RESEARCH

The research record is the methodology.

We do not cite academic credentials as background decoration. The papers listed here are the direct intellectual foundation for how we analyse smart contracts and ZK programs. Solidifier is how we think about EVM state reachability, and NAVe is how we reason about Noir programs.

Formal analysis of Ethereum smart contracts

Solidifier: Bounded Model Checking Solidity Using Lazy Contract Deployment and Precise Memory Modelling

SAC 2021 (ACM Symposium on Applied Computing), pp. 1788–1797 · 2021

Pedro Antonino, A. W. Roscoe

Presents Solidifier, a bounded model checker for Solidity/Ethereum that translates a desugared Solidity representation into Boogie and verifies it with the Corral model checker. Introduces lazy contract deployment and precise memory modelling to over-approximate Solidity/Ethereum semantics without targeting specific vulnerability patterns.

Specification is Law: Safe Creation and Upgrade of Ethereum Smart Contracts

SEFM 2022 (Software Engineering and Formal Methods), pp. 227–243, Springer · 2022

Pedro Antonino, Juliandson Ferreira, Augusto Sampaio, A. W. Roscoe

Introduces a 'specification is law' paradigm for Ethereum smart contracts, combining auditing tools with upgradeability patterns to allow contract code to be created and upgraded provided it always satisfies a fixed formal specification, addressing the catastrophic risks of faulty immutable contracts.

A Refinement-Based Approach to Safe Smart Contract Deployment and Evolution

Software and Systems Modeling, pp. 1–37, Springer · 2024

Pedro Antonino, Juliandson Ferreira, Augusto Sampaio, A. W. Roscoe, Filipe Arruda

Extends the 'specification is law' framework to allow smart contract specifications themselves to evolve, provided a formal refinement notion is preserved. Introduces specification refinement tailored for contracts (including data representation changes and interface extension) centred around a trusted off-chain deployer. Applied to ERC20, ERC3156, ERC1155, and ERC721 Ethereum standards.

Extracting Formal Smart-Contract Specifications from Natural Language with LLMs

FACS 2024 — Formal Aspects of Component Software (LNCS 15189), pp. 109–126, Springer · 2024

Gabriel Leite, Filipe Arruda, Pedro Antonino, Augusto Sampaio, A. W. Roscoe

Implements DbC-GPT, a framework that uses LLMs (primarily ChatGPT) to automatically infer formal design-by-contract specifications (invariants, pre- and postconditions), expressed in the solc-verify notation, for Solidity smart contract functions from natural-language descriptions.

Generating Formal Smart-Contract Specifications: Comparing Few-Shot Learning and Fine-Tuned LLMs

Science of Computer Programming, vol. 252, article 103464, Elsevier · July 2026

Gabriel Leite, Filipe Arruda, Pedro Antonino, Augusto Sampaio, A. W. Roscoe

Extended journal version of the FACS 2024 paper. Extends DbC-GPT to use fine-tuned large-language models and conducts a systematic comparison of few-shot prompting against fine-tuned LLM approaches, evaluating generated specifications for syntactic and semantic correctness against Ethereum standards.

Formal verification of zero-knowledge programs

NAVe: Formally Verifying Noir Zero Knowledge Programs

arXiv · January 2026

Pedro Antonino, Namrata Jain

Formalises ACIR (Noir's intermediate representation) using SMT-LIB and the theory of finite fields, then verifies programs using the cvc5 SMT solver. Open-source.

A Soundness Bug in cvc5's Finite Field Split Solver

March 2026 · The Blockhouse Technology Ltd

How a misplaced overflow check in cvc5's split solver could invalidate the formal verification of zero-knowledge programs, discovered during NAVe testing and fixed in PR #12457.

The wider research foundation

The audit practice sits within a broader body of work at TBTL spanning distributed systems, applied cryptography, and trusted execution environments.

