# The Circles Web of Trust: When It Works, When It Fails, and What to Do About It

- Channel: [Dappcon](https://streameth.org/dappcon)
- Date: 2025-06-19
- Duration: 27:10
- Watch: https://streameth.org/watch/6853f64790bd41297b790ebc

## Transcript

Hi everybody, very happy to be talking to you guys today. My name is Paul, I'm with the Circles team, and I want to talk to you about the Circles Web of Trust. This is not an introductory level talk, Martin yesterday gave a presentation on this, there's been several workshops, in fact, as I guess you all know, it's been hard to kind of evade us in this conference and over the last couple of weeks. Instead this talk builds on the understanding, the basic understanding that we hopefully all have about this protocol. So let's just recall what Circles is about, here are the core three rules of Circles, everybody can come and create an account, there's no gatekeeping, and then as soon as you have an account, you're going to be enabled to create your own personal CSC token at a fixed rate of one circle an hour, and all the circle tokens that are being created by everybody suffer from a sort of constant evaporation, which is called the Demirage, at a rate of 7% per year. And on top of that, we have the notion of trusting, users trusting each other, that's just a relation that is sort of registered on chain, and by Alice trusting Bob, Alice is kind of communicating to the rest of the network that she is willing to accept Bob's circles at face value. And so if we think about it, we have one rule relating to money creation, one rule relating to money destruction, and then one rule relating to money flow, by which I mean, if you sort of took a snapshot of all the balances, the third rule would kind of govern the dynamics of how these circles can then move about. And in this talk, I'm not going to be caring at all about the first two rules, I'm only going to be interested in the third rule. And all these trust connections, they produce a trust graph, and people get really excited about the trust graph. We talk to them, we ask people, you know, what do you like about circles, and we often hear that the trust graph is what they find most exciting. And I think there's several reasons for that. One of them is that people just love graphs, full stop. One of them is that people really dig the idea of kind of curating and engineering their personal currency, and so projects like Trustlines or Ripple Pay have always excited lots of people in the community. There's also this social and local nature of trusting, so circles are fully sort of breathing decentralization and sort of giving sovereignty to the user, so not just at the tech stack level, but also clearly at how the protocol works. And people get really excited to think about secondary use cases, things they can do with this trust graph that go beyond just the circle currency. But the trust graph also brings up questions, it brings up some questions or kind of confusion. There's questions like, why do we need the trust graph to begin with? And whom should I trust? Just in practice, like, somebody comes up with a QR code, should I scan it, should I not scan it, what should be the criteria based on which I decide? Is the graph providing a proof of personhood? If so, how strong is that guarantee? And if not, what is it providing a proof of? And I think we, as a team or the communities, we're also not yet quite decided, so there's conflicting advice floating around, such as, trust all accounts you think aren't bots, or trust those who you think will provide the most value to the community, trust the people you trust in real life, or don't trust accounts by people that you believe might run a second account. So they all kind of pull in the same direction, but they target different underlying criteria. So this talk tries to sort of help answer these questions, it's just sort of my two cents on these questions. They don't represent necessarily the views of the Circle team or kind of what we're going to be rolling out or whatever, it's just me thinking about them. But why I think it matters to have this discussion and think about it is because A, lack of alignment on these questions among the Circle's user base undermines trust in the protocol and hinder adoption. So if it seems like an arbitrary thing, everybody's just trusting based on different criteria, how can this become sort of the next big on-chain money? And B, if we're really committed to thinking about secondary use cases, things that we can do with this trust graph, I think we need to be sort of confident in the properties of the graph and the value proposition, right, so that if everybody's doing things based on different criteria, what is it that the trust graph is encoding? So the main points I kind of want to drive home over the next 15 minutes or so is that A, what we prove depends on our collective trusting strategy. And that's kind of a truism, I mean of course that is true, but what I mean is that if what we're after is a proof of personhood, then depending on the collective accuracy of our beliefs and the personhood of others, we might achieve that. If our criterion is something other, another property that is sort of hidden that we want to provide proof of, then again depending on the accuracy of our collective beliefs, we might do that. If all of us are doing different things, then probably we're not really proving anything. The second point I want to drive home is that there are some good reasons, in my view, to kind of move beyond thinking of what we're doing with the web of trust approach in circles as providing this proof of personhood. And the third point is I want to quickly talk about how groups, which are a new construct in Circles v2, allow us to go beyond the web of trust approach. So let me start by addressing my view of the question why we have the web of trust and the trust graph in circles in the first place. So I think that within circles, the web of trust is primarily a technical solution to a technical problem. Circles is an ambitious project, and we kind of want to have the cake and eat it. We want to have a currency that is universally accessible in the