💻 computer science

Towards Decentralized Searcher Competition in MEV Markets

This paper addresses MEV market centralization by proposing a Shapley-capped auction mechanism with Bayesian security constraints to ensure fairer reward distribution and prevent Sybil attacks and validator collusion among heterogeneous searchers.

Original authors: Roozbeh Sarenche, Yunwen Liu

Published 2026-08-12

📖 8 min read🧠 Deep dive

CC BY 4.0

Original authors: Roozbeh Sarenche, Yunwen Liu

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a bustling digital marketplace where invisible racers, known as "searchers," are constantly hunting for tiny, fleeting opportunities to make money. This isn't a normal market; it's the hidden engine of a blockchain, a system designed to be run by everyone, not just a few big bosses. In this world, "Maximal Extractable Value" (MEV) is the prize: it's the extra profit you can squeeze out of a transaction by rearranging the order of events, like a sneaky shopper who buys an item just before a price hike and sells it immediately after. To get these prizes, searchers race to be the first to submit their code to the blockchain's "block builders," who act like the race officials deciding who gets to run the next lap. The current rule of the game is simple but brutal: whoever bids the most money to the builder wins the whole prize, and everyone else gets nothing. This "winner-take-all" system has a nasty side effect: it encourages a few super-fast, super-rich racers to dominate everything, pushing out the smaller players and turning a fair, open race into a closed club.

This paper dives deep into that race to see if we can change the rules to make it fairer without letting cheaters take over. The authors, Roozbeh Sarenche and Yunwen Liu, argue that the current "highest bidder wins" approach is great at stopping one specific type of cheating (where a racer copies their own code to pretend to be many people), but it's terrible at keeping the competition open. They show that when racers have different speeds and skills, the current system lets the fastest few hog all the rewards, leaving everyone else with crumbs. To fix this, they propose a new way to split the prize money called the "Shapley-capped auction." Instead of giving everything to the single winner, this new system looks at how much each racer actually contributed to the group's success and shares the loot accordingly, but with strict safety locks to prevent cheaters from gaming the system. Through computer simulations and looking at real data from the Ethereum blockchain, they suggest that this new method could turn a race dominated by a handful of giants into a lively competition where dozens of racers can actually win.

The Problem: The "King of the Hill" Race

To understand the issue, imagine a video game tournament where the prize is a giant bag of gold coins. In the current version of this game, there are many players, but only the one who pays the biggest "entry fee" to the referee gets the bag. Everyone else pays nothing and gets nothing.

The paper points out that this setup creates a problem called "centralization." In the real world, some players have better computers, faster internet, and more money than others. In a "winner-take-all" race, these advantages compound. If you are slightly faster or have slightly better information, you win almost every time. If you win almost every time, you get rich. If you get rich, you can buy even better computers. Soon, you have a situation where one or two players win 90% of the gold, and the rest of the players are just watching from the sidelines. This is bad for the blockchain because the whole point of these systems is to be decentralized—run by many people, not a few kings.

The authors also looked at a specific type of cheating called a "Sybil attack." Imagine a player trying to rig the game by creating 100 fake accounts, all using the same strategy, just to increase their chances of winning. In the current "highest bidder" system, this doesn't really work because the player still has to pay the full entry fee for every single fake account, and they only get the prize once. So, the current system is actually quite good at stopping this specific kind of cheating. However, the authors argue that being good at stopping cheaters isn't enough if the game is still rigged against the little guys.

The Solution: The "Fair Share" Pot

The authors propose a new way to run the race, which they call the Entry-Filtered Shapley-Capped Auction. Let's break down this fancy name into a story about a group of treasure hunters.

1. The Entry Filter (The Gatekeeper):
First, not everyone gets to play. The system sets a minimum quality bar. If a treasure hunter's map isn't good enough (their code isn't efficient enough), they are politely asked to leave. This keeps the game from getting clogged with garbage submissions and helps stop cheaters from flooding the system with low-quality copies.

