Fair transaction ordering and privacy-preserving verifiable computation in decentralized systems

Abstract

Decentralized systems promote the distribution of control between numerous independent parties instead of a single entity, achieving openness and censorship resistance. However, the positive attributes of decentralized systems may not hold when powerful participants are allowed to influence the order of transactions or off-chain computation trust. Fair transaction ordering aims to promote the ordering of transactions via transparent and rule-based mechanisms rather than being manipulated by validators. Maximal Extractable Value (MEV) is a significant challenge to the achievement of fair transaction ordering in blockchains including Ethereum Layer 1 and Layer 2 chains. We first present a unifying overview of MEV on Ethereum and any L1 and L2 chains, discussing mitigation strategies along the dimensions of sequencing, cryptography, and application-layer solutions, and their respective limitations and practical constraints. Following this framework, we examine ordering fairness on single-sequencer-based rollups with First-Come-First-Served (FCFS) ordering. We develop a new ordering mechanism that achieves fairness granularity, minimizing the benefits of latency-based front-running and spam, and introduce a statistical approach to calculate the interval value from network parameters. Experiments on Arbitrum, as a L2 chain system, show that our ordering follows the optimal ordering while minimizing the benefits of front running compared to the standard FCFS approach. Next, we examine the problem of MEV from premature information disclosure on the ERC4337 Account Abstraction, where the user operations reveal their intent to bundlers before the transaction is included. To address the problem of selective inclusion and adversarial ordering, we develop Blind Ordering for ERC4337, a commit-reveal protocol that conceals the contents of the operations during ordering, while cryptographically binding the operations in a bundle to a fixed order that can be verified on-chain. A prototype implementation shows low gas costs while achieving a substantial reduction in intent-based MEV through plaintext disclosure. Furthermore, privacy-preserving verifiable computation seeks to resolve the trade-off between privacy and verifiability by allowing users to verify that computations were carried out correctly without leaking sensitive user data or model parameters. This is typically achieved by means of recursive zero-knowledge proofs, to efficiently combine multiple proofs into a single, compact, and easily verifiable proof, in a decentralized machine learning process. Thus, we introduce RzkFL, an end-to-end solution that combines federated learning with recursive zero knowledge proofs for verifiable inference computation on private inputs enabling efficient verification on the blockchain. Collectively, the above contributions provide practical solutions for fair transaction ordering and privacy-preserving verifiable computation in decentralized systems improving decentralization by limiting order manipulation and also preserving user privacy and verifiable correctness, to improve the overall fairness, security, and efficiency in these systems.

Date	7 Apr 2026
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Kaiwen Zhang (Supervisor)