Skip to content
Crypto.world

Research

MAD-DAG: Protecting Blockchain Consensus from MEV

Selfish mining lets a miner withhold blocks, release them at the right moment, and earn more than their fair share — and the attack gets much worse when the miner has a network-propagation edge…

MAD-DAG: Protecting Blockchain Consensus from MEV
arXiv
2511.21552
Published
2025-11-26
Authors
Roi Bar-Zur, Aviv Tamar, Ittay Eyal

AI summary

MAD-DAG: Protecting Blockchain Consensus from MEV — Summary

Overview

Research area: Blockchain consensus security — specifically proof-of-work (PoW) protocol design and the economics of selfish mining.

Technical level: Advanced. The paper combines DAG-based consensus protocol design with Markov Decision Process (MDP) modeling and dynamic programming; readers need familiarity with PoW, forks, and basic game theory to follow the analysis.

Scope: The paper designs and analyzes MAD-DAG, a PoW DAG protocol whose novel "MAD" ledger function discards the contents of equal-length competing chains, and evaluates it against Colordag and Nakamoto Consensus (Bitcoin) under rushing, block-reward variability (MEV/fees), and petty-compliant miners.

What This Paper Is About

Selfish mining lets a miner withhold blocks, release them at the right moment, and earn more than their fair share — and the attack gets much worse when the miner has a network-propagation edge ("rushing"), when block rewards vary because of fees or MEV, or when other miners can be implicitly bribed to mine on the attacker's chain ("petty-compliant" miners). The state-of-the-art defense, Colordag, was only proven secure under idealized conditions and an impractically high latency parameter. The goal is a practical-latency protocol that keeps selfish mining unprofitable even under all of these adverse conditions.

Key Contributions

  1. MAD-DAG protocol and the MAD ledger function. A PoW DAG protocol whose ledger function marks a block "destructed" when there exist two chains of equal length to the canonical chain, one containing the block and one not — so the contents (and therefore fees and MEV) of both competing chains are discarded. This removes any gain from rushing or from bribing petty-compliant miners into a tie.
  2. MDP models of selfish mining in Colordag and MAD-DAG. The first tractable model of selfish mining in a practical DAG-based blockchain, solved with the Probabilistic Termination Optimization (PTO) method using dynamic programming, yielding a lower bound on the security threshold (unlike deep-RL approaches, which give upper bounds).
  3. Conservative ("upper-bound") reward rules that favor the selfish miner. A modified subsidy/acceptability mechanism that collapses many states into equivalent ones, making the model tractable; this upper-bounds attacker revenue and therefore lower-bounds the security threshold.
  4. First analysis of Colordag at practical latency and under adverse conditions, showing MAD-DAG is more secure than Colordag and Bitcoin while otherwise maintaining comparable security.

Main Findings

  • Rushing + practical latency: MAD-DAG's difficulty-adjustment mechanism raises the security threshold from 25% in NC and 34% in Colordag to 39% in MAD-DAG, with a rushing adversary and fork sensitivity N_ℓ = 25.
  • Low subsidy, high fees: When the subsidy is significantly lower than transaction fees, Colordag's security threshold drops significantly below NC's, while MAD-DAG's drops but remains higher than NC's.
  • Petty-compliant miners and reward variability: With petty-compliant miners and high block-reward variability, MAD-DAG's security threshold ranges from 11% to 31%, whereas both Colordag and Bitcoin achieve 0% for all levels.
  • Colordag collapses under mild variability: With petty-compliant miners present, Colordag's security threshold drops to 0% at a mild level of block reward variability, while MAD-DAG's is unaffected by petty-compliant miners and stays above 11% even with extreme reward variability.
  • Incentive compatibility for petty compliance: If a selfish miner tries to bribe a petty-compliant miner onto an equal-length chain, the petty-compliant miner can point to both chains, destruct their contents, and then include those transactions in their own block — removing the bribe's value.
  • Why MEV needs more than fee-burning: Burning transaction fees while keeping the transactions valid would suffice for fees, but MEV is invisible to the consensus protocol, motivating the stronger destruction of contents.
  • Tractability: The full model's state space is roughly 100,000,000 states at chain-length cap L = 5, which is why the conservative upper-bound model is required.