Blockchain

Hierarchical Consensus: Scalability Through Optimism and Weak Liveness

DISC 2025 — 39th International Symposium on Distributed Computing (LIPIcs, vol. 356), pp. 6:1–6:20 · 2025

Pedro Antonino, Antoine Durand, A. W. Roscoe

Introduces a hierarchical consensus protocol that achieves scalability by being optimistic, operating fast when conditions are favourable, while relaxing liveness to a weak form under adversarial conditions. Formalises handover and justifiability properties and proves safety and liveness for blockchain and distributed systems settings.

Scaling committee-based consensus

April 2026 · The Blockhouse Technology Ltd

How to safely reach agreement with small committees, an accessible overview of the DISC 2025 research on justifiability and multi-threshold BFT consensus.

Hooks: A Simple and Modular Checkpointing Protocol for Blockchains

IEEE NCA 2024 (23rd IEEE International Symposium on Network Computing and Applications) · 2024

Pedro Antonino, Antoine Durand, A. W. Roscoe

Presents Hooks, a modular checkpointing protocol for Nakamoto-style blockchains. A hook is a forward-pointing digital signature from a block to one of its descendants, the forward analogue of a parent hash link. A quorum of hooks forms a checkpoint, enabling trustless bootstrap via public verifiability. Unlike prior protocols, Hooks does not solve an agreement problem; it extends the input chain's safety guarantees while requiring only weak safety, converting unsatisfied safety requirements into liveness issues.

Abstracting and Verifying Decentralised Systems in CSP

The Application of Formal Methods (LNCS 14900), Springer · 2024

A. W. Roscoe, Pedro Antonino, Jonathan Lawrence

Uses CSP to formalise assemblies of distributed agents (including blockchain nodes) operating as consensus machines, sequential state machines that achieve predictable outcomes despite Byzantine agents communicating via message passing.

The Consensus Machine: Formalising Consensus in the Presence of Malign Agents

Theories of Programming and Formal Methods (LNCS 14080), Springer · 2023

A. W. Roscoe, Pedro Antonino, Jonathan Lawrence

Applies formal CSP modelling and verification to blockchain consensus, addressing the problem of ensuring that decentralisation cannot allow two valid decisions to arise simultaneously. Models a primary consensus mechanism supplemented by a backup triggered when the primary is blocked.

TEEs

Flexible Remote Attestation of Pre-SNP SEV VMs Using SGX Enclaves

IEEE Access, vol. 11, pp. 90839–90856 · 2023

Pedro Antonino, Ante Derek, Wojciech Aleksander Wołoszyn

Proposes a protocol combining two TEE technologies: an SGX enclave acts as a trusted guest owner to deploy, attest, and provision AMD SEV VMs, bringing SGX-style flexible remote attestation to SEV. Formally verified with the Tamarin prover; demonstrated on accountable ML models.

Guardian: Symbolic Validation of Orderliness in SGX Enclaves

CCSW '21 — Proceedings of the 2021 ACM Cloud Computing Security Workshop (co-located with CCS), Virtual Event, pp. 111–123, ACM · November 2021

Pedro Antonino, Wojciech Aleksander Wołoszyn, A. W. Roscoe

Introduces the notion of an orderly SGX enclave, splitting enclave behaviour into execution phases, each enforcing a policy over accesses to untrusted memory, phase transitions, and register sanitisation. Implements Guardian, an open-source tool using symbolic execution (via angr) to automatically validate whether an enclave binary is orderly and to detect typical enclave attack primitives. Distinct from vulnerability-pattern tools like TeeRex: Guardian checks structural policy adherence rather than searching for known exploit patterns.

Trusted And Confidential Program Analysis (TCPA)

arXiv · 2021

Han Liu, Pedro Antonino, Zhiqiang Yang, Chao Liu, A. W. Roscoe

Develops a protocol for program certification where neither the producer nor an external body needs to be fully trusted. Using encrypted sources inside a TEE, TCPA allows all parties to trust delivered object code without the producer ever releasing unencrypted source to any external entity.