SoK: The Evolution of Maximal Extractable Value, From Miners to Cross-Chain

Source: arXiv:2603.07716 · Published 2026-03-08 · By Davide Mancino, Hasret Ozan Sevim

TL;DR

This Systematization of Knowledge (SoK) traces the conceptual and empirical evolution of Maximal Extractable Value (MEV) in blockchain systems through three distinct chronological eras from 2014 to early 2026. The paper organizes fragmented MEV research and infrastructure developments into Era I (2014–2020), characterized by the birth of Miner Extractable Value (MEV) in Proof-of-Work systems with public mempools and gas auctions; Era II (2020–2024), marked by the generalization to Maximal Extractable Value across PoW and PoS systems including the Ethereum Merge, Proposer-Builder Separation (PBS) with Flashbots infrastructure, and non-atomic cross-venue extraction; and Era III (2024–present), focusing on emerging cross-chain MEV involving coordination across Layer-1, Layer-2 rollups, bridges, and centralized exchanges. The paper synthesizes taxonomies, formal definitions, empirical measurements, economic models, and mitigations from the literature, concluding with a research agenda prioritizing standardized MEV metrics, detection benchmarks, and cross-chain infrastructure security. Empirically, MEV extraction has grown from millions to hundreds of millions in USD with increasing ecosystem complexity and infrastructural sophistication. The work clarifies key distinctions such as potential versus realized MEV, single-domain versus cross-domain extraction, and the evolving supply chain from miners to proposers, builders, relays, and searchers.

Key findings

• Miner Extractable Value (MEV) first formally quantified by Daian et al. showed over $6 million in atomic arbitrages on Ethereum before 2020 (Era I).
• Flashbots infrastructure deployment (2021) captured 48% of sandwich attacks via private channels within months, with mining pools controlling 97-99.9% of hash rate adopting mev-geth.
• Proposer-Builder Separation (PBS) launched post-Ethereum Merge (September 2022) led to rapid PBS block adoption (>85%), concentrating over 85% of block building on just two builders by early 2024.
• Cross-domain MEV arbitrage identified by Oz et al. (January 2025) with 242,535 cross-chain arbitrages generating at least $8.65 million USD profit.
• Wu et al. (2023–2025) document $233.8 million USD extracted via CEX-DEX non-atomic arbitrage on Ethereum, indicating significant off-chain MEV sources.
• Torres et al.'s Layer-2 rollup analysis (April 2024) initiated Era III, showing early complexities in multi-chain sequencing, latency, and trust assumptions for MEV extraction.
• Game-theoretic models (Capponi et al.) indicate MEV induces inefficient resource allocation with partial welfare improvements through private pools but equilibrium adoption barriers persist.
• Empirical studies report distinct negative externalities of MEV including block congestion from spam, elevated transaction fees, price manipulation on DEX trades, and threats to consensus security like time-bandit attacks.

Threat model

The adversary is any entity controlling transaction ordering within a blockchain domain such as miners in PoW, validators in PoS, sequencers in Layer-2 rollups, or relayers in bridges. They can observe pending transactions, reorder, delay, insert, or censor transactions to maximize profit through front-running, back-running, sandwich attacks, arbitrage, or reorganization (time-bandit) attacks. In cross-domain MEV, adversaries can coordinate across multiple chains or domains to unlock additional value. The adversary cannot rewrite or censor transactions outside their sequencing domain without collusion, nor typically control 100% of the ordering power across all domains simultaneously.

Methodology — deep read

The SoK is a comprehensive literature synthesis and conceptual analysis rather than a novel empirical study, but it rigorously reviews and connects multiple empirical, theoretical, and infrastructural studies on MEV. 1. Threat Model & Assumptions: MEV involves adversaries who control transaction ordering within blockchain domains (miners in PoW, validators in PoS, sequencers in rollups, relayers in bridges, or centralized exchanges). These sequencers observe pending transactions and reorder, insert, or censor transactions to extract profits from arbitrage, front-running, sandwich attacks, liquidation triggers, or reorganizations (time-bandit attacks). Cross-domain MEV involves coordination or collusion among multiple such ordering authorities across heterogeneous blockchain domains. Adversaries lack unlimited control—e.g., do not control all hash power or sequencing across all domains simultaneously—but do have privileged ordering power and private communication channels. 2. Data: The SoK references multiple datasets such as Ethereum mempool and block data pre- and post-Merge, Flashbots bundle records (blocks 10,000,000 to 14,444,725), Uniswap V2 and V3 trading cycles (>292,000 arbitrage cycles), CEX-DEX arbitrage detection data spanning 19 months (Aug 2023–Mar 2025), and cross-chain arbitrage event logs (242,535 trades). Data provenance typically involves on-chain traces, mempool captures, Flashbots private relay data, and decentralized exchange trade logs. 3. Architecture / Algorithm: Various MEV extraction strategies are analyzed including primitive ordering manipulations (front-running, back-running), composed strategies (sandwich attacks, JIT liquidity provision), and reorganization-based extraction (time-bandit attacks). Formal models incorporate game theory, economic equilibria, and mechanized proofs in tools like Lean theorem prover to bound extractable value and prove optimal attack patterns. The Proposer-Builder Separation architecture is described as a layered supply chain where searchers submit bundles to builders, builders create blocks, relays broker blocks to proposers, and proposers select and publish blocks, changing the incentives and concentration dynamics. 4. Training Regime: Not applicable as the SoK is a meta-analysis not involving model training. 5. Evaluation Protocol: The SoK surveys existing papers’ metrics including total MEV profit extracted (USD), number of arbitrage cycles detected, percent of blocks using PBS, concentraton metrics of builders or miners, latency and revenue per block, and measurement of replay and front-running strategies. It notes the divergence between potential MEV (theoretical maxima from optimal orderings) and realized MEV (actual extracted profits), highlighting varying heuristic accuracy and detection challenges especially post-Merge with private transaction submission. 6. Reproducibility: Most referenced works release code or datasets openly, e.g., Flashbots publish mev-geth client code and empirical dashboards, but cross-chain MEV remains nascent with limited standardized benchmarks or reproducible datasets. The SoK calls for standardized measurement pipelines and detection benchmarks to improve reproducibility of MEV research.