sense that everybody can create an account. It should be democratized. The issuance should be democratized in the sense that all of these accounts can start participating in the money creation, but we also want what I would call sound tokenomics. And I know this is a loaded term, sound tokenomics, I hear means something very modest by it. I mean the rate at which a person should be able to produce circles should be one an hour. Not more, but not less. And so if you're willing to just take any two of these, it's relatively easy to see how you can do it, but getting all three of them under one hood is kind of difficult because of this obvious civil attack that you kind of open up for. And the web of trust is meant to address this issue. So when Martin, for example, gives this kind of advice, what he's getting at is that he has obviously an account that has, on the media, is anonymous. We don't know who they are. They have no form of verification. And they evidently have at least two accounts on Metri that they're sort of mixing up, right? I mean, so there's several red flags right there for why we think they are not using the protocol as intended. Right, so now how does the web of trust approach help us address this problem in detail? It helps me to construct a kind of a toy model, a mental model that I use to think about how we as a collective use our beliefs about the personhood or the properties of other people to kind of produce a currency that is universally accessible, issued democratizedly, if that is the thing, and also has sound tokenomics. So let's take the example of an online community that wants to set up a currency with democratized issuance and universal access. And the community knows that it's somehow infiltrated by accounts that have some hidden property which, for lack of creativity, we're just going to call bad. They're just so bad accounts. But bad could also be a person. It could also be people with, I don't know, things that you can't see, like certain chromosome structures or something. And we're going to define the tokenomics of the currency that they're going to come up with as sound if these accounts are somehow successfully occluded from the economy of this online community. So in thinking about what they should do, they come up with the toy rules of circles, which look as follows. First of all, give everybody some circles. And they're all non-fungible tokens, right? Everybody gets circles. And the second step, each of the participants in the community gets to decide whom to trust based on some strategy. So they have articulated a strategy and they decide, well, this is how I'm going to trust. And then third, they're going to sample a random good exchange partner. So this is this red arrow here. And so this is how this sort of day looks, or this round looks, from the point of view of this one player. But now they all participate in this game. And the task is for the good accounts to come up with a collective strategy such that the economy among them is liquid, but bad accounts are isolated. So we can now kind of formally evaluate the quality of a given strategy by liquidity, which I'm going to take here as the average fraction of their circles that good users can spend, right, spend with the transaction partner that was chosen, and adversarial exposure, which is how many circles can bad accounts pay good accounts. So we now have a mapping from every possible strategy to the quality of their strategy. And there's a simple optimal strategy, which is, well, all accounts just trust each other and none of them trust any of the bad accounts. That's the perfect strategy. Liquidity is 100%, adversarial exposure is 0%, so far so good. The only catch is that to come up with that strategy, you need to have known who's who in the first place. So that kind of begs the question, right? But people might not have full knowledge of this hidden property, but they have informed beliefs. They might have interacted with them in the past, so they might have applied some Bayes rule or whatever, and they might run around with some indication of whether a given account has that property or doesn't. And one possible strategy that they could kind of agree on is that each of them samples their trust as a function of their current beliefs, and they update them accordingly. Right? So here we have these accounts kind of having some roughly correct beliefs. You can then show in this toy model that if there are islands of knowledge for every account, this strategy will work well. So what do I mean by islands of knowledge? I'm not going to tell you, so I'm hiding some technical lingo here. But so quite clearly, the web of trust approach in this sense, in this toy model, builds on the wisdom of the crowd. And it can only work if the crowd is wise in that sense, right? If they have no clue who's who, then that approach is not going to work. Now a couple of interesting properties of this approach are that A, beliefs don't need to be optimal for a strategy to work well. Here we have sort of a circular trust network, and so that is not the same. You know, so this person here, do I have a pointer? No. The person on the bottom left, she's only trusting the one at the top, but not on the right. So because she might not have interacted with them, but nevertheless, they get good liquidity and good adversarial exposure. Now, this approach is also naturally resilient, because you can have subsets of the economy, like the two accounts here at the bottom, that if they have accurate beliefs, and they can kind of shield themselves from other parts of the honest side of the network that have fallen prey to bad accounts, they can get sound tokenomics going among them, and they are not sort of exposed to the bad decisions of others. So to summarize, the web of trust approach works as follows. If a collective agrees on some property X of people that people should have that they want to include, and they hold collectively accurate beliefs about the actions of others, and they also choose their trust relationships according to their beliefs, then we can achieve, so then we can have the cake and eat it, right? That's the kind of idea. Now what's the