Cardano Leios Models Find MEV Exposure in Just 2.6% of Analyzed Volume

Input Output Research has presented new findings on the security and performance modeling behind Ouroboros Leios, Cardano’s proposed Layer 1 scaling protocol. Research fellow Giorgos Panagiotakos shared the results during a Cardano R&D session on July 16, 2026.

The presentation formed part of the midyear update for WP1.1 Consensus, a research package covering Leios for higher throughput and Ouroboros Peras for faster transaction settlement. The work includes network modeling under load, formal security analysis, voting certificate optimization and parameter selection for engineering implementation.

Leios Models Find Limited MEV Exposure Under Load

The research team analyzed how the higher transaction capacity planned for Leios could affect mempool fragmentation, transaction ordering and front-running opportunities.

MEV, or maximal extractable value, refers to value that can be captured by changing the order, inclusion or exclusion of transactions before confirmation. Front-running is one form of MEV in which an attacker observes a pending transaction and attempts to place a competing transaction ahead of it.

According to Panagiotakos’ presentation, approximately 2.6% of the volume examined in the research was potentially exploitable through MEV. Front-running was generally unprofitable in the tested scenarios, with the result linked to Cardano’s EUTXO architecture and the way Leios distributes transaction processing.

The estimate applies to the models and conditions included in the research. The public Post-CIP R&D findings support the broader conclusion that only a fraction of transactions are susceptible to front-running, although they do not include the dataset or calculation behind the 2.6% figure.

The public findings also state that the size of the attack surface depends on individual decentralized application design. A mempool-based attack would require control over relay infrastructure and the ability to construct a competing transaction quickly, while a block producer’s probability of success would be proportional to its stake.

Leios mempool simulations found that Cardano’s highly connected network topology supports rapid transaction diffusion and limits fragmentation. When demand remains within network capacity, the models report global mempool synchronization above 90% and fragmentation below 10%. Under heavier demand, diffusion can slow and fragmentation increases in proportion to the load.

Markov and Resource Models Guide Leios Parameters

Ouroboros Leios is designed to increase Cardano’s throughput by allowing ranking blocks to include certified references to endorser blocks, which represent larger batches of transactions. Committees vote on those blocks before their certificates can be included in the ledger.

The research includes a Markov model that estimates the probability of endorser block certification as ranking blocks are produced. It tracks block production, certificate availability and whether sufficient votes are collected before the next relevant block-forging opportunity.

These results help researchers evaluate Leios efficiency and select protocol parameters based on modeled network behavior.

A separate constraint model measures processor and bandwidth requirements. According to the published findings, waiting for transaction bodies and applying transactions to the ledger are the two largest sources of delay. Signature verification and Plutus execution are less restrictive because much of that work can be performed in parallel.

The models indicate that three CPU cores are sufficient for a typical 12 MB endorser block. An attacker intentionally releasing transaction bodies late could double peak processor demand.

Those results inform limits including the maximum number of transactions per endorser block, maximum block size and Plutus execution limits. The targeted throughput and hardware requirements can therefore be evaluated against measured resource behavior before production parameters are selected.

The security work also includes modeling the timely diffusion of endorser blocks under adversarial conditions, formal proofs of safety and liveness properties in Agda and a trace verifier. The verifier compares implementation behavior with the formal protocol specification to identify deviations in generated blocks, certificates and ledger states.

The official Leios repository contains the formal specification, simulations, network visualizers, cost analysis and trace verification tools. The project now combines protocol research with active design and prototyping work for future Cardano node implementations.

New Voting Certificates Connect Leios and Peras

Input Output Research is also developing an optimized voting certificate design intended for both Leios and Peras. The objectives are to reduce certificate size and voting traffic while improving resistance to stronger adversarial models.

In Leios, certificates confirm that endorser blocks received the required committee support. In Peras, voting certificates support block boosting, a mechanism that gives additional weight to a sufficiently endorsed chain and accelerates settlement.

The design builds on earlier Input Output Research work, including Mithril’s stake-based threshold multisignatures and the Fait Accompli approach to committee selection. A research paper describing the optimized certificates is scheduled for completion by the end of 2026.

The second-half research plan also includes a technical report on components that could be shared between Leios and Peras, along with a separate analysis of timely endorser block diffusion under attack. Andre Knispel, Brian Bush, Matthias Fitzi, William Wolff, Yves Hauser, David Rosales and Sandro Coretti-Drayton were among the researchers identified as contributors.

The three scheduled outputs will provide engineering teams with a proposed certificate format, documented links between Leios and Peras, and a security model for endorser block diffusion. Together, they will define several of the parameters and adversarial assumptions needed for the next stage of both protocol implementations.