for me to make a bunch of Dye, but I was at a much higher risk of being liquidated because of the volatility of F and finding those types of boundaries, like you said earlier, like no one really knows how much you should be doing, is an interesting thing to do because, I mean in the wild can turn out that like maybe one of the middle layers ends up being the safest place to be because because of this Wistar cladization, and we don't it's really hard to get that type of information out of these systems because it's basically the world wild west. And adding systeance like this, if these gives you some finer grain resolution as to where the hot spot might be. So are other people's collateral used to support your your your yours...

...is is that was going on here? So, like, yeah, how can you get down to like I noticed a picture. I don't know how relevant the pictures, but look like the collateral drops to like literally one to one. tunable. It's one of the situations or it's just phighly tunable system. Understand. I understand. What how could you possibly do that if, like, yeah, like with the volatility at all, unless the like you can get people paying two to one, were kind of supporting the people paying one to one. Is that even what's going on there or what? What's the situation? How does that my misunderstanding something? Yeah, so the great questions the there's a couple points here. So one point for sure is where's that sweet spot of collateral? Where can we actually find that? And maybe it's even in the middle layers like you set carry. So it might not be at the lowest slaver and that's also why you need to carefully consider which actions do you actually want to reward for reducing the collateral, because if we take, for example, die right, so say you get close to that liquidation threshold and now you fill up your collateral again to make it above or to be have a saver margin, do we actually want to reward that kind of behavior? Because people would actually, I think, start gaming the system in such a way that they purposely go really close to the liquidation threshold just to get that reward by, you know, increasing their collateral, and I think thereby we make the system overall less safe in a sense that we encourage this behavior. So that's why these two quite careful consideration. How we actually want to apply balance. And then the second question is how much can we actually reduce collateral? So what we do in the paper is basically okay, let's say we have hundred percent collateral at the start. Can we reduce the collateral? And then in the paper we basically say no, we can't, because you should never fall below a hundred percent collateralization, at least in the mathematical model that we built for balance. And I think the knee trick in what you can do with balance is you have the absolute collateral and the relative collateral and in the Nice property with balance is if at the lowest layer where you collateral is like the highest. So if you compared to start collateral, say one point five and two, or like hundred fifty percent to an percent, if you start with a hundred fifty percent, your overall reduction will be less than you can do with two hundred percent. So you can reduce the Colacchal more when you start out two hundred percent, but in absolute terms your two hundred percent starting point will still be higher at the highest layer then if you have started with a hundred fifty percent. So in turn I think that means you could build systems that start with higher collateral and have a higher reward for doing good actions and you have overall more safety margin, since your absolute collateral is still higher. I see what you made relative to the the the relative being the layer difference as opposed to like absolute, being how much you put in and how much you're getting out. Yeah, exactly, so, basically. If so, in one example parameterization, if you sat like with two hundred percent and you have thirty layers, that's example, if you have...

...the paper and figure four, basically you can reduce your overall collateral to hundred seventy five percent if you have started with two hundred percent. So you get twenty five percent points less collateral. If you start with a hundred fifty percent, you end up at, I think, one hundred thirty five, one hundred thirty seven, something like that. So you have less reduction overall and you have a like a lower collateral. But that means if you start with something that's higher, you could actually you know, you have a higher reward for agents that are actually good. So if you're a protocol designer and you want to have agents that we have well and you want to sort of re what them relatively strongly. Then you can start with a higher collateral and in turn you can reduce it even more. Okay, so let me, let me, let me just see if I understand this a little better, because, again, like it's hard to pass through a full paper, we do show every week, Yada, Yada, Yada. This very absolute. So I'm trying to go based off what you're saying in understand the, you know, very basic dynamics of this. So you can come into a system and you can drop in, say, two times a collateral and get one, you know, two to one. Die Eth right, let's just say that. Right. Die Value Ether. Okay, as use a system more the reputation of the say, die system would give feedback to the balance system, which would then do see amount of collateral needed in the future to pull out die is. This is just accurate. It's just a making amusing and picking on die, but it could be anything. Is that accurate? Now the what the is that correct? Yes, that's correct. Yea. So now the month, the the collateral you've already put in, the two to one, is still in holding as collateral while you're doing this. Correct. Yeah, okay, so that will remain in the next time you want to take out die, it costs you less because of that, and and the more that you start doing this good behavior and staying staying above your keeping your head above water, the lower the amount that you can you need to collateralize in order to pull out more die so you can put more ethids of the contract and then pull out. You know, I'm sorry, less into the contract to pull out and near equal amount of die. You might not always get to like one too one practice probably not optimal in most situations that you do, but you get very close. Whereas right now it's a hundred fifty percent with die, you can probably get down to maybe a hundred five percent. Now that's the that's the point that that's you. If you start with two hundred percent, you will probably be able to go down to like one hundred seventy five, but you will not be able to go down you've in with like really aggressive parameter setting. I think you shouldn't be able to go down below like one hundred forty or something. Why? Because you have you can only reduce by these well, depending on a couple of parameters, but the strongest parameter that sets this boundary is the expected return that you can earn somewhere else. So let's let's make that example. We start with two hundred percent collateral and let's say on a daily basis we can earn a return of ten percent in another protocol. Like we can. We can increase our earnings by ten percent if we participate in some other protocol. Then you can only reduce...

