Toni Wahrstätter 🛸 Exploring Censorship Across The PBS Stack

[Music] thanks everyone for showing up um let me quickly introduce myself my name is Tony I work as a researcher at the Firm Foundation and today I want to dive a little deeper into the PBS deck and especially focusing on censorship within the PBS deck just to give you a very quick outline what I'm going to talk about I will start with very shortly explaining what type of censorship I'm meaning when I talk about censorship then we will look into the PBS stack so relays Builders validators and finally I want to quickly cover uh in my opinion quite interesting example which will involve a comparison of Bitcoin and ethereum so if you ask jbt you get this very very general um definition of what censorship is so it says like censorship is the suppression and restriction of speech communication other information and so on but of course this is much too broad for our context and indeed I want to mainly focus on the beacon chain so every type of censorship we can see on the beacon chain will which will also be part of this talk now so let's get into it the first type of censorship is not proposing blocks with transactions to or from targeted entities and with meaning transactions I'm of course also speaking of internal transactions so it's not only the um the normal transactions that I'm talking about but more or less everyall to any address that can be censored in this case next a much harsher form of censorship would be not attesting to blocks with transactions that con um that um interact with a certain ENT um entity and finally a very harsh form of sensorship would be even rearing out certain blocks or yeah excluding them ignoring them um if they contain um a transaction of a certain entity and if we if we think about censoring right we also have to answer the question what is censoring first of course if you don't like your neighbor you can censor him which will be which would all already be censorship but on in the eum context we mainly when we talk about censorship we are mainly talking about ofac right but surprisingly ofak is actually not the only list existing right there are other countries that also have their enemies and here on this slide you can see some countries with their own s um sanction lists and although we only talk about ofac all the time there are actually all the countries have their own lists and the OAC list is therefore so dominant because it was the first that really first included some Bitcoin addresses and then later on also um targeted some etherial addresses like tornado cash so for the for my talk I will also mainly focus on the OAC list just because the OAC list is the only one that we um that has a real impact on the today's ethereum landscape to quickly tell you something about the methodology I use so how do I even know that a certain entity is censoring or not right this is not that straightforward but there are actually um a bunch of cool ways how you can determine is a certain entity sensoring or not and one cool way is just using a binominal distribution so we assume that the probability of a block containing a certain transaction so containing oact sanction content is binomial distributed and by doing so we can just um calculate what is the probability the likelihood of a certain entity proposing a certain number of blocks without any OFA sanction content um P here is the probability of a block containing a certain containing ofac sanctioned transactions this means 2% so 2.18% of all blocks since the merge contained OFA sanction content and now it's basically very simple um the larger the sample size the more sure we can be that a certain entity is sensoring so for example if you propose 10 blocks then it's not that easy to tell if you're censoring or not right this would this would not be very accurate but if you propose um a thousand blocks and there was no of sanctioned content in your block then we can be quite sure that you might be censoring the chain so this just as a starter and now we can dive into the actual censorship of the current Network so starting with the validators if we look at the validators it looks actually quite good you might say so we are the graph shows you the distribution uh so the share of slots and we can see around 15% of all slots as of today are produced by sensoring validators and now you might ask who are these validators basically you can go to tornado. pic um and there you can see which entities are including torado cach transactions and which are not and in the first four columns you can see the number of blocks of those entities um the number of blocks with sanctioned content and then um let's focus on the last column it tells you the percentage of ofac blocks of a certain entity and I want to highlight especially those entities so let's look into them figment is the first entity which has a very low um ofac inclusion rate right so we can see around 0.32% of all figment blocks contain oact sanctioned content if you compare it to Lio which has 1.78 it's much lower right now you might say okay why is it still be um above zero but we will come we will talk about that later then kaken right so if you look at kaken it looks like they started censoring very recently right so you can see also kaken has a very low of ofac inclusion um ofac inclusion um same as other validators like Celsius and some smaller ones like Staker bitstamp qoin bit finex ever stake pool bitpie or crypt stake.