2. The Shapley Value (The Fair Splitter):
This is the heart of the new idea. Instead of giving the whole treasure to the person who found the best map, the system looks at how much each person contributed to the group's success. Imagine a group of friends trying to solve a puzzle. If one person brings the box lid, another brings the corner pieces, and a third brings the edge pieces, they all helped. The "Shapley value" is a mathematical way to calculate exactly how much credit each person deserves based on their unique contribution. In this new auction, if multiple searchers find good opportunities, the profit is split among them based on how much they added to the mix, rather than just going to the single "winner."

3. The Cap and the Burn (The Safety Valve):
Here is the tricky part. If the system just splits the money among everyone, a cheater might try to create 1,000 fake accounts to get 1,000 tiny slices of the pie. To stop this, the new system has a "cap." It says, "Okay, we will split the money fairly, but only among the top few best submissions." If too many people try to enter, the system switches to a "fallback" mode where it only pays the very best one and "burns" (destroys) the rest of the money. This makes it a losing strategy to create too many fake accounts.

The "burn" is like a penalty fee. If a cheater tries to split their strategy into many copies to game the system, the math is designed so that the penalty (the burned money) is always bigger than the extra profit they might get. This keeps the game secure against cheaters while still being fair to the honest players.

What the Numbers Say

The authors didn't just dream this up; they tested it. They built a computer model of how these searchers compete and ran thousands of simulations. They also looked at real data from the Ethereum blockchain, analyzing over half a million transactions to see how real searchers behave.

Their findings were clear:

The Old Way (First-Price Auction): In their simulations, when the race became competitive, the "winner-take-all" system became very unfair. In the most concentrated scenarios, they found that effectively only one searcher was winning all the rewards, even though there were many players. The "fairness score" dropped to almost zero.
The New Way (Shapley-Capped): When they applied their new system, the results changed dramatically. In the same concentrated scenarios, the new system allowed 3 to 4 (and sometimes more) searchers to share the rewards fairly. The fairness score jumped up significantly.

For example, in a scenario where the old system gave a fairness score of 0.075 (meaning almost no fairness), the new system boosted it to 0.726 (meaning much more fairness) while still keeping the system safe from cheaters.

They also looked at real-world data from Ethereum. They found two types of treasure hunts:

High-Margin Hunts: These are rare, big prizes. Here, the old system was very concentrated, with a few big players dominating. The new system would have made this much fairer, spreading the rewards to more players.
High-Volume Hunts: These are small, frequent prizes. Here, the old system was already somewhat fair because there were so many players. The new system didn't change the fairness much here, but it did ensure that the system remained secure against cheaters.

The Bottom Line

The paper suggests that the current way of running these blockchain auctions is like a race where the fastest runner gets the whole trophy, and everyone else goes home empty-handed. This encourages a few super-teams to take over. The authors propose a new rulebook where the trophy is shared based on how much everyone helped, but with a strict limit to stop cheaters from faking their way in.

Their work suggests that we can have our cake and eat it too: a system that is both fair (rewarding many different players) and secure (stopping cheaters). It's not a magic wand that fixes everything instantly, but it offers a promising new path to keep the blockchain race open, competitive, and fun for everyone, not just the richest few. The authors emphasize that this is a theoretical and simulated solution that needs to be tested in the real world, but the math and the data so far point in a very hopeful direction.

Technical Summary: Towards Decentralized Searcher Competition in MEV Markets

1. Problem Statement

Maximal Extractable Value (MEV) markets on permissionless blockchains face significant centralization risks. While existing literature has extensively analyzed centralization among block builders and validators, this paper identifies searcher competition as a distinct and critical locus of centralization.

Current MEV extraction relies on a first-price, winner-take-all auction mechanism where searchers bid for block inclusion. The paper argues that this design, combined with inherent searcher heterogeneity (differences in opportunity coverage, execution efficiency, capital, and information), leads to severe reward concentration. In this environment, rewards accrue to the searcher with the highest rank (highest bid or best execution) rather than being distributed proportionally to marginal contributions. This "rank dominance" creates high barriers to entry, reinforces the dominance of specialized infrastructure, and undermines the decentralization goals of permissionless protocols.