...the collateral up to a lower bound of around on hundred and forty percent. So you can go from two hundred percent two hundred forty if under the assumption that on a daily basis, you can earn a return of ten percent somewhere else. If the earning is much lower somewhere else, you can only reduce by a much lower margin. So say you earn like five percent in another in a potential other protocol, then you can only reduce to one hundred seventy five percent, but you can. So this is why it's that's necessary for the security of the protocol is that you can only reduce up to the bound of this opportunity cost if you start reducing the collateral lower than this opportunity costs. o. So in your example, if you go from two hundred percent eventually to like a hundred five percent, then that breaks the security of the protocol. That's interesting because this, this opportunity cost, is dynamic in reality, which means that the parameterization of whatever your system is would also need to be dynamic. Is that true? Yes, so there's a couple of extensions that they are not yet or that are currently not in the paper that that we have in the back of our minds. So one of them is like, yeah, so this is dynamic. Do we need to actually adjust a system on a dynamic basis? But that would introduce a lot of problems, like we need to have some sort of oracle for for feeding in that kind of data. So that's a bit tricky. Maybe it's also, you know, fine if we assume that there's an average and if we don't break out that average too much, it should still be okay. But yeah, these are this is one of the sort of parameter questions that that makes it, I think, exciting to try this out in real world and try to determine the parameters. Yeah, so we have some done some work on this for the x time protocol to sort of fine tune these parameters, but it's so it's a bit tricky. To to sort of pin down the actual value. But again, the I think the Nice property or the advantage of really the adventure of using balance is that if you start with a relatively high buffer, so with a very high starting culture, then you can reduce more. So it kind of makes sense to kind of push a bit up the the the initial collashule that you have to provide, given that you can reduce over time. Let's that's a tough proposition to sell to somebody, though, especially if you try you're trying to vet whether or not a particular asset has value. So it literally in your your you're putting up a barrier to entry on something where we're trying to reduce barrier to entries. How would you suggest somebody boots draft a system that uses your system? So the question is if that's actually the case, because if you look at die as an example, I think it. That is a great example because we can actually we have data right and the liquidation threshold is hundred fifty percent and die, but in reality you see the actual collateral ratio somewhere between three hundred two, almost six hundred percent. So I think the people that open the CDPIECE, they try to protect themselves against getting even remotely close to the to the threshold. So in die you could, I think, easily set a collateral ratio of two hundred percent...

