Workshop on Rollup Sequencer Economics: Small Group Discussion, organized by Ben Fisch | a16z crypto

Transcript

okay so welcome to uh day two um so just quick reminder of the agenda for today uh we're going to um have basically one hour for group discussions we're going to split into three groups um there were three questions that we introduced yesterday so each group is going to work on one of these questions and then we'll come together at the end of the hour and have to hear from someone from each group for 10 minutes um so let's try to solve these problems uh okay so just remember reminder of what the questions are so that um you can decide which group you want to

be a part of so the first question is about Revenue sharing and specifically this question of if you have multiple Roll-Ups or users or applications which are contributing to the Mev of the blockchain then how do you how do you figure out what each indiv what each participant's marginal contribution to Med is is that discoverable is it possible to design is it discoverable if you're also going to then give a rebate to each of these applications roll ups users based on their marginal contribution um so that was the the first question um the second question was about and and

remember for this question it's specifically focused on uh because this this question has appeared in multiple contexts and and and uh and I as I mentioned it's connected to user uh to to order flow auctions but specifically the problem here is we're looking at if Roll-Ups were to run their own chain then they would be generating some profit from Mev and if they're now going to be sharing a common chain for sequencing the uh the we they want to be convinced that they're going to be making at least the same profit as if they were running their own chain right so the question is comparing uh these two different worlds where each roll-up is running its own auction I had this comparison here

um with the Searchers versus uh giving their um order flow to a sequencing layer and then having the sequencing layer run the auction for them with the Searchers and then give them rebates the second problem is about Mev mechanism Independence so different Roll-Ups may have different approaches to Mev some want to run auctions to try to you know optimize and democratize the distribution or use it for public uh for funding of public goods others want to prevent it entirely some favor of first come first serve ordering policy some favor some content oblivious policy some may favor a random ordering policy who

knows right the question here is to what degree can a shared sequencing layer be agnostic to these different approaches and simultaneously support Roll-Ups that can choose their own ordering policies um and also caveats of that right what what uh if if we did have a sequencing layer which does that what are the trade-offs does it impact interoperability uh and finally there's a question of how a sequencing layer that is not executing for each roll up should charge for transactions so one challenge here is that it doesn't because it's not executing on each roll up it doesn't necessarily have the ability to charge based on the gas consumed by transactions because in order to check

how much gas transaction is consuming you have to execute it in general so it could be charging just based on the size of data but again now everything that gets sequenced has to be then executed by the execution layer that's the whole idea here the execution layer can't choose what to execute and whatnot otherwise it would have the power of deciding what gets included rather than the sequencing layer so how how should transaction fee mechanisms work with shared sequencing layers um there's another question of uh how the sequencing layer verifies the correctness of this fees that could be an independent roll up some minimal State that's just being used for for fees but that also could create a ux

challenge where users now have to pay in two different currencies so there's a question of how to deal with that great so those are the three problems and we're going to split up into three groups I think one is going to be in here and then two are going to be in conference rooms I think the one in here is the only one that's going to be recorded so um see everyone at the hour I think there's like two different problems both of which are really hard here um one is even just to know what the Mev is so first of all at the level of the shared sequencer like when you combine everything how much any of you is there and then maybe even like when it's more

hard to understand how you'd ever know what it is is the counter factual if we're brand on its own what would the me be if it was Standalone right so question one of you is just like elicitation like how do you even find out what these numbers are question two which is completely separate is like suppose magically you knew all of these numbers how would you do the distribution that's maybe the easier problem I kind of feel like that's that's connecting to the stuff quintus talked about yesterday around shapley value and that kind of thing yeah and there's not going to be any like perfect solution but I think there at least you probably get a handle on like different trade-offs if you actually knew knew them but um so yeah so just you know I feel like it's good to just like whenever we're working on something to

be clear in our minds which of the two things we're trying to tackle because they're both I think pretty hard problems so if you had a trusted shared sequencer and the Searchers just put in bids for bundles of transactions and there was kind of like a bid for each individual transaction in the bundle um seems like that would be a starting point for figuring out what would have happened in the Standalone roll-up yeah like a lower bound on how well it would have done yeah so you're saying users at least are well if there's a competitive market for Searchers then you should actually get some sense of how much each individual transaction is adding to the value

so so this would be listening right the shared sequencer level or for an individual rollout uh but the shared sequencer about right that makes sense yeah yeah I agree so that seems so because you could always have like bids for just an individual transaction at one roll up and if there's a higher bid to include it as part of some cross roll up bundle you know then that that's a signal about how much about how much mvv is being added by that sort of so I guess can you really differentiate any of the per transaction because you know it's it's not often that like this one transaction is causing that immediate sometimes it's the case but sometimes it's like you need two other transactions in order to actually

extract that and be correctly I think there's another good point that like um you know how much uh how much maybe is generated by like the interaction between transactions as opposed to just like but like if you think about like the order flow auction Vision the way I hear about it it's often like an a transaction has something maybe and then the OFA is designed to elicit what that is right so that's sort of what you were saying Joe and then conceivably you know there's actually like a ton of extra maybe that's generated if you take different transactions and assemble them in just the right like magical way and that again seems like a harder version of the problem so I sort of agree with Joe's suggestion that just like maybe it's not true but to get started let's just

imagine the transactions kind of have their own well-defined mov and like then then what we then what could we do and then we can worry about sort of complex interactions that'll be my maybe the market would fail for some reason but like you could start with the assumption that you figure out the individual mov because that some Searcher will put in a bid for you know just a bundle like one chance of one transaction right right and then but you could imagine so here's you could imagine like say there's like three different Searchers and they each tell you kind of the Standalone Mev of three different transactions yeah but maybe then a builder is like oh if I sequence them B and then the bundles B and then a and then C I'm gonna get all this extra stuff for myself right so some Searchers see that too and put it

in a bid and say like if I can run a b and c I'm willing to X Y and Z for each of them and if there's a competitive market that should sort of they should bid the entire Surplus out yeah this is great this is a great question I mean the um maybe I mean I think it's a good I mean they have some Alpha right right well no there's something else which is not so hard right which is just that um if you auction off the transaction separately you're sort of the amount of information you're getting total like if it's in transactions you're getting n numbers back um as far as the transactions whereas if you actually have these interactions this kind of like n factorial different

outcomes you might want information about so there's this there's inherent information loss if you only ask people about transactions separately yeah I mean I was thinking you might hope that in a competitive market they sort of figure out how to business more way and extract most of it maybe maybe people could put in bids for any set of transactions yeah and then you do get into this exponential blowout so I'm wondering if there's a similarity between uh like a problem of thinking about Mev for one domain there's another domain versus the combined domain versus the problem of thinking about Meb for one transaction another transaction in the combination of the transaction there's something similarity there because you could make the same argument like we're trying to solve a problem of you know whether it's MVP from this domain or this domain or this role for