The core challenge is designing a mechanism that distributes rewards more broadly and fairly among searchers while strictly maintaining security guarantees in a permissionless environment. Specifically, any mechanism that rewards multiple submissions must prevent:

Sybil Attacks: Searchers splitting a single execution strategy into multiple identities (copied or degraded code) to capture a larger share of rewards.
Validator-Searcher Collusion: Coalitions where validators and searchers coordinate to increase joint payoffs through identity splitting.

2. Methodology

System Model

The authors develop a heterogeneous model of searcher competition involving:

Entities: A finite set of searchers $S$ and one validator.
Opportunity Classes: MEV opportunities are grouped into classes (e.g., arbitrage, liquidation).
Heterogeneity: Searchers differ in:
- Opportunity Coverage: The probability $\alpha_{i,\tau}$ of discovering an opportunity in class $\tau$ .
- Execution Quality: The certified normalized net value $z_{i,o}$ of their execution code, derived from gross value minus execution costs.
Information Structure: A Bayesian framework where searchers know their own code and value but only know the distributional primitives of competitors' participation and values before the submission deadline.

Evaluation Metrics

The paper introduces two primary metrics to evaluate auction mechanisms:

Shapley-weighted Jain Fairness Index ( $J_{Sh}$ ): Measures whether rewards are proportional to searchers' marginal contributions (Shapley values) to the expected surplus. A value of 1 indicates perfect contribution-adjusted fairness; lower values indicate concentration.
Expected-Reward Herfindahl–Hirschman Index (ER-HHI): Measures the concentration of long-run expected rewards among searchers. The reciprocal ( $N_{eff}$ ) represents the "effective number of rewarded searchers."

Security Definitions

The paper defines security against deviations in a Bayesian context (evaluating expected payoffs conditional on a searcher's own participation and value):

Bayesian Copied-Code Sybil Resistance: Prevents a searcher from increasing their expected payoff by submitting multiple copies or degraded versions of their own code.
Bayesian Validator–Searcher Coalition Resistance: Prevents a searcher and validator from increasing their joint expected payoff through copied-code deviations.

3. Key Contributions

A. Analysis of the First-Price Auction (FPA)

The paper establishes the standard FPA as a benchmark.

Security: The FPA is proven to be robust against copied-code Sybil deviations and validator-searcher coalitions in an ex-post sense (and thus Bayesian). Since only the highest bidder wins, splitting a strategy into multiple bids does not increase the probability of winning or the payoff.
Economic Failure: Under heterogeneity, the FPA rewards rank dominance rather than marginal contribution.
- If a small set of "dominant" searchers (high coverage or high execution quality) exists, they almost always win.
- The paper proves that as the participation probability of dominant searchers approaches 1, the effective number of rewarded searchers ( $N_{eff}$ ) converges to the size of the dominant class (potentially 1), and the fairness index ( $J_{Sh}$ ) collapses.
- This leads to a "winner-take-all" outcome where lower-quality searchers are effectively excluded from rewards, reinforcing centralization.

B. Proposed Mechanism: Entry-Filtered Shapley-Capped Auction (SCA)

To address FPA limitations, the authors propose the Entry-Filtered Shapley-Capped Auction.

Mechanism Design:
1. Entry Filter: Submissions with certified net values below a threshold $\theta$ are rejected.
2. Shapley Sharing: If the number of admitted submissions ( $t_\theta$ ) is below a cap $H$ , rewards are distributed according to the Shapley value of the validator-dependent max game, scaled by retention parameters ( $\eta_S, \eta_V$ ).
3. Capped Fallback: If $t_\theta > H$ , the mechanism switches to a conservative branch where only the top submission receives a reward (scaled down), and the rest of the surplus is burned.
Security via Burning: The mechanism uses burn functions (burning a portion of the surplus) to eliminate the economic incentive for Sybil attacks. By carefully calibrating the burn parameters ( $\kappa_S, \kappa_V$ ), the mechanism ensures that the expected payoff of a searcher (or coalition) does not increase by splitting their submission into multiple identities.
Theoretical Guarantees: The paper provides a Bayesian security characterization, proving that with suitable parameter choices, the SCA is resistant to copied-code Sybil deviations and validator-searcher coalitions. It also proves budget feasibility (total payments + burn $\leq$ total value).