implication for circles? If we as a collective circle's user base want to have sound tokenomics mean for us one circle per person per hour, independent of what that person is, what they do, how they live their lives, whatever, then what we should be doing, we should trust all those accounts that we are confident represent the primary account of somebody, and that's the only criterion. We shouldn't consider any other information. But we might of course decide to go beyond that. And so I want to kind of quickly introduce the idea that what we should be after is maybe not proof of personhood, but proof of economically trustworthy personhood, right? So if, I'm going to sort of tell you what I mean by that, but if that is what we're after, then what we should be basing our trust decisions on is we should trust those accounts that we are confident represent the primary account of somebody. We're still interested in personhood, but we're not just trusting any person, we're trusting people whose integrity we believe we can rely on when it comes to sort of trading or dealing with them economically. And I think this is sort of an interesting way to look at it, at least for me, because I think it builds sort of on the strengths of the approach, right? I mean personhood to my knowledge is sort of pretty much a binary property that people either have or they don't, but economic trustworthiness is super fuzzy, it depends on what you mean by it, so it requires human judgment. So that's exactly what a web of trust approach gives you, and that's not something that an iris scan can even attempt to substitute. It's intrinsically relative, right? So the fact that different people have different understanding of what trustworthiness means isn't really a problem, right? It just means that you're going to be dealing with people who ideally share sort of a similar idea of this fuzzy concept. I think there's really lots of secondary use cases, so the idea of us being able to produce a trust graph that isn't just proving personhood but something stronger is very interesting, and whoever of you was here when Sebastian gave his talk and we talked about how can we sort of choose which nodes we should expose our public IP address to, and we could sort of incorporate this graph, then I think this is already one interesting use case. And I actually think that a lot of us are already doing that. A lot of us are basing our trust decisions already not just on the question of is this a person or is this their primary account, but to what extent can I rely on them? So as a simple thought experiment, if you are not sure which of the two camps you would fall into, what is your sort of circle strategy, you consider the following example. Let's say there's an account that runs a whole bunch of Sybil accounts, and that gets discovered. So now what do you do? Do you entrust all the Sybils and not the central account, or do you entrust all of them at once? Depending on where you fall, you've already made a decision. And I'm not saying one is necessarily better than the other. I just think that it's both viable and valid properties we might be after. All right, so that's a lot of sort of theoretical blah, blah. Let's try and evaluate this question looking at the actual circle's user base. So in our circles in the team, we have a blacklist of around 12,000 accounts. And these have been, I mean, they hold around 26 million circles. Right now, and that was last week. And we put together this list, essentially based on trustworthy personhood. So for example, if we discovered a network of puppets, we didn't just include the puppets, we also included the Puppet Master, the account that actually drove this thing. And just to clarify, this blacklist, I mean, is not limiting anybody on the protocol level. We as Gnosis, we cannot censor what you guys are doing in circles. But of course, we own groups, we run groups, and as owners of these groups, we can make decisions about letting people on this blacklist in or not, right? So this is important to clarify that we have no sort of backdoor entrance and sort of sanctioning anybody here. Right, there's two groups that are kind of prominent in the current scheme. There's the circle's backers. So the group of those people that have put their money where their mouth is, and they've invested $100, and they hold around 1 million circles right now. And there's the Metricorps group, which is a loose group of 800 people. I think today it's actually 1,200 people. And they also hold around 1 million circles. And so we didn't, weren't very careful in who we let into the Metricorps group. But the backers, of course, they have an incentive to be very conservative with their trust. And here's sort of the numbers for the adversarial exposure, by which I mean how many circles could accounts on the blacklist today channel into either one of these two groups. And we can evaluate this in terms of absolute numbers, but we can also evaluate this relative to the total holdings of the blacklist accounts and those of the two groups, respectively. And the answer is that for the backers, we're actually looking pretty good. The backers have successfully sort of shielded themselves from the blacklist. There's only sort of 2K out of 26 mil that can go in there. For the core group, the numbers look slightly more concerning because the absolute number is 91,000 circles, and that constitutes around 7% of the current holdings of the core group. So either we are pretty bad at putting our blacklist together, or the Metricorps group isn't aligned on what they're basing their trust on, or they're not doing a very good job in trusting accordingly. Right, before I finish, I want to talk to you about one way that, if any of you was now sort of starting to be concerned about the web of trust, or like, does this mean that it undermines the Circles project overall? I think the answer is a resounding no. And one of the reasons for this is the existence of groups that we introduced. So, for example, projects like, protocols like Proof of Humanity or self.ck, they can create groups and circles that dynamically add all accounts to the group that have gone through their