Sealed execution auction

[aelowsson]

Thanks for the good feedback and references.

@quintuskilbourn

1. I realize the architecture you propose doesn’t need to be tied to the Vickrey auction, but since you mention it, I would point out that the incentives for the Vickrey auction aren’t that clean cut in this setting as argued in this paper . The problem is basically that you can force people to bid higher by placing a bunch of sealed bids and then revealing them conditionally on what everyone else reveals (e.g. so that you don’t win but force them to pay more or maybe because you want to deal with adverse selection in a certain way)

It seems that the linked paper deals with an auction setting where the auctioneer derives the proceeds of the auction. However, there is no revenue for the auctioneer in the sealed execution auction (the revenue is burned), strongly limiting the incentives for placing bids that are revealed conditionally in an attempt to take advantage of the Vickrey design. I therefore assume that the outlined concern is not applicable. Correct me if I am wrong.

Note that I am in my answer specifically referring to (my interpretation of) the context in which Ferreira and Weinberg investigate online auctions, and Chitra, Ferreira, and Kulkarni extend this to an open ledger in the link you provided. The concern of proposer—builder collusion outlined in the post instead emerges from value-in-flight, and/or the desire to grief the winning builder among losers, or an ultimatum game.

@quintuskilbourn

2. this design relies on censorship resistance derived from attesters operating based on their local view of (effectively mempool) messages. How realistic is this? Is there research into these kinds of approaches? Eclipse attacks and just shoddy gossip layers could be problematic

The auction design is an alternative to what I will refer to as the English MEV burn auction, which also relies on validators attesting in accordance with their local view of bids. There are then two questions; (a) if Vickrey MEV burn is more viable in this regard than an English MEV burn, and (b) if any of these designs are viable.

(a) A benefit of English MEV burn is that the proposer only needs to get one bid right, picking a bid that has an equal or higher base fee than the highest base fee before the observation deadline. In the Vickrey design, the proposer must include all received bids. In English MEV burn, builders should try to be attentive to the level of the base fee floor as they continue bidding, making many, if not all, of the last incoming bids eligible. There may however be circumstances where the proposer has an incentive to “gamble” on the bid base fee floor, even when not interacting with a colluding bidder.

A downside of English MEV burn is that there probably will be a higher quantity of bids. The quantity of bids in the Vickrey auction should be rather modest, as long as there is a small fee attached for bidding (and each builder only can place one bid). The English MEV burn design does not naturally lend itself to taking out a bid fee or penalizing builders that place more than one bid (although it can be modified).

(b) The P2P layer is not my expertise, and I would welcome feedback on viability. Your and Barnabé’s discussion on FOCIL from yesterday seems relevant (1, 2).

@quintuskilbourn

3. there could be very many bids and its not clear how we could do congestion pricing in this setting since we require attesters to view bids before they are “collated”.

When it comes to the bid fee, there are many designs that are viable. As a basic example, we could rely on a similar mechanism as in the dynamic pricing auction, such that the bid fee increases with the quantity of bids, but still keep the fee rather modest. The bid fee curve could for example be designed to reach 1/50th of the average sales price of the previous auctions at a desirable quantity of bids. The fee curve can peak well below the average sales price even at a large quantity of bids.

I am not perfectly sure which phase you refer to concerning attesters viewing bids before collation, but assuming here you are referring to P2P handling before commitments are collated at time T_2. In this case, my assumption is that attesters could verify the commitment and its signature using the builder’s public key, e.g., \text{Verify}_{pk}(C, \sigma), and only propagate valid commitments. Does it make sense? The point is to ensure that anything propagated across the P2P layer would eventually incur a fee. There is then a further nuance to this around if the proposer might be more likely to see its block rejected in the case where valid bids are spammed. I would actually find it reasonable to still take out the fee for all bids even if the auction fails to conclude, even if it might marginally hurt honest builders, because it would hurt dishonest spamming builders more. Note also that if builders are staked with 32 ETH each, then placing 100 bids requires 3200 ETH (over 10 million dollars worth of capital). Our experience with staked entities in general has been that they do not wish to pursue the outlined activities en masse due to implicit risks. Further note that placing more than one bid can be associated with a penalty much higher than the bid fee.

@quintuskilbourn

Also, if sealed bids pay congestion fees then this could reveal info about the underlying bids, as argued in this paper

The sealed execution auction does not allow the proposer to exclude bids from the auction (if propagated before T_1), and thus does not lead to tips or bribes as described in the linked post. I therefore assume that the outlined concern is not applicable. Correct me if I am wrong. The purpose of SEA is to strip the proposer of autonomy when dealing with timely builders: to restrict its extractable MEV for facilitating the auction. It can be noted that MEV burn in general derives from oversight of attesters. Note further the parallels to multiple concurrent proposers, and its potential viability.