this roll up but I think a similar problem that we haven't uh first solved at all is is like that maybe for this yeah just restrict to one domain I don't see how the cross domain changes the economic picture I think it should be pretty much the same problem to figure out like did the med come from this transaction or this transaction um within the same domain itself because I think it's kind of the same problem um and that itself does not assume some like normal distribution of alpha or Surplus which is not the way we'd see it right like a uniswap roll-up would be 95 of the Mev in the current context and you would say if I'm a sequencer and I know that that let's see who's your unit

swap versus like a payments change you're probably going to have the shared state that wants to be near each other with all the interview me on one roll up and then the cross to me the distribution is unlikely to be nice and uniform it's going to be very concentrated around like hot spots of State conventional accidents no it's fine I mean you can make the same argument for like uniform transactions versus transactions from some other contract on one domain and solid free there's some mbb that comes about just from extracting units auctions themselves and then there's some MVD that corresponds to maybe like pulling transactions from some other contract maybe like a flashlight contract or like an away or something like that and the

combined me from both of those contracts contracts I think it's kind of a similar problem I guess you are adding a third entity because there's the users who submit the transactions and then there's this shared sequencer plus the Roll-Ups that sit on top of the shared sequencer so you do have to like split the pie with a third party in this case right with the sequencer yeah so there's kind of maybe these are the two separate questions there's like how you split the Mev between different transactions in a bundle that are dependent on each other in some way and then how you split you know the total Med between the shared sequencer and all the constituent Roll-Ups that are sort of providing more of like a flow over time of transactions yeah so

if you imagine a world where the shared sequencer takes zero then they sort of become sort of the same problem I think right but you're sort of except there's still the question of like these these individual roll-up servers could sort of like walk from the whole thing and say like I'm better off not being part of the shared sequencer then I can capture all the Mev and have to think about like how much they should get over time so that they don't want to do that yeah that was my there's a selection bias for what would want to be on the shared sequencer in in that selection bias it's like why am I using a shared sequencer probably have some properties natural to the fact that we're in multiple domains right like that's why we're getting used in the shared sequence presumably when you're using

the shared sequencer because you think it will grow your business overall right because like more users will want to use your roll up if it's part of a shared sequencer because they can get better atomicity or bridging or whatever you know that they're like maturation of these things have changed right when you launch YouTube no liquidity that's a very differently pre-saturated unit swap versus post-saturated is a very different why I would use to share a sequencer right yeah like as these things mature they have different properties in different and you start to see different infrastructure to support yeah makes sense I mean I would think using a shared sequencer like you you have to give something up like some percentage of the total Mev flow but

you're hoping like the whole pie will be bigger by being part of the assured sequencer ecosystem and yeah so making some meta decision to stay with it yeah that's actually something um I'm very curious about right and we can tie it back to that so so back to the question of just like suppose you even knew the Mev of people combined and what they would be separately right so like really simple two Roll-Ups each one can get like mev10 by themselves now the question is what's that maybe if you put them together right if it was 20 that would be really simple you just refund it like 50 50.

right if it should be like neutral about joining yeah then they're like indifferent now if it's 30 right then maybe the shared sequencer takes some and maybe they each get back more than 10 and so they're sort of happy but like one thing I'm worried about like what if it's 15. right like do we have a sense of that do you have a sense that it's like always sort of synergies between people in the same sequencer like is it clear that more is better always or I'm not sure is it so super additive how it could be where it works it's the worst case scenario is that there's no cross chain mov and then they should still have the same amount of separate mov right yeah I mean if it's I don't know if it's too robust maybe okay maybe that'll be a good sanity check

as you change the properties of block times or something that had some interactive thing like hey we have to inherit the block time to get this network effect thus we have different immediate properties we could end up with like pathological edge cases where um it would be different because if you do the state update on one chain first and that kind of affects like it's using the second chain is using this one as like an Oracle and that's why uh the state gets updated on the second chain as well which makes the Meb there smaller when you're trying to execute that transaction later I see so basically like doing both of them separately one after the other might not actually result in

the sum but that's kind of maybe a pathological case and then on your point about the network effects I'm actually it's just sort of coming out of the Josh's presentation yesterday it occurred to me that could you imagine a world where people are using a shared sequencer just because they wanted a really lightweight launch of the roller yeah and so you have an immature launch you don't want to launch your own set of sequencers because you can't launch a centralized one single sequence there and so you get a lot of immature things launching off the chairs earlier just for a while among which there may be no network effects among which it may just be purely additive that maybe kind of qualities okay that's why we go through moderation cycle properties of ethereum with no State versus

ethereum with all the rich state is very different function so you can see your winner take all dynamic in one of the shares sequencer networks if it gets the place everyone wants a bunch of marginal role it might also give you some credibility when you're launching too like it's a tested stack or whatever yeah also just like credible neutrality that you know people to trust your own um right which is the sequencers because they have devops is fixed let's say that's a fixed cost and they're scaling that might make them more efficient more higher on time so four months it's also maybe possible that you get more MVP out of being part of a shared

ecosystem because the the Searcher Market is better and more mature thank you let's squeeze more out I guess so what is the vision there like if people had to guess I suppose you have a shared sequencer with like 30 roll-ups is the thinking that probably you would have a Searcher that just looks across all of them at the same time or they'd be Specialists I have no idea so you feel like searching liquidity which is if I if I have this pretty complicated cross chain R and it's doing a solver whatever we're calling the searcher's algorithm to start to figure out what it's going to include it needs more complex things to get these like nine roll-up Meb transactions and that search for liquidity might be working on the

transactions could have better you're confirmed faster more likely so one kind of potential Counterpoint to your argument uh where you have new Roll-Ups sort of joint shared sequencing is the fees can potentially be higher like because um let's say there's like limited uh you know processing power the shared sequencer has and there's like it's handling 30 different Roll-Ups what can happen is um if a new roll-up wants to do like lower fees then maybe those transactions don't get included in the shared sequencer is creating it um sequence versus like if they did their own roll up they could probably give users like much lower fees yeah it's a life cycle question again

early in the life cycle that's not a problem but I mean once we get saturated block space or saturated sequencer then we would have to then what we're full what do we do now yeah like adding a new roll up to an already full sequencer like shared sequencers build here well basically sequence for free right we don't care we'll give everything back to you such that we want to boost your network effects to get these Searcher liquidities and other things because it's worth it for us to pay at the beginning it almost feels like designing an optimal currency area would just like what's happening well