protocol. And now vendors or other accounts can start to decide to accept only their group circles. So by doing so, you can decide that you only ever get paid in coins that have been minted by people who've gone through some protocol, right? And so there's also a hybrid strategy, where you can decide to rely to the web of trust to some extent, but you also want to kind of incorporate information from external protocols. So that goes to say that Circles, in my view, is the language, not the book, right? I mean, of course, we built the protocol based on opinions and convictions, but it's an open protocol and everybody can build stuff on top of it. We can't force people to choose certain criteria for making their trust decisions, but I think it's a good idea to start coordinating this to kind of drive the value of this project. And yeah, with that, I'm done. And if you still haven't created a Metri account, please, this is your chance. And I'm open for questions. APPLAUSE Thank you so much, Paul. Before we open for questions, I just want to say that the next Farsat chat will be about 10 minutes late. Not everyone, not every speaker is here yet. So we'll start about 3.20 more or less. Does that mean I have to take questions for 15 minutes? Yeah, so you're going to have to take questions. Let's go for it. I see many hands. Yeah. Hi, Ivan from Reactive Network here. Thank you very much for a clear and interesting presentation. So the question is, from my understanding, trust cannot be expressed as just one number. So let's say me and Bob have 10 transactions, each worth of $1,000, right? I can be confident that if I deliver the goods without an upfront payment, he will pay me later $1,000, right? But if I deliver goods for $100,000, in that case, Bob might reconsider and not pay. So I think you probably thought about that, and it would be great to hear your thoughts. Yeah, I mean, so I think in Ripple Pay and in Trustlines, this sort of additional degree of kind of quantifying your confidence of the trust or something like this was present. And it's absent in the B2 of the protocol. So right now, trusting somebody is a binary decision. I think we definitely have voices in the team that are in favor of doing that. And I think the sort of Bayesian view of looking at everybody running around having these degrees of belief between zero and one about everybody else sort of pulls in that direction. But at the same time, I think the protocol is complicated enough as it is, both in terms of UX, but also in terms of, yeah, just like technical implementation. And I would assume I wasn't the one making the decision about this. I would assume that we've made the decision in favor of UX and that. And so we'll maybe see something like that on top of Circles in the future. More questions then. One there. I enjoyed how you talked about the concept of trustworthy personhood as a way to define humanness. Or a form of humanness. So there was an opinionated view on what constitutes humanness. In the instance where, like, let's say in the future, I form a relationship with agentic AI that feels aligned with me, and that agent then creates a Circles account, I want to trust that agent. Does that pose a threat to the protocol, to the Circles protocol? No. I mean, again, you know, we don't write the rules, we don't impose the trusting. If you get a collective, like a critical mass of people who share your feelings about how they should trust each other and what sort of sound tokenomics means for them, then you just do your thing, right? And so we can't control that. We do have something we call the Circles etiquette. And we've made a decision to say that Circles is primarily a currency created by humans. And so for now we say, you know, AI agents can be part of the economy, but they should have organization accounts. And they can be paid and they can transact, but they shouldn't create. And so this is, I think, our current stance on this. But this is sort of what the future will see and whether, you know, anybody will move away from this. I mean, it makes total sense to think about this, yeah. Okay, maybe one more question. Oh, one more question. Oh, okay. Oh, I see one there. No, yeah. So my question is, with the current graph, so on the Circles white paper, there is this discussion of this possibility that maybe there's going to be a fractioning into different subgroups, where there's a subgroup of people who trust each other and their currency is worth a dollar. And there is another one group that the currency is worth 20 cents and there's limited liquidity between the two. This is a possibility mentioned on the white paper. I want to ask you, with the real data, is it actually happening or is it not happening? I think the short answer is we see that it's kind of happening. We definitely have groups on there that trade at different prices right now. And we're making sense of the details of this. But fundamentally, I mean, what Kelvin is referring to is something we refer to as liquidity clusters in the paper. So basically, any group of people where there's a trust graph between any two people, we kind of have a guarantee that at equilibrium, the prices of their tokens should be the same. But when you don't have those trust graphs between all connections, then you can have this sort of bifurcation or sort of triplification and whatever. And yeah, I think we have this right now. Thank you very much. This is a comment that I think is actually very interesting and not a problem. I think it's actually very interesting that for this thing to happen. It's like the economy in the real world. I agree. I mean, so in the white paper, we're like, yes, the main goal is for this to become a globally fungible currency. And it's only one price. But it was sort of almost pedagogical, the decision to present it as that, because that's the simplest way to think about it. But of course, reality is always more complicated. And I think there's an interesting future of circles in which we do have a handful of these groups. I think we shouldn't have as many as users. That would be sort of not the intention, but there can definitely be a handful. Thank you. Cool. Thank you.