...and then if you have good agents and they sort of go through the layers and say their liquidation threshold could be reduced to one hundred seventy five percent, that gives them sort of more safety margin is, I see, provides also a bit more safety for the die system in if there's some sort of big market crash. And in reality it seems like that the majority, at least from what we see in die, is willing to pay a lot more collateral to stay as far away from that liquidation threshold as possible. But then again, I see your point totally. If you want to have efficient collateral, saying that, you know, we should actually increase the starting collateral is a bit of a tricky one, but then again, I think it's the it's sort of the tradeoff, you know, and I guess it only also depends a bit what kind of risk you have behind it. So in Trubet, for example, it's a bit true. It is a great example for this case because and true, but really, the damage that can be done is to the person who provided a computation to solve, whereas in die it's kind of to yourself, right, if you, if your CDP gets liquidated, it's only you suffering from the damage at first. But if, like a lot of agents do that, you might get some sort of systematic failures. So really we have to be quite careful and how to implement balance and in which way to implement balance and see where these tradeoffs come in because, for example, and true, but I don't see a lot of risk for other for other computation providers. So what? Okay, so, so now my brains go into the product mode and like trying to understand like how this would actually work in reality, and I've come up with like one example of where I could see the collateral being so high and yet they still would want to do this, and that's it's actually an n FT example and I don't know if your protocol will support that. So let's just say that you keep software licenses on the blockchain and this software license you you want to rent this software from someone else. Okay, so you have to give them the NFT. They have possession of the NFT, but you want them. They want to be able to just like pay a small amount to actually use it for x number of ex period right then they have to give it back or occur a fee. I wouldn't I would probably see a really good way of doing that. It's just like collateralized the NFT. Let them hold it and then, if they're a good actor, they return the NFT. They get reputation to system and they have they don't have to necessarily collateralize as much going forward. But the problem the system to stay in the layers. Is that correct your pet yourself in that when you broke up on my mind. Oh, they I don't know where I broke up, just that last part. They'd have to keep the collateral locked up in order to go up the layers. So if they go down to layer one, right. So is this even a valid use case? What would you think? Again, it's also take going out of the normal like coin to coin exchange rate system like. These are the kind of things where I think I can kind of like make real world analogies for how people actually use collateral R so just to make sure I end the same card. So basically, you have a sufur license, that's an NFT and you are renting this out to someone and that someone provides a collateral to make sure that it's returned once you're finished with it. Yes, yes, okay, and then if I return it correctly, and then I can...

...reduce my collateral over time with other software providers. Yeah, but do what? Does that actually work that way? Because the reason I ask is, don't you have to keep the collateral locked up in order to reduce the value of the collateral that you're that using? Or can I release my own collateral and then still have the credit in the system to actually get less? That's like it's making it's making the reputation longer than temporary, and its system that Collins talking about. He wants to maintain the reputation over time through multiple collateralizations and because, okay, and returns of that type of thing, so that he's he ends up being a good actor on a system. So it's more just it's more like a reputation a long term reputation system and a short term one. Yeah, so essentially what you can do is right now, how balance is designed. It's it's synchronous, so we have a set durrection of the rounds and if you don't do anything in the next round you get actually up to the or what you get demoted to the previous layer. So say there's a pause and you don't do anything and you don't interact, then your collateral in the next rounds will be higher again. I think on the there's a couple of things here. So one is like an implementation thing. What you would kind of need to is in your pre a call to allow actors actually to withdraw parts of the collateral so that if you have a lot of collateral provided and you can reduce them, that you can actually get that collateral out, because then only then you can actually sort of reduce the the cost factor. So that's a method that needs to be implemented in the contract. And the other thing is, obviously, you know, do we have some sort of long term collateral? And this is it's a great question. It's an open question that I am having as well. Is, you know what, if you only have these very sporadic interactions with protocols, then balance in its current form may not be kind of the solution to the problem, because you have the synchronous, continuous rounds and if you don't do anything, like you did something Nice but then in the next round or, you know, you participate every half a year or so in the protocol, then it's only or it doesn't really add anything. What do you lose by bastardizing your protocol and turning it into a long term reputation system? Yeah, so you can't just, you know, doing it like long term, because we don't want to end up in situations where you turn militias at just some point in time. What might work, but we haven't tested it, is either making the protocol asynchronous so that we really we just consider your actions and in a sense, only when you interact and we get information about you as an agent. Only then we adjust your layers. Yeah, that's that's a bit of a question. So right now it's again it's synchronous. So we need to analyze exactly how that would work in an ASYNCHRONO system. The other question that is coming up in that matter is can we actually do reputation transitivity? So what if we would implement balance, not like in a single protocol, but what if we would implement balance on like a etherium level and you can have reputation across different protocols that decide to feed into the balance contracts. So individual protocols...