there's this big question like in economics I guess like the first thing I've understood well you could probably explain it much better than me that's just a question like should the EU keep adding members like should all countries be on one currency when is it better for a country a small country to have its own currency and yeah it's like you you know you have the opportunity to be part of network effects from being part of a much bigger currency area but you give up local control the monetary policy might not you know always be tuned to what you want specifically and so any lessons from traditional economics we can take for that reason about these well one thing which is a a little bit puzzled by and that is

um are we assuming that there's multiple uh shared sequencers so there's competition in this space and then I think then kind of what you were saying is that uh that division will happen through the competitive process right so um uh the people who are offering the flow will have some knowledge of what they can get on their own and the sequencers have some knowledge of uh the economies of scale or whatever it happens to be and they'll be willing then to pay right and they'll pay each one of the potential inflows a different amount depending upon which has the most

value so that'll come out of the competitive process I think we're assuming there could be competition like there's it's not clearly a winner take all Market but it would probably be a small number I think it would be like maybe the analogies like Airline alliances where there's like a couple and plus one or two Independents and probably like you said the I think in that case they've like hammered out you know some negotiation between every member Airline and the alliance and they're probably all getting a slightly different take of everything right so that's a good analogy in that

there are the the different airlines uh want to partner with an airline which fills in the Gap that they have right and presumably uh for these uh sequencers there's two things which are very similar and there's two things which are different and you want to get the things which are different in order to maximize uh the agglomeration benefits like if you have buyers and sellers and you want both of those right but you don't want all buyers or something like that on the other hand if you cross chain and maybe you want

the similar ones together right because that's where the Arbitrage opportunities are right like if you have two two chains of amms you want on the same C Class here so people get efficiently Arbor so that makes it sound like it might be a winner take all market then with the one caveat that I think in cryptocurrency space sometimes like natural winner take all markets people don't want to appear like that there's one big centralized piece so there might be some like pressure for appearances to not seem like totally unified in one shared sequencer

yeah we didn't talk about the returns like vertical integration saying like it's the same entity all the way through the stock and how how much Advantage does that give somebody again we try to design these things so there is no Advantage inherently so that you get no benefit from knowing things that no one else knows but we've seen it in PBS a lot that are you know some of these things are integrating and getting access to flow that other people's arm and because of that they can quote better they have a cheaper rate on finance so they can quote tighter on their on-chain off GNR I think those are things that aren't naturally obvious when we're designing some of these things and they might they might appear into what what

um like so what what what what caution should we have then thinking about the design from that like what are the lessons that we should think okay because some of these things were emerging not obvious I guess it's the bigger thing I think people were obviously aware that PBS could have some centralizing forces obviously so I'll make a conjecture that we talked about like the two different Revenue split questions like you know between individual transactions and between the Roll-Ups I think the first one needs to be solved algorithmically because that will happen like a million times an hour um

like different transactions from different chains will be bundled together and you have to sort of split the you know the fees or the Mev like rebate the second one might be might play out in a very slow timeline like I don't think that Roll-Ups will switch between shared sequencing layers like multiple times a day or even like you renegotiate your contract every couple years or something like that yes I think that could be solved in like human level Yeah by negotiation around it I suspects those ties will change very slowly so you don't need some like automated mechanism necessarily I guess we solved it well I was gonna say I mean it feels

like we could still attack the first problem independently which doesn't even have to do with shared sequencing necessarily elicitation or which part yeah yeah um in the second part you could still try to come up with guidelines for like if you are the shared sequencer or the roll up and the negotiation like what should you ask for when should you decide you're better off independently or yeah like you say that's probably not programmatic right that's probably outside of that yeah well you still might want to agree to some programmatics but though so that like as your volume changes over time your payout changes right like you probably won't just agree to like some static enough per month but

yeah the interesting thing is it's kind of like information review in game two which is so let's assume that the people who are making this decision on the role of service providers and they say our contract is five percent and whether this is public or not although five percent of what we still have to answer that question right yeah and then I guess my point was more on the interactivity of the next person say well we'll do it for two percent or we'll do it for one right if a lot of these contracts as far as I know are done occasionally and people don't really actually know the competitive Dynamics unless they're in negotiation with private information there's not like actual selective pressure on the fees there okay so you so Joe you suggested as far

as like five percent of what so One initial stab would be like the sum of the winning Searcher bids basically right I guess that I mean that I guess there's pretty weird incentives for the shared sequencer to collude with uh Searchers to keep those bids low right so how do you audit the not if the shared sequencer is also getting a percent of the same thing although maybe they haven't okay the worry would be to extract a lot off chain right that'd be the concern right um so there's a question like auditability of the shared sequencer is kind of overall Revenue which seems tricky I see so you're saying like they they collude to have like lower nominal bids but they're getting paid off there's

like a side agreement for this you need like a crms basically and say like this is what we expect this is what we're seeing and this is the Delta that means that this person's polluting and not forgotten yeah dydx is working on this for their kind of like shared order book yeah I don't know how many they have too 20 notes but like they have like eight order books you have to figure out is one node extracting a bunch of ndv but to see that they have to see some common order book but they also need some very important queer design decisions or isn't it like only posting trades on treatment so like all the ads have canceled over

there so it's like I can spoof isn't that the point to do spoofing Mev that's like completely unrealized effort reviewers insurance and then like the dream of the intense model though is like that you don't have to publish on chain all this stuff that's never matched or consummated or whatever yeah but if you're trying to get an audit Trail for for games you definitely need to know something yes just kind of kills your efficiency though Ryan sure good that you are being entering there's probably something possible yeah I mean I suspect in practice people

will go for the efficiency rely on some like trusted brand you know that which would make it even more winner take all but like if the shared sequencer has a brand where people like trust that it's not colluding or it's giving a legit audit Trail without it being Unchained and isn't that how flashbots basically works today well PBS you pass and in fact by accident because like none of the none of the builders have turned off of the historical data from all the options I I think like eventually they will

someone who will be like okay but right now it's like social consensus on everyone broadcasts the historical data involved into November how's that authenticated though but you're saying taking those at face value you could check that the highest beer was the one that was selected yeah so long ago you can also use that to see if people are stopped tonight you actually feel like all the small