com all of those entities have a very low inclusion of ofx sanction content which might point to them sensoring next up we have relays right relays were a big topic especially especially around the merge um whole meth boost ecosystem was launched in parallel to the merge m CL and it started with mainly flashs being the only very dominant relay so we can see that at the beginning um of the year we had almost 80% of all relayed blocks being censored this was just because the flash BS relay was so dominant and almost all the blocks were relayed by flash Bots this meant that around 80% of all um blocks relayed um were in the end censored by those entities then over the time we saw many other relays coming up like ultrasound relay agnostic agnosis relay the Y estos relay also block start um there's a block start relay that is non sensoring and the situation improved quite dramatically if we look into the numbers we can see the same picture um that we saw that I just explained so we can see that flash plots block native Eden they are all censoring right so they are on the the very top uh on the very bottom of this of this table on the very top you can see the non- sensoring relays and something that is quite interesting for example let's focus on block rout regulated blockout regulated is also a sensoring relay but you can see there was a block in the last 24 hours right so in the last 24 hours there of course mistakes happen and from time to time some ofac transactions just slip through right and we have all seen all these Security checks at our hotel and stuff and you know that this ofac block might have so he he it felt very similar for him right so it was not that it was not that sophisticated so the check was not that sophisticated and from time to time even the non- sensoring relays relay sensor blocks or 2B sensor blocks and in order to do sensoring properly you would really have to simulate that that transaction right it's not enough to just check where the transaction comes from or what's the first Contact so which contract is touched first it's you really have to look into the state and see does a certain transaction touch the state of a certain um sanctioned entity and only then you can you can be sure um that it that I have to censor it or not next up we have the builders and the Builder landscape looks very very scary right at the beginning it was quite good Builders were non sensoring but over time many of the builders um yeah felt the pressure of ofac and now we are around 80% of sensoring Builders this means even if you connect to a non- sensoring relay the probability that you get a sensor blocks are extremely high and more or less as a validator you can do nothing about that you can connect exclusively to non- sensoring relays but it won't help you because the builders are sensoring again let's look at the table and we can especially focusing on the top Builders here so we can see our sync Builder Beaver build flash BS Builder 69 Gambit labs and so on those are all sensoring Builders right and it's not like they were censoring um since the merch but many of them started very recently on the other side of the spectrum we can also see builders like payload dode or moneyold that have a very very large um tornado cach inclusion rate so for example for moneyold it's almost every second block includes a torado cach transaction and for payload de it's also at 28% so we can see both sides of that so some are sensoring very good and some are really trying uh probably really searching for tornado cach transactions to make sure they get included very quickly finally I want to quickly cover a topic um that is quite old um maybe um to ask the audience who has ever heard of a Fed Fork it's it's from Bitcoin very old concept okay one hand nice then let let me explain you what a feder fork is feder Fork is a it was first proposed in the Bitcoin talk Forum by Andrew Miller in 2013 and what a Fed Fork is it's like how can I censor the whole network with having less than 50% of the hash rate and how this works it's actually quite simple let's assume we have two entities right we have Ellis and Bob and Ellis doesn't like Bob so she will not build on top of blocks that contain a transaction of Bob very simple importantly Al Ellis announces that publicly so everyone knows that Ellis with her 20% harh rate she will not build on any block that contains the transaction of Bob this also means Ellis will reor blocks out if they contain a transaction of Bob next we know because Alis has a hash rate of 20% right we know that the probability of having two consecutive slots is around 4% right because of Alis controls a hash rate of 20% and 0.2 to the^ of two is um 0.04 so you arrive at the 4% so now as a minor you might think okay if I propose a block that contains a transaction of Bob then there is a 4% chance that Ellis will reor me out because the the 4% so she can build two blocks and then has the longest chain so for all miners um this creates an incentive to also sensor Bob just to make sure that they will not be reared by the 20% hash rate of Lis and what about fed foring on etherum right what happens if some big stakers announce that they want to reor out every block that contains OFA sanctioned content currently you need five entities to get up to over 50% of the total Stak so if you have Lio stake fish coinbase cracking and figment um all um for example saying that they will reor out blocks that contain ofac sanction content we are already above 30% of the stake and this would then mean so we assume the strongest form of censorship they will reor out and and and announce that publicly and the result will be that economically rational validators would also have to censor just to make sure that they were not being re Reed out the cool thing is if we compare Bitcoin to ethereum in this situation for Bitcoin you can do that with below 50% of the total hash rate for eum you need at least 51% in order to win the fork choice finally this brings me to a very important topic because we' have not seen such a harsh form of censorship yet right so there were no entities at least as of my knowledge that were really trying to reor out of sanction content from the chain but if that would happen we would still have social slashing right so if that would happen we could still in in conscious to bitcoin counter that in um by social slashing those entities taking their stake away if they try to attack the chain through censorship and this is basically all I have here you can find some more dashboards and stuff especially tornado. pic and censorship. pics some open source data if you want to experiment and look into uh the math boost stuff yourself and yeah you can find my contacts and I'm happy to answer any questions if there are [Applause] any [Applause]