C. Empirical Calibration

The authors calibrate their model using Ethereum on-chain arbitrage data (577,264 transactions).

They classify searchers into three strategy profiles: Margin-oriented (high profit per tx), Volume-oriented (high tx count), and Mixed.
They extract truncated log-normal distributions for execution costs and participation probabilities for "Elite," "Middle," and "Tail" tiers within each strategy.
These empirical distributions are used to simulate and compare the FPA and SCA under realistic conditions.

4. Results

Theoretical Benchmarks

FPA Centralization: In scenarios with asymmetric searcher populations (e.g., a small elite class with high coverage/quality), the FPA leads to extreme concentration. In the "Single Elite Searcher" scenario, $N_{eff}$ drops to $\approx 1.04$ and $J_{Sh}$ to $0.075$, indicating near-total monopolization of rewards.
SCA Decentralization: The Entry-Filtered SCA significantly improves decentralization. In the same "Single Elite" scenario, the SCA maintains an $N_{eff}$ of $\approx 3.78$ and a $J_{Sh}$ of $0.726$. It successfully distributes rewards to a broader set of searchers based on their marginal contributions.
Trade-off: There is a security-coverage tradeoff. Lower entry thresholds ( $\theta$ ) admit more searchers (improving decentralization) but require higher burn rates ( $\kappa_S$ ) to maintain Sybil resistance.

Empirical Findings

Concentrated Opportunities (C1 - Margin-oriented): In opportunity classes where competition is naturally concentrated (few elite contracts dominate), the FPA performs poorly ( $J_{Sh} = 0.175$ ). The SCA dramatically improves fairness ( $J_{Sh} = 0.983$ ) and increases the effective number of rewarded searchers.
Diffuse Opportunities (C2 - Volume-oriented): In highly competitive, volume-driven markets, the FPA is already relatively decentralized ( $J_{Sh} = 0.997$ ). The SCA offers marginal improvements in fairness but ensures that reward distribution remains broad even under strict Sybil resistance constraints.
Security Validation: In all calibrated scenarios, the binding deviation for security is a two-copy exact-copy attack at the admission boundary. The proposed parameters successfully deter this deviation.

5. Significance and Claims

The paper claims that decentralization in MEV markets is fundamentally a mechanism design problem, not just an issue of infrastructure or order-flow access.

Limitation of Current Design: The authors demonstrate that the standard first-price, winner-take-all auction, while secure against Sybil attacks, is economically centralizing under realistic heterogeneity. It rewards "rank" rather than "contribution," allowing dominant actors to capture the vast majority of MEV.
Feasibility of Alternatives: The paper argues that it is possible to design a mechanism that simultaneously achieves:
1. Contribution-Adjusted Fairness: Rewards proportional to Shapley values.
2. Reward Decentralization: A broader distribution of rewards among searchers.
3. Permissionless Security: Robustness against Sybil attacks and validator collusion without requiring trusted third parties.
The Role of Burning: The introduction of burning (discarding surplus) is presented as a necessary trade-off to align incentives. By burning a portion of the surplus, the mechanism removes the profit motive for identity splitting, enabling fairer distribution without compromising security.

The authors conclude that the Entry-Filtered Shapley-Capped Auction offers a viable path toward more decentralized MEV markets, particularly in opportunity classes where current mechanisms lead to monopolistic reward concentration. They emphasize that the mechanism's parameters (thresholds, caps, burn rates) must be carefully calibrated to balance the trade-off between broad participation and security against strategic deviations.