Builders are paying subsidies right now like pretty high sub studies like 20 0 . so forget about me you could probably make that slightly better cryptographically if you like committed to the full set and then you could sort of proof individually to people that their bids were actually included right I just don't think yeah I feel like the latency aspect of this week well you could do that as some like post-hoc auditing thing right that's not really yeah yeah I mean yeah right now as we you could actually assuming you trust them so if you're worried about this

collusion problem or like accurate data reporting though like I'm trying to think if you could solve that by changing what the the five percent or what the take is actually computed over because there's some other signal that's harder for them less incentive for them to manipulate should there be competitive forces at some point though like assume the Searchers can pick from different share sequencing layers if they're I guess you're saying that the there's a solution The Searchers would be in on this like collusion right so the Searchers and the secrets here would be taking the cut that should be going to the Roll-Ups because it's basically it's like a three-party deal right um between like Searchers Roll-Ups and

shared sequencer and there might even be pressure on like other pairings of two of those to collude right I mean you do already see this where like the builders that are Searchers definitely are shaving their business this is what I'm saying it's a crazy historical news yeah I don't know how long this will last it's probably only another eight weeks until the next update but isn't that is this like an inherent principle that anytime you have like three parties trying to split some part like two of them any two of them were incentivized to collude there is a divide the dollar kind of example which is just like a very very cartoonish version of this so three people try to

split a dollar right the point is like no matter how you try to do it there's always an incentive for two of them to make a side agreement so that that implies like even if it was five percent of something else like there's kind of this fundamental problem in this space that like this all is supposed to happen pre-consensus that's sort of for I guess is what Turin was saying a second ago but like either you commit to everything at the consensus layer in which case he was like new to the benefits of having l2s

or you sort of have to trust somebody because not the data is not all going to be like authentically committed I think at that point you're running everything on chain yeah Ben's argument would be you get some heterogeneity versus hey we're just specialize in English and we're making some trade-off that's very optimized for the specific kind of back and forth and that will be the marginal group versus just committing to that I think there's also some Communications

I mean if everything is fhu it's unfortunately that's not tremendous yeah I do like the combat I think who made this but the idea that like you know competitive forces plus kind of like you know like if you build a reputation as a sequencer you will probably be able to extract more um from that and you know there's a question of like how do you get that reputation maybe you sort of enable audits or what have you but I wonder if like one from my perspective like one sort of maybe weird analogy is when I think about like Publishers like for books and like the contracts I sign with them right so like For Us distribution I might work with

Cambridge University press and I sort of trust them more or less but then when I do like the Russian translation at one of my books right I just sort of say like pay me a good Advance because I don't expect to ever see another dollar after the advance basically and so you could imagine similar kind of see some shade sequencing layers out there with the same trade-offs yeah empirically on l1's there's the biggest trading infrastructure service provider who's generating transactions for most of the big trading desks I always spend time with developing me like look our customers care mostly about inclusion they need to get their transactions in as fast as possible they don't care if they're getting they're

going after the public medical whatever gets us included as fast as possible that's what everyone cares about the emotional value of that I think they do like a couple billion dollars a buying a day through that specific kind of condition is that we don't care it's all we care about is reputation and conclusion like that's what they really that's how they're making their transaction program just just the two you're saying they don't care about me or they don't care about ndv because their customers like all their customer cares I won't get this included as fast as possible and so if they were putting it through some shielded pool with 10 of the hash rate or inclusion right they would be they would not do that because they want everyone to see it they want to get as good as possible so they don't care

about getting interesting not really is that are we sure that's rational like do they understand what they're actually giving up like getting Fun Run technology their flow does not have a ton of MVD associated with it it's generally the second leg of a settlement of the trade like you're just settling it here sending you this asset so it's just a transference it could be just a transfer right like generally most trading is like here like here you want to see one and so front running is not a giant then if they're running much more complicated on chain trading strategies properly but they don't care about that which I was shocked at but I see so

they're they're doing the price Discovery and so forth off chain possibly so there's no information in your uh but again their customers care just they want to make sure everyone's using victims basketball I guess the reason I brought that up was really these transactions are very heterogeneous in terms of what people care about and why and why they're going to broadcast into one set of circus versus another sometimes just exclusion School is it possible that different Roll-Ups would somehow work with multiple shared sequencing layers or is the only model we could think of that like they're tied to one shared sequencer and they're sort

of exclusive with it really I don't understand how you reconcile a frog-proof layers in terms like somehow you have to get a mistake so they're not like multiple redundant databilities because there's a model there's a there's like a raid zero model where you just like write the data separately to every data availability layer and then it seems pretty obvious but if you're like striping in half half then it I agree it's a bit weird the the bring your idea availability

thing is allowing the developer to choose not a portion of their stated so you could write it I I feel like it's it's it to me it seems to me right so what would it look like though if the role of server like had multiple shared sequencers that it was partying to it could just kind of alternate blocks or do you think the shared sequencers are

ultimately both posting to ethereum say or well you mean like all the Roll-Ups like are they in the same ecosystem like uh the two I guess I mean imagine it's a venture well-ups that sit on top of ethereum yeah and then there's like multiple shared sequencers and you're like for you know I don't want to be like tied to Coke or Pepsi here I want to use both of them because if clients just care about Max inclusion you could say like oh will we take we take transaction bundles from multiple shared sequencer layers I guess maybe a simpler question is can the roll-up have a shared sequencer

and also have their own sequencer so that like clients can decide whether they want to collect them to share distance or just yeah I mean the trick would be like what happens when some Searcher you know bids a lot for a bundle that includes transaction X and then transaction X is like preemptively executed you know through the direct sequencer or some other shared sequencer and then then it kind of like preclude some entire transaction bundle like there would be these weird dependencies I guess I don't

even heard of this concept because is the thinking more that um the types of transactions that you would want to sequence together uh is that going to be changing across different Roll-Ups all of the time or is it going to be the case that sort of two Roll-Ups have a natural kind of uh connection and that it's going to be a more permanent thing that these are the ones you're going to want to sequence together I think it will probably like drift over time but you know in the short run I think there will be some stable sort of pairs

that see a lot of you know uh synergy but I guess like I mean you were saying a second ago like some some clients like have a very simple flow it's just a bunch of Trades so maybe they're like we don't care about the shared sequencer we just want to submit direct to the roll-up and sort of be done with it and there's not a lot I may be so who cares and and then others might be like oh we should go through this shared sequencer so that there's more you know cross chain and maybe you captured the non-obvious thing if you're hearing these like equities where you're saying like Okay I'm just doing Arbitrage I'm buying Google here and selling Google here is that we don't have one-to-one fungibility of assets

actually if you're on different Roll-Ups on different l1s right like they are actually literally different security assumptions trust assumptions throughout the stack and so if you're trying to do this cross domain R you're actually buying and selling two different assets at the end of the day you're buying like uh our return youth in optimism which are not the same thing and so Arbitrage is almost the wrong word for it because it is not a one-to-one swap and so a lot of these like intents these cross-domain ethnicity ideas people think they can be deterministic like zero one or binary but they're really like a probabilistic curve of like how like are these two assets it's like how like are usdc well usdc is one to one right it breaks this