Technical innovations

• Chronological framing of MEV development across three distinct eras (2014–2020, 2020–2024, 2024–present) clarifies causal evolution of concepts and infrastructure.
• Unified taxonomy distinguishing potential vs. realized MEV and single-domain vs. cross-domain MEV provides conceptual clarity beyond prior fragmented definitions.
• Formalization of cross-domain MEV with domains, sequencers, and joint ordering maximization extends prior single-domain MEV models (e.g., Babel et al.) to multi-chain ecosystems.
• Detailed mapping of the Ethereum block production supply chain evolution—from public mempools and miners in Era I to Proposer-Builder Separation and private bundle submission in Era II—reveals architectural and economic shifts.
• Synthesis of empirical MEV measurement challenges and negative externalities highlights tradeoffs in fairness, security, and efficiency remaining underexplored in the literature.

Datasets

• Ethereum mempool and block data (2014–2025) — public on-chain data
• Flashbots bundle records (blocks 10,000,000 to 14,444,725) — publicly accessible via Flashbots
• Uniswap V2 cycle arbitrage dataset (292,606 cycles) — extracted from public DEX data
• CEX-DEX arbitrage data (Aug 2023–Mar 2025, $233.8M extracted) — derived from public exchange logs
• Cross-chain arbitrage events (242,535 trades) — analyzed by Öz et al., dataset not publicly available

Baselines vs proposed

• Traditional public mempool miner ordering (Era I): ~$6 million atomic arbitrage (lower bound) vs Flashbots private submission (Era II): 48% of sandwich attacks via private relays, leading to greater extraction centralization.
• Post-Merge validator block proposing without PBS: baseline ordering vs PBS-enabled block production: >85% of blocks PBS within two months, >85% blocks built by 2 major builders indicating high market concentration.
• Single-domain MEV extraction measurement vs cross-domain MEV extraction: Cross-chain arbitrage events generated at least $8.65 million USD profit, indicating value beyond any single chain.
• Empirical DEX volume (132 billion USD) traded with non-atomic arbitrage (~30% of volume) vs atomic arbitrage only: non-atomic MEV substantially larger in realized value.

Limitations

• Cross-chain MEV remains underexplored with sparse empirical measurement and lacks standardized detection methodologies.
• Many MEV detection pipelines rely on heuristics, leading to variance and potential under-/over-counting of realized MEV, especially post-Merge with private mempools.
• Empirical studies often focus on Ethereum and EVM-compatible chains, limiting generalization to other blockchain architectures such as Algorand or ZK rollups.
• Concentration analyses of builders and relays highlight centralization but lack detailed causal attribution or impact on censorship and fairness.
• Limited consideration of adversarial adaptive strategies on emerging PBS or cross-domain infrastructure beyond static analyses.
• Time-bandit and chain reorganization attacks remain theoretically concerning but empirically unobserved or rare.

Open questions / follow-ons

• How to design robust cross-chain MEV detection pipelines that can handle heterogeneous consensus protocols, latencies, and partial observability?
• What formal guarantees or incentive mechanisms can mitigate MEV-induced centralization pressures at the proposer-builder-relay level?
• How to accurately quantify the economic and security impact of non-atomic, multi-block, and off-chain MEV strategies across evolving DeFi architectures?
• What are effective defense architectures for MEV in multi-domain environments, balancing transparency, fairness, and censorship resistance?

Why it matters for bot defense

For bot-defense and CAPTCHA practitioners, this SoK highlights the evolving complexity of adversarial strategies around transaction ordering and multi-domain coordination in blockchain systems. Understanding MEV is crucial because transaction ordering manipulation resembles a strategic corruption of fair sequencing which parallels sophisticated bot interference in web requests. The taxonomy and empirical insights into negative externalities like user cost inflation, congestion, and consensus instability reveal attack impact that may analogously affect decentralized oracles or authentication pipelines. As extractors evolve from single-chain focus to cross-chain mechanisms, defenses must consider collusion and composite pipelines, motivating layered detection and mitigation strategies. Furthermore, the architectural evolution from public, transparent mempools to private submission channels and multi-party auctions suggests that detection must incorporate off-chain, privileged communication channels beyond simple on-chain monitoring. Overall, integrating MEV-aware adversary models into bot-defense frameworks can improve robustness against transaction manipulation and multi-domain exploit chaining.

Cite

@article{arxiv2603_07716,
  title={ SoK: The Evolution of Maximal Extractable Value, From Miners to Cross-Chain },
  author={ Davide Mancino and Hasret Ozan Sevim },
  journal={arXiv preprint arXiv:2603.07716},
  year={ 2026 },
  url={https://arxiv.org/abs/2603.07716}
}

Read the full paper

• arXiv:2603.07716 (abstract)
• arXiv:2603.07716 (PDF)