...would sort of define what are good actions and would get feedback to the balance contracts. I think that's a really interesting question. It's one that needs a lot of careful consideration, because you could just start deploying random protocols that would report that your action, that your agents, is doing all sorts of great things and then that hasn't picked on other Proto calls. So I think it's quite a delicate problem, but it's one that could work around this problem with or where you could have protocols where you interact quite often and some protocols where you just have very few interactions, but they could both profit from from the collateral reduction. Yeah, but this, again, is that's sort of going further than what the current balance paper is. Yeah, you, so this is it. This is really like, this is really good for incentivizing things like true bit, where you are supposed to constantly be interacting with with the protocol itself in order to like, for instance, a validate or create computation for somebody else can use something on their on their behalf, you know, run a smart contract off chain on their behalf. Like this is something you should be doing frequently and get rewarded for that kind of thing. And if you wanted to get Seepu Tokens, you know you'd have to find some way of breaking into that system, and one way to do that would be collateralization. And then that gives you a reputation system. will not sort of a reputation system in in through balance, so you can actually keep doing that more and more and more and actually earn more true tokens as a result. Is that? Is that like an accurate, like reasoning behind the balance protocol? That's what that's the ideal case for what you currently built. Yeah, exactly, play it. So we set it out with applying it for for the excellent protocol, and there you have like these these vaults that need to lock up collateral and then we need want to ensure that that day release bitcoin back on the bitcoin chain to from the bitcoin back tokens. And it's really about this being an interactive protocol, and we're currently looking into die, because die is already the first problem where when you open up your CDP like it kind of it's just there right. You don't necessarily have these continuous interactions. So it's really a question how to apply it to say in quotation marks, like passive protocols. Yeah, I would like to also chip in. It's a good point that you mentioned. If you have an initial very high culitorization ratio, I think it depends a lot on how fast can you move down to lower layers. Right then the the initial higher cost might not be that bad because you might just get get the get the different specularia. Yeah, it definitely the like. When you hear collateralization in the typical sense it's to me it's like a one off. Like in my brain it's like a oneoff. You know, I'm doing this once. Okay, I want to get some die, so I go out and I claterize and I get my die. Okay, now I have died right. But this is a repeated this bent this is a system that that that is built and designed around the idea of a repeated in constant use and in it's for high volumes to systems or high high utility systems. That might not traditionally fit into people's mental model around collateral, which is evolved Rut which when most people hear colladal I think that it's fair to say that they kind of associate it with like, Oh, I'm putting up my house, I could get alone, you know what I mean, or that kind of thing, so I could get a business land. You know, my house is my collateral for my business loan. You know that could that's like a one off, but in this case is like a repeated like system for high volume, like are not even necessarily high Voune, but frequent uses, constantiality. Yeah, yeah, I guess where the sort of the big difference is where if you would put up collateral right with your...

...bank to get a financing for your house, we kind of in the decentralized finance system, like the bank would need to put up the collateral. So really balance is more for the bank kind of set of things, for like a service provider, because we don't want to trust the bank anymore, but we want to trust the bank because the bank has actually provided collateral, and balance is the one that can help the banks. Or then didn't new type of banks, or the new service providers in that case, with for them to be more efficient and to reduce their opportunity cost? But you're right, like how balance is designed at the moment, it's for these continuous interactions and it's aimed at like service providers that actually have continuous interactions. All right, I think it's a it's a decent way to start wrapping up. Are there any any questions that you wish I would have asked you that? I didn't get around asking either. One of you got them all perfect. Yeah, we got the coverage. Awesome. Yeah, I know it's not something like yeah, really great cresents. I guess one one other point that that we can add on the on the balance protocol is going back to that idea of applying it to like a blockchain whyte system. I think that could be a really cool idea because in that sense you have a little deserve service providers. Also, when you think about things like Kessper FFG, where you have these on chain staking things and you have validators that really interact on a continuous basis, I think these are sort of the interesting use cases. But yeah, so we will have to look a bit more in detail how that will actually work up. It does kind of sound like a staking mechanism a little bit, you know, but you have to be able to get Roi out of it in order for there to be value on the steak, and I don't know if there's currently a mechanism to do that. New Protocol Right. Well, a goal is to reduce the like. Say, there's a given. I'll maybe take a just a an idea out there. If we talk about foo and the staking mechanism for being a validator, there's it's there's a certain Roi on being a validator. Maybe there's some type of thing built into this and such a way we're like, over, the goal is to find the minimal amount of money that you can stake to get this type of thing. But also disincentivized people from entering the system and misbehaving immediately. So, like, in order to get to that optimal amount, you just start with a larger collateralization, or staking for that matter, for that Roi and then slowly move down to one that's like the opt the optimal number, so that you know the people who are at that optimal level are relatively good behavior and good behavior people in that system. So, like, it just it just takes a certain amount of time to get to the point where you're up are you're operating at an optimal level. It's opposed to entering operating and then you know Ames and what could be also very cool. Well, the problem is if then they have to continually input more steak in order to reduce well, you know, I mean, they only get benefit from that. So really, though, I mean, if you think about it, it's kind of more like then you have the issue of like those are interest out way the amount of, you know, bad behavior that they're putting in. Yes, so if they just side to be Byzantine for whatever reason, if they've acquired enough interest to outweigh the collateral, then you know, or it could way way to force an exit, like if you're if you're interested in your collateral are over this threshold, then you must no longer be a valid or you must...