concept because it's all racking back to Native yeah exactly so like the syntheticness of it is actually a probabilistic score that when you're doing Arbitrage you have to assign the likelihood that you're going to get back to the actual data so that's actually much trickier than I think most people realize the Searchers have to have a tolerance between zero and one it could be 0.7 it could be 0.2 when they're doing these cross chain cross domain things and that brings in a lot of discretion and choice in like Market courses which is not as synonymous as people think with our church if we assume number one form of Meb is centralized decentralized R for the same

asset that actually makes sense like ethon finance and Ethan yeah it's counterparty risk but when you end up with the assets the same thing and so my the reason I bring that up is if it's possible for the sequencer to improve that in some way or give you better I don't know how I even thought through enough why share sequencer would provide that would be something that would be an edge and be interesting people so right now the shared secret server just give you the ability the topically do the two trades but they wouldn't let against each other um where's their da I think it's it gets

much more complex the more we break apart what the asset is like how much heterogeneity we have in assets makes Atomic Arbitrage harder am I will also decentralizes them guaranteed at the same that's right you're saying like two different things yeah yeah exactly they sort of have to agree on those elements and get produced back to like oh and graduate yourself once equals okay which is where a lot of these Network effects would show up it's like

I can get better and better and better more liquidity I have or sequence did anyone know how about okay stacked for changing the semester work or do they have they chosen uh because like I don't understand if there's all these different development as far as people each will sequence you're stacking somehow you eventually can go back to op mainland is that the same sequence or do it does like every sequencer sequencies open Snapchat and have to guaranteed update possible I don't know what the guarantees are too bad now Kevin from optimism on Zoom address theoretically

hilarious oh yeah wow wow yeah sorry honestly it's with this the audio is not great uh I didn't do it with the stack though uh yeah the question was uh how does like Bridget how does like baby virgin but how how to cross a psac role of transactions work does a sequencer who's sequencing an OP stack roll up have to also have be sequencing the mean open stack chain or do they have some guarantees on like happy inclusion are there any attempts that guarantee is only inclusion to the mean

okay yeah we are still here to create a like honestly the approach right now with these topics and everyone's using the exact same integration and you know all the same modules that like really easier for us to figure this out in the future so it's kind of it's kind of annoying so I think right now it is it's in the research basis that's having a lot more limited at Carl um but yeah I don't think I don't think they would need to use the same sequencer though social division yeah I think um

to verify you know caution messages between a little distractions but do I get any verification do I get any guarantees that the main sequencer will verify proof within K blocks on the main chain at more than the the Arbitrage standpoint of like suppose I was doing Arbitrage between op stack chain and the main chain I want some sort of like latency guarantee or maybe guarantees the wrong word but some type of probabilistic measurement of how long until I get included is there any like form of

trying to bound that or make that like yeah like like do you think the like main chain sequencer will have some inclusion guarantees where like some portion of a block is allocated to verifying other op stack chain proofs or something like on every block oh sorry I was cutting out um basically the question was like on the main op stack chain will there be some portion of block space like allocated just to verifying these messages versus it being competing

with on both being made yeah yeah that's currently not a part of the design of them like kind of like functioning us um like settlement change or something like value here just exclusive access like something reserved for the the cross domain stuff versus it competed with all the other uh blocks yeah it was really not a plan to be a distinction of like oh he may not sponsorship you know that's true well that's a cartel that's interesting

sequencers that made that sequencer have to have come to some economic agreement to guarantee this verification otherwise you it's just like two separate chains then that's just like the normal Bridge you know I mean is that a simple problem even if

you put the economics sort of aside I mean it's well a form of this Arrangement might be like other MP stack chains if you if you pay some amount of op you're guaranteed include you you know some block spaces of Reserve metal defense verification but like you even suppose the other of each game could prove it was running like exactly though he's stack with their body who's presenting from the first group I think to summarize so we talked about how to how to share uh revenue from Mev um I think we said they're kind of two

distinct problems that we were maybe getting muddled together a little bit which is uh how to how to split the movie between different transactions that are included in a bundle um and then how to how to share it kind of three ways between Searchers Roll-Ups that are on top of a shared sequencer and the shared sequencer itself um I mean the proportion that uh Searchers get is probably tied to solving the first problem and figuring out how much each transaction is really adding any V wise and um write what each transaction should be paid so the the first problem I think we said it's hard because there's kind of an exponential space of different

combinations of transactions and it's hard to elicit the value of each one before you can even start thinking about like a you know uh shapley values or ways to split the value up but that problem kind of exists even without a shared sequencer so maybe it's more interesting in the shared sequencer case because the space gets even more complex and more transactions go into a bundle maybe and maybe the values are you know in different assets on different chains but you know fundamentally the same problem but more complex and then the second problem of how to solve kind of the three-way split between a shared sequencer the constituent Roll-Ups and the Searchers um the good news I guess on that problem

we felt like we had maybe even less of a sense of how to you know what a good split is but that problem between different Roll-Ups that are sort of joining as part of a shared sequencer yeah the shared sequencer itself um and then also Searchers that are submitting uh bundles of transactions um but at least the split between the different roll-up servers on the shared sequencer and the shared sequencer itself um we are thinking that that won't need to be solved nearly as often or as quickly because roll ups won't be switching between shared sequencers really quickly that will be much more of like a sticky Arrangement that they need for a long time

um we had some questions about I guess how permanent that would be like what the feasibility would be what the friction would be for uh roll-up server to switch from one shared sequence one shared sequencer to another we're also curious if it made sense if there was any model for a roll up to take transactions from different shared sequencers in some kind of arrangement or if they're sort of inherently tied in a one-to-one relationship with a shared sequencer that we kind of came up with at the end so I was curious if anybody else has thought of that before thought if there's any models just repeat that last question uh if there's a model for a roll-up server to um work with multiple shared sequencers

in some way or if it's inherently kind of a one-to-one relationship like a roll-up server takes all of its flow from one shared sequencer and that's it it could be multiple shared sequencers or it could be one shared sequencer plus it has its own roll-up specific sequencer kind of like a direct submission pathway well you're essentially asking if you can if there's a way in which you can use multiple consensus protocols to agree on your order well I mean you could take some kind of logical composition of what the outputs are but or you could you could possibly like round robin between them um right I mean I guess you could you're like essentially designing like a meta