Methodology in Plain English

  • Protocol design. MAD-DAG keeps a DAG of blocks (each block can point to multiple previous blocks). It borrows Colordag's subsidy rule — only uncontested blocks on the canonical (longest) chain are rewarded, where a block is uncontested if no other acceptable block sits at the same height, and acceptable means within N_ℓ blocks (symmetric difference) of the canonical chain. It adds its own ledger function: any block that appears in one of two equal-length top chains is destroyed along with its contents. Its difficulty adjusts after a fixed number of blocks on the longest chain, exactly as in Nakamoto Consensus.
  • Modeling the attacker. The miner with fraction α of hashing power is modeled as a single rational agent in a Markov Decision Process. The state tracks the attacker's blocks, the public chain, whether rushing is currently relevant, how many attacker blocks were revealed, the index of the attacker's last canonical block, and the pool of pending whale transactions. Actions are Mine, Reveal, and Merge.
  • Objective. The miner maximizes the expected ratio of total rewards (subsidy + guaranteed fees f + whale fees F) to total difficulty contribution — a non-standard utility that PTO converts into a stochastic shortest-path MDP solvable by dynamic programming.
  • Making it solvable. Because the honest model's state space explodes, the authors define a variant that always favors the attacker (attacker blocks always "acceptable," attacker may destroy the public chain without revealing an equal-length chain, attacker may include whale transactions retroactively). A larger attacker revenue means a smaller computed threshold — so results are safe lower bounds on security.
  • Mode switches. Different combinations of tie-breaking (first-heard, random, worst-case), difficulty adjustment (uncontested = Colordag, canonical = MAD-DAG), and ledger function (canonical vs. MAD) instantiate either protocol, allowing direct comparison.
  • Sanity check. The authors verify their closed-form honest-mining revenue against a simulation of honest behavior.

Why This Matters

Impact on research: The paper provides the first tractable optimal-strategy analysis of selfish mining in a practical DAG-based blockchain, and the first security analysis of Colordag at realistic latency. It reframes ledger-content selection — not just block ordering or subsidy denial — as a consensus-level defense against MEV-driven attacks, and it supplies a reusable MDP-plus-upper-bound technique for analyzing other DAG protocols.

Real-world applications:

  • Bitcoin and Bitcoin-like PoW chains facing declining subsidies (which halve every 4 years) and growing fee/MEV dependence.
  • Chains where large "whale" transactions create highly variable per-block rewards.
  • Networks with well-connected miners using relay networks, where propagation advantages ("rushing") are real.
  • MEV-heavy environments where transaction ordering/insertion profits are not visible to consensus and therefore cannot be fixed by fee smoothing or fee burning.

Industry relevance: Mining pools, protocol designers, and clients evaluating consensus changes get a concrete design that resists bribery of rational miners on ties. The paper also notes that projects like BitVM aiming to bring smart contracts to Bitcoin would further increase reward variability, making this defense more relevant.

Future Directions

  • Network delay and benign forks. The model assumes zero network delay except for rushing and analyzes only the one-color version of Colordag, leaving the multicolor mechanism and non-zero-delay effects unexamined.
  • Combining MAD-DAG with fee-smoothing or alternative fee mechanisms, which the paper analyzes separately rather than jointly and which cannot address MEV on their own.
  • Extending the MDP abstraction beyond a single selfish miner — e.g., multiple strategic miners, or miners who are simultaneously petty-compliant and selfish — remains open.
  • Deployment and empirical validation. The paper is an analytical/model-based study; real-world measurement of MEV-driven reward variability and of the practical consequences of destructing equal-length chain contents is a natural next step.

Target Audience

Protocol researchers and designers working on PoW consensus and MEV resistance; security analysts modeling miner incentives; and graduate students or engineers with a background in Markov Decision Processes and blockchain fundamentals who want to understand how ledger-level rules — rather than just reward rules — can neutralize selfish mining. Readers without that background can still follow the qualitative argument, but the quantitative results require it.

Authors’ abstract

Blockchain security is threatened by selfish mining, where a miner (operator) deviates from the protocol to increase their revenue. Selfish mining is exacerbated by adverse conditions: rushing (network propagation advantage for the selfish miner), varying block rewards due to block contents, called miner extractable value (MEV), and petty-compliant miners who accept bribes from the selfish miner. The state-of-the-art selfish-mining-resistant blockchain protocol, Colordag,…

The full abstract on arXiv
Read the original paper