...restake. I don't know, it's just kind of a thought. It's just it seems like another it's the tool in the tool box that people can use to try and aligned sentence appropriately while also being efficient, for like optimizing capital lock up right. You want to make sure that people make the most money and a fair way out of the money that they're putting into a system without, yeah, exactly reducing security, of course, which is kind of the whole goal of this whole paper. Is To do this in such a way where you don't reduce these security of people doing this type of thing. Yeah, exactly, so sorry. So, yeah, I just wanted to think it's important that we make sure you kind of it. Not Everybody's on the edge Ron. So it's still very important that this transpendency and everybody can can kind of optimize the particular steak and revenue, but without its kind of putting at risk the overall system, where by rational or maybe a partially irrigtional juststory could could come in and cause major damage. Yeah, put on the work I figured to be a rational yeah. Yeah, I mean essentially, when you think about like a Validator in a prof of state protocol kind of thing, and let's say the VALIDATA puts in two hundred percent collateral at the beginning, so to two hundred eight, and then the valid teritor over time, can reduce to a hundred seventy five eight, then the validator doesn't necessarily have to put up new collateral. He can basically take out existing collateral, like the twenty five eight, over time and have sort of less opportunity go. So either, you know, using the other twenty five eight at a rate of one hundred seventy five percent to continue staking, or use it in some other protocol as well. Maybe get some die, maybe go to compound and an interest. They're over time by being a good behavior by being a good behavior agent. I don't know writ term for that. That's terrible water say it. You get to reduce your operty or opportunity costs of the money you lock up, and that's that's that's a good thing. That's something you would want to do as someone who's locking money up over over a long period of time. Yes, exactly, that's the that's the whole point. And then, yeah, the the the cool thing would be you, you know, if the valader has two hundred eight locked up or hundred seventy five ee, you get the same motivation to be have honestly from those two two agents, and without slicing conditions too. So it's not like, you know, if they were, if there was like a if this were to apply to any sneak staking scenario, will be a good way to stake without slashing and that you just losing opportunity value and nobody wants to lose money. Must Well, now you get like say so, you you misbehaved. You get pushed back to another level. I think that's what it says in the at least the website. Yeah, and in that case you have to put money in. That's kind of like, I don't know, it's you're right, you don't lose anything. You're slashing is more of an opportunity cost type thing than, I am like an actual losing money. Cool. Where do people go to learn more? Sorry, I didn't get your question that. Where do people go to learn more? So on the balance Resert Duc h website is like a quick overview of the protocol and there is a link to the getup and there's a link to the paper, so there's like a proved concept implementation. On the GITHUB in solidity there's, if anybody's really into Mathematica, there's a whole mathematical model of the thing and you can play around with your own values. There's Hey, Yay, there's a xt Ame application in a mathematical model.

So if you kind of want to have your own protocol in there, you can use that and otherwise, I guess reaching out to author and me. We're happy to help. Awesome. We'll add that stuff in the show notes so people can play around with it and get in touch you guys. That was has been fun. Thanks for Promo show. Yeah, thanks so much, cold in Gary for the great questions, cool, yeah, thank you very much.

In-Stream Audio Search

NEW

Search across all episodes within this podcast

Episodes (128)