level consensus protocol that's now taking individual consensus protocols as input so you're saying it sounds hard no it's not hard it's not clear what I mean it may compromise on the properties that each give to you but there are compositions of consensus protocols right I mean even ethereum is a composition of two consensus protocols essentially right yeah so I mean we're thinking about that because that might change the Dynamics of basically like the easier it is for Roll-Ups to switch between shared sequencers or take transactions from multiple they might be able to demand sort of a higher cut of the Mev because they would have like more ability to leave and get a better deal on another shared sequencer or just do

it themselves was the conclusion that is the difficult to just figure out a way of um estimating the marginal contribution of each roll up to the Mev or in other words how much they would be making on their own and and therefore it the conclusion is that it's it's at least we don't know how to solve that yet yeah I think the conclusion was like we don't know how to solve it you know we were thinking of really simple things like uh the shared sequencer and also the individual Roll-Ups get some fixed take of Searcher bids for different transactions that are included in a bundle um and then you know somehow some of that

goes to the shared sequencer and some of it goes to the specific roll-up server for each transaction that's something that you could at least you know so so like sorry to make this interactive but I feel like the simplest the simplest question that is related to this would be let's say that you're you have two transactions two items that you're auctioning off and uh nobody really wants these items individually they just want them together if you were to run if you were to run two separate auctions for them then you still might have um bidders bid on both because there's a

certain probability that they will get both and they're not going to bid as much as they would for the pair of them if you ran joint auction because of there's some price to that risk so if you were is there a way to run an auction where you where where where you can discover how much they would bid in the separate auction case even though you're giving as an as an option yeah the the bundle okay so let's maybe we could go through that because that I think can illustrate the two different problems we were thinking about so I think the classic things like a left shoe and a right shoe yeah that's great yeah yeah um it's like I I've I want both shoes so

it's value I the pair of shoes to me is a hundred dollars I don't want the left I don't want the right but I'm willing to bid on both because there's a certain chance I'll win both functions and say like for some reason the left shoe and the right shoe are being auctioned on two different chains yeah two different roll-ups um so yeah so I mean we're thinking like the first problem would be uh basically figuring out you know what value those two combined really gives and that was the first problem we were saying that's the same whether they're auctioned on the same chain or on two different chains right um you still have like Searchers can put in a bid for the bundle of two and you know in the simple case like you can

notice that Searchers are willing to bid much more uh for a bit on both uh right than for separate bids I guess my question if you're running the auction though for both why would anyone bid individually on each if they can just submit a bid on both like why would they truthfully report like if we were running in separate auctions this is how much I would bid on each if we're running an auction now that's giving the pair so I mean if the value is like truly zero then you would see Zero bids and I guess you would impute like a value of zero if you don't see any bids but somebody might bid some Epsilon right like you said there's a probability that they end up with both or they end up with one and sell to the person who has the other so

there's probably like a non-zero price even just for the right shoe um oh interesting um so you're saying that you would you could still have if you give people the option to bid on both or either one or individually then you think that some people might still bid on one because they can then sell it to someone else but they won't have to compete for the bid on both or something yeah exactly so like suppose someone puts in a bid for the right shoe for like a dollar and someone puts in a bid for the left shoe for a dollar and then you see like a bid for both is like a hundred dollars right um then that gives you like some signal that like there's a lot of value of the two together and then you know you you have to figure out like a way to

to split that Surplus but at least you'd have some idea of what it is is there like a theory of this problem in auction Theory or yeah it connects pretty strongly to the literature on combinatorial auctions um there you don't have a notion of an ordering like you do on transactions but it's the same features where you have like an exponential size set of things that you're worried about and so people's preferences might be over that very rich exponential size set and then you have to decide on the bidding language to do just like elicit bids on individual items or do you allow them to bid on bundles and if so how do you compute prices so that's all I think somewhat relevant to this discussion the problem is just uh like we said like getting bids on every sort of subset of items and also like exponential number

of subsets and yes so I guess the the usual if you think like say Spectrum auctions or something like that which are kind of a real world deployment of Commerce real auctions there a lot of what you do is you're kind of like lazy about it so you sort of like if people you don't you don't sort of upfront ask everybody for their bid on all exponentially many things you're like look if you feel like bidding on this bundle like feel free and then like we'll figure it out at the end and so there are like there are these rules about how do you charge prices based on bids on different bundles in that literature which perhaps one could draw on I don't I don't know that I'd be as concerned about the exponential number of uh combinations because if we have let's say five

Roll-Ups and that we're looking at sharing Revenue with I mean there can be many more of course but let's say there's five in particular that are trying to come up with a revenue sharing model and the builders are building blocks wholesale for each so it's really just bidding between either building a block for an individual roll-up or some combination of subsets of the five which is tractable yeah but one problem with some of these things is they're communication complexity is not low like you might have to have many rounds like that the the resale of the shoe is the second is adding extra communication complexity right because you have to have another option to retail well what level of communication are we talking about here if there's only five items

that are being possibly exponential in five well basically potentially five 32 round are you willing to do it okay 32 rounds would be a lot that's what I'm saying right like you have to you you shouldn't kind of underestimate that right right and actually one high level takeaway from that commercial auctions literature is it this specific case they're like zero zero and ten together yeah is like really hard in like many senses actually I mean if you know it's only going to be a bounded number of um like transactions where you have this complementarity but like if it might be like large sets of bundles that all of a sudden unlock all this value that's just like very very difficult to deal with options so you know at the moment we haven't even

solved the case where in my opinion we haven't even properly solved the case where the transactions don't interact at all right and then you can also Imagine okay let's go to the case or maybe you get like a 20 percent you know increase if these things go together what can we do there right that would be my advice to focus there it seems like this problem arises in other Industries right like you have shops in a mall I mean there's other okay maybe it's not addressed through through auction Theory but but it's a similar situation where they you could have one flagship store that's contributing the most in terms of Revenue and should get a discount on its rent but like you say you would not do that with an auction right right long negotiations right lots of people at the same time at the same

time right cool okay maybe yeah go ahead yeah I was just gonna say yeah so I mean so we were the the stores in a mall is probably a place to end on because yeah we were thinking like this uh the combinatorial auction part is kind of like needs to be solved algorithmically quickly it's a hard problem it exists in some form shared sequencer or not yeah and then with shared sequencers uh with Roll-Ups joining your shared sequencer or splitting off on their own deciding what to do that's something that you could start by just like hammering out an agreement you know like a paper contract slowly around the table and assume that you know your roll up's not going to change more than yearly right or at least not super fast

