Serokell is bringing its formal verification work under a dedicated team, building on nearly a decade of research and practical experience in the field. Formal methods have been part of our work since Serokell’s early years, from proving properties of blockchain systems to developing programming languages and verification tools. We’re now giving this expertise a more defined place within the company, with a focus on applying mathematical methods to software where correctness is especially important.
Nearly a decade of formal methods
Our work with formal verification dates back to some of Serokