The Silhouette of Resilience

Data Always - March 23, 2024

tl;dr

By examining the distribution of MEV block values we footprint min-bid usage across major proposer sets. We combine this footprinting with Tornado Cash transaction data and find that min-bid contributes at least 7% of all censorship resistance, while the majority (79%) comes from non-censoring MEV-Boost builders. Among these builders, Titan builder stands out, facilitating the inclusion of 56% of all sanctioned transactions on the network.

Assumed Audience: Ethereum researchers and validators. You are aware of the current state of censorship on Ethereum and know what the min-bid parameter in MEV-Boost does.

Data Sources:

• MEV block value data from dataalways/mevboost-data on GitHub
• Top-level entities from Hildobby's proposer pubkey set on Dune
• Lido node operator pubkeys from the pubkey_mapping dataset on mevboost.pics
• Tornado Cash data forked from coldfire on Dune and then merged with decoded extra data to identify block builders.

Conflicts of Interest: This research was entirely unfunded. The authors have exposure to ether, but declare no other meaningful conflicts of interest.

---

Implemented at the height of relay censorship after The Merge, the min-bid parameter allows proposers to opt out of MEV-Boost blocks and revert to local block building when the value of a block is below a custom threshold. This leaves a distinct footprint in MEV data, which we can use to analyze which proposer sets use the option.

Staking Entities

Starting with a control, we plot the distribution of MEV block values for the second-largest proposer set, Coinbase. Comparing the distribution to the rest of the network, we see that both are log-normally distributed in block value and have complete overlap. This indicates that Coinbase does not use the min-bid parameter across its proposer set.

Coinbase is by no means the exception–almost no large entities appear to leverage the min-bid parameter across their proposer set. The four biggest entities who actively use the parameter are Upbit, Daniel Wang, Octant and KuCoin. In each distribution below, we see a hard cutoff representing the block value at which the proposer set reverts to local building.

Combined, these entities represent 2.5% of staked ether, but their configurations vary dramatically. KuCoin appears to have the strongest setting, building blocks valued at less than 0.05 ETH locally, while Octant has the weakest setting, reverting to local building only for block values smaller than 0.01 ETH.

Lido Node Operators

We observe higher adoption of min-bid among Lido node operators (each controlling about 1% of total stake) compared to the rest of the network. Plotting a time series of the least valuable MEV blocks for each operator, we see clear changes in behavior over time. Recently, Prysmatic Labs and ChainSafe have adopted min-bid, while Sigma Prime discarded their configuration earlier this year.

ChainSafe, Prysmatic Labs, and Stakin all have aggressive min-bid configurations, reverting to local block building when block values are less than 0.07 ETH. Attestant uses a weaker configuration of 0.035 ETH, which is still relatively strong compared to the rest of the network.

Theoretical Impact

We were able to tag 6.5% of proposers, weighted by staked ether, as users of min-bid. This represents a lower bound of min-bid usage. Lido operators account for over half of this total and have the most aggressive configurations.

In total, local building reversions due to min-bid should account for 2.65% of all blocks. Given that 91% of blocks are currently built with MEV-Boost, this suggests that min-bid represents 30% of locally built blocks. As 64% of blocks are currently built by censoring builders, min-bid should be reponsible for about 7% of non-censored blocks.

Empirical Distribution of Censorship Resistance

The largest source and cleanest dataset of censored transactions is Tornado Cash withdrawals, therefore we use them as a proxy for all censored transactions on the network in this analysis. Plotting Tornado Cash transaction counts versus inclusion delays in a log-y scale, we observe a linear relationship that reflects the average rate of censorship of the network.

Source: https://dune.com/queries/3541826/5959146

Although the social layer tends to frame local block building as the censorship-resistant backbone of the community, about 80% of blocks containing Tornado Cash withdrawals are actually built through MEV-Boost.

Source: https://dune.com/queries/3535198/5947751

We see that the majority (70%) of MEV-Boost blocks containing Tornado Cash withdrawals built over the past six months were built by Titan. Jet builder and Penguin builder have built 11% and 6% respectively, while the remaining 13% comes from a diverse set of smaller builders. Taking these rates into account along with PBS permeation, Titan builds 56% of all blocks containing Tornado Cash withdrawals.

Source: https://dune.com/queries/3541759/5958996

Titan is the dominant OFAC non-compliant PBS block builder. In the Tornado Cash withdrawals dataset we see that bloXroute and beaverbuild stopped including these transactions in their blocks in response to the sanctioning of Tornado Cash by the US Department of Treasury in August 2023. Titan's market share of these transactions has been relatively stable since the Summer of 2023.

Source: https://dune.com/queries/3542844/5960978

Isolating the subset of locally built blocks, we find that 33% of those containing Tornado Cash withdrawals are the result of min-bid (6.9% of all blocks). These blocks have top MEV bids at the end of the slot valued lower than the proposer's parameter. An additional 40% of locally built blocks come from untagged proposers, while the final 27% are built locally for indeterminate reasons. The majority of these come from the top ten staking entities by size, suggesting that the primary causes are minor configuration issues, payload timeouts, etc.

Source: https://dune.com/queries/3541805/5959086

Summary

Approximating block building censorship from the distribution of Tornado Cash withdrawals, we find that the network's censorship resistance comes from the following sources:

• 79% of censorship resistance comes from PBS
  • 56% from Titan builder (71% of PBS blocks)
  • 23% from other MEV-Boost builders (29% of PBS blocks)
• 21% of censorship resistance comes from local block building
  • 8% from untagged proposers building blocks locally (40% of local block building)
  • 7% from min-bid usage (33% of local block building)
  • 6% from reversions to local block building for indeterminate reasons (27% of local block building)

We believe that these sources will likely all trend down in time. As the gap closes between traditional and decentralized finance, non-atomic MEV will grow and cause beaverbuild, rsync-builder, and beelder to build a greater share of blocks (all of whom censor). Further, as MEV strategies continue to evolve and Ethereum grows in popularity, block values should trend up causing a decrease in min-bid reversions for any given configuration. Finally, new decentralized entrants to the staking ecosystem are rare, with most capital flowing into already established enties.

The proposed solution to censorship resistance is to add complexity to the protocol by adding support for inclusion lists. However, a simpler and faster solution could be to change the MEV-Boost default to make min-bid opt-out. If the common belief is that defaults are sticky and enabling inclusion lists by default will cause adoption to stay high, then we should be able to extend that logic to min-bid.

Open Questions

• How often do OFAC non-compliant transactions make a difference and cause non-censoring builders to win blocks?
• Empirically, how expensive are min-bid reversions? How do final block values for these locally built blocks compare to MEV bid values?
• Is there support for changing the min-bid default in MEV-Boost to make censorship resistance opt-out instead of opt-in? What value should defaultRelayMinBidEth be set to?