so that's maybe more like the stores in the mall aren't literally having an auction but they do have some like slow back and forth and they do give the big corner store better rent and it's probably all not very transparent and maybe not like completely efficient but they approximate it with people negotiating so yeah we were thinking like maybe that maybe that happens with Roll-Ups deciding to join the shared sequencer cool all right thanks did I miss anything from the people who are in the group great okay um I'll do the group two um we can pull this white board this we this is the mvv

um agnostic sequencer so um it's the idea here is that we have some shared sequencing layer okay again we can just think of the sequencing layer as ethereum itself it could be something that's a module on top of ethereum which is uh you know maybe more optimized for data availability and ordering but we can for for purpose of the discussion we can we can just think of the base roll-up model where basically Roll-Ups are doing computation but they want to use the

decentralized layer 1 for ordering and availability of data right so um let's say now because this because ethereum itself is not really implementing threshold encryption or time boost like first come first serve ordering policies or other kinds of ordering policies uh to what degree can these different Roll-Ups running on top of this common sequencing layer have their own ordering policies so let's just look at a few examples so let's say that we have this roll up one which wants to do some threshold encryption so what we talked about is how this roll up one could could designate its own and

we'll talk about the caveats of this in a moment but it designates its own um foreign threshold encryption set okay it designates its own set of servers this could be involving the governance of the roll-up or not whatever it could be again it could be decentralized as well the the concept is simply that if each roll-up were running on its own trying to decentralize it they wouldn't each be able to end up as decentralized as the sequencing layer itself so for purpose of the discussion we're going to assume that every roll up can designate a set of servers it won't necessarily be as as

decentralized as the common sequencing layer but they can still introduce it and that will be important for the discussion of what are the trade-offs here of having this separate set of servers that's that's in charge of something so in this case the set of servers that's specific to this roll up will be threshold encrypting every transaction so every transaction that a user submits uh so the transaction is valid for the roll up if and only if it's threshold encrypted under the correct key and uh the users would would just be submitting their transactions still to the sequencing layer threshold encrypted under this under this shared key and uh

it would be ordered along with everything else that the sequencing layer is ordering and at the end when the roll-up reads from the sequencing layer uh then this set of servers will threshold decrypt the transactions at execution time okay and if something is not threshold encrypted it would just be dropped it wouldn't be considered valid uh let's let's go for another example first before we talk about what could be the the uh in other words what are we losing by introducing this extra set of servers are we still gaining a lot from the shared sequencing layer what what properties are we gaining which are we losing so uh let's say we have a roll up two that's doing first come first serve

so in first come first serve generally there's uh two components right there's some Atomic broadcasts or consensus protocol there's also some set of servers that need to agree on timestamps for transactions um a transaction needs to be sent to every server participating in the time stamping and then they need to broadcast those timestamps to each other and there's some mechanism to end up agreeing on what the the outcome is um of course there's the condorsa Paradox but we're just you know ignoring those uh those those uh details so the idea would be the same designate a set of

um time stamping servers and the transaction would have to be sent to each of these servers to get time stamped before it would be considered valid by this roll-up so that we'd have this you know time stamped transactions would then be sent to the common sequencing layer an important thing to note here is that when you have some kind of first come first of ordering mechanism there is a much sharper trade-off between the latency of

the consensus protocol and its decentralization because the transaction needs to be sent to individually to every single server to get time stamped uh and and then they need to broadcast their timestamps to each other whereas in other consensus protocols there can just be a leader that collects all the transactions and then broadcasts uh you know a hash of the transaction to like basically Erasure codes and shards and makes available the data and collect signature so point being there's more restrictions here in terms of what needs to happen uh in a very decentralized uh set of servers in before a transaction gets approved so here we would definitely be looking at a

set of timestamping servers that's much smaller than the set of servers that are participating in the sequencing layer um so now the question is okay well sequencing layers that are common they're supposed to provide us with a few different properties so first censorship resistance right these are things that come from decentralization censorship resistance um certain economic properties the fact that the leader in a consensus protocol is elected once in a blue moon if the distribution of stake is very

flat then uh it behaves in a short-sighted way so it will just take whatever it can get to build the most valuable block it won't try to engage in long-term Revenue maximizing strategy so this is important for for like EIP 1559 um it's important for PBS it's important for proposer Builder separation and then finally there is the finality guarantee which or non-occupation right one something has been stamped as final buy this sequencing layer then based on this the the the fact that

uh there's a lot of stake behind this that will get slashed if there's some equification later um the user knows that the transaction is final okay so now let's look at which of these properties are preserved when Roll-Ups are introducing this extra set of servers so the first thing that goes out of the window a little bit is censorship resistance I'll say at the end maybe one way of recovering it but this is important to recognize this uh set of the servers that are in charge of threshold decrypting can now censor transactions this is not entirely true for threshold encryption because if we introduce some kind of time lock encryption mechanism because the user doesn't the user threshold encrypts the transaction on its own it doesn't have

to first send it to the servers for approval um so if if this is somehow time lock encrypted but the trapdoor is secret shared among these servers then that might be okay for censorship resistance but certainly with first come first serve sequencing because the set of time saving servers needs to receive the transaction first they can the the weakest link is now for censorship resistance is now this set of time stamping servers okay but censorship resistance is not the only thing that the shared sequencing layer is providing um so to the extent that these ordering policies like threshold encryption or time stamping don't entirely control uh the ordering and they don't necessarily

control like Atomic dependencies between transactions other things that a builder Network might handle the ability to be compatible with PBS and also the pricing of getting things sequenced on the on on this common sequencing layer um can still benefit from the decentralization of this in terms of anti-monopolistic properties on the other hand it's important to recognize that if you have the ability to censor for your roll up then for just transaction inclusion you can jack up prices right so the time stamping servers could could raise a base fee that could be well above whatever the base fee is for the sequencing layer but on the other hand when it comes to this to Mev that cannot

be extracted just by these time stamping servers or just by these special encryption servers we're still potentially getting the benefit of this common sequencing layer in controlling for Mev um then finally uh finality so we still have the finality from the sequencing layer one way of thinking about this is like let's say that a roll-up has a designated Builder a centralized builder for that roll up right would there still be a benefit to running on this common sequencing layer in terms of the finality that it's giving and the answer is yes so a builder can censor transactions but it's not as trusted to provide finality right the Builder could still cause double spending if it tells

someone hey your transaction is final and then equivocates later before settlement on the L1 so in in many ways the sequencing layer is now acting as a notarization service which says okay Builder what do you want to uh you know have final okay now this is final right so the sequencing layer is still providing this final stamp of approval which gives users confirmation that their transaction is done and won't change before it's settled on the L1 um and lastly when it comes to censorship resistance we talked about how what we could do is have an ordering policy that's just for the beginning of a block and not the tail and not regains some censorship resistance so for example we have a policy that's implemented in the rollup which says the beginning of every block has to be

threshold encrypted but the end doesn't need to be or the beginning has to have these time stamps on it but then there's some tail of the block that doesn't have to go through the timestamping servers and that's interesting to explore more in terms of interoperability we talked about ways that we can still have a builder a shared Builder that can say get make some a promise to a user that will only include transactions if they execute on both Roll-Ups that is still possible but we need to be careful about making sure that the threshold encryption or the time stamping isn't on an already valid transaction that there's some step where the Builder gets something threshold encrypted or gets it

um it's easy for threshold encryption that gets like a part of the transaction timestamp before it adds a signature so that these sets can't submit the transactions independently the Builder can still participate in an auction and only submit both transactions if it's able to do both uh okay so that's that's problem two all right we'll try to go through who is this working uh problem three like really quick um so problem three like it's somewhat similar to like what was discussed in problem one um what we roughly talked about with kind of the user experience of how transactions are going to be submitted through like a shared sequencer and a lot of the questions was around payments and so how would a user pay for a transaction to be included on a on um one roll up or many roll up door like a shared sequencer and what is the chain

of payments that needs to happen if we assume there's kind of three layers you know where you have like a roll up and you have a shared sequence or you have like a dla or a settlement layer whatever you want to call it later right presumably the user probably wants to own like one token um and they want to pay for that token if we think about the existing user experience using like you know like uh eth right if you're gonna use like unit Swap and you're going to submit a payment for uniswap uniswap have the front end the user is actually interacting with the front end and signing a transaction that front end needs to interact with some stateful RPC to actually generate this kind of transaction bundle to give the user the opportunity so we assume that flow is probably going to be the same and the

user is still going to interact with the front end or something to generate a transaction and then there's a question of when you get a collection of roll-up transactions how do those get submitted to the shared sequencer where we Trend it to just trying to like really summarize this it's like like essentially like a PBS Market where we assume there will be some Market whether it is off chain or whether the shared sequencer itself has some kind of you know protocol owned building or you know enshrined PBS solution such that the transaction they're submitted to the shared sequencer would be submitted by sophisticated actors and then the key benefit they provide and that these sophisticated actors can be assumed to be running this kind of multi-roll-up knapsack problem to determine what is like a good block to submit to the sequencer because one of

the core things you want to resolve here is that um like we use like the shared sequence like a roll up up here right a roll-up is going to have some stateful execution Loop that is going to have some limit on the amount of execution it will do for like a given amount of like fees and or like it should have like a strict computational limit right to like be able to run a node the shared sequencer is strictly like not going to be aware of that limit so if you include a bunch of transaction in the shared sequencer the subset of those for any given roll-up May mean that there's like 100 transactions and only like 50 of them execute that's undesirable that's structurally pretty similar to like getting reverting transactions and so we model this like similarly to what you saw with like priority gas auctions

prior to like Mev guest right where you can say you can resolve this by having off-chain actors that are economically incentivized to be State aware of a given roll up such that when they construct what like a bundle or like you know a block for the shared sequencer they are actually aware of the state and thus the limitations of the execution that would actually occur on one or more Roll-Ups um from the shared sequencer side again you can question of like whether we actually end up with ever having transactions as in like a single roll-up transaction submitted a shared sequencer or if the steady state is a shared sequencer is only ever really doing what we see with these protocol owned um and like optimistic relays where the shared sequencer says I don't care what is in the block that's up to like the roll up

that's not my job I want a set of pipes and the highest bid for that and that's what we're seeing in like East validator for like optimistic relays right now they just want what is the block that pays the highest bid that's what we see a steady state potentially reaching the other questions that we got into a little bit was what is like the fee payment mechanism on the sequencer right you can say the sequencer presumably has to still have some mechanism where it can do accounting on it such that it'll actually include a thing kind of giving us these kind of like economic properties and like finality of like you know non-equivocation like if it actually really accepts like a bid why would it do that if it didn't have a guarantee that but it was actually paid that's easiest if it's just on the

native state machine or the shared sequencer but it could also be some you know various the models that like the Suave guys have like proposed right where you can say okay you're gonna post this on this bulletin board and the payment settlement is going to happen it could happen on the roll-up it could happen on some settlement layer it could happen on like eth it could happen like out of band on like venmo whatever but there's various designs for how the actual payment is going to occur of this is the proposer for this round of the shared sequencer they're going to choose to accept the block how they do so is a pretty broad design space but yeah so in somewhere somebody's like users will have normal interactions with RPC nodes for given Roll-Ups and you will probably have some sophisticated group of PBS f

incentivized actors who will say I'm only going to create and submit you know blocks of transactions to the sequencing layer that are not going to revert for the execution layer because why would they bother to include these transactions that are presumably not valid someone else on some other layer and various designs for how you can kind of hold that accountable um useful things to note right it's like standard as like you know uh Epps it's like a still pretty novel kind of design space we've seen some stuff from like the skip guys and protocol and building on like osmosis um question of how much State needs to happen in the sequencing layer and how generic that needs to be to satisfy um various kind of inclusion mechanisms um or you know how you're getting a guarantee of like the execution the

guarantees generally we should assume that these like off-chain Builders are like semi-trusted people that do have the opportunity to like grief an end user but it's going to be essentially like a reputation based system as we have with Searchers builders in like East today did I miss anything from the group I know we had like five or six things I think there are a couple of like Newsradio I remember the one that we missed no I think that's pretty compressive the only thing was uh we also thought about what if like there's Builders might be malicious and like and for that there are like different methods like I think one of them is like that's short-term T using Tes like having them run being run on tees the other one is like later on proving that this is the gas cost via succinct proof but these are like future steps uh

generally this converges to the normal kind of PBS design space which is rather broad but we think there's actually a lot of like good parallels it just is okay does your knapsack problem because like combinatorily more expensive because now you say I'm not just simulating one state machine I'm Sydney like one state machine against many others and this gets into kind of the revenue sharing mechanism right where you say well how do you determine what was like the optimal block how do you determine the sharing well you have to calculate a bunch of counterfactuals and that can be an expansive set depending on like how many Roll-Ups you have using a given shared sequencer cool all right so thanks so much um if the discussion leads from each group could maybe just send me some notes this is what you discussed and I'll share them with everyone in case

that was too fast um but yeah thanks everyone this is really fun and hopefully you got convinced that there's some interesting problems here and if you're continuing to think about it and then you know please uh reach out to anyone here who's everyone who came in terms of a guest I think is very interested in this problem space so hopefully this can be the start of a longer discussion that we all end up having great okay thanks so much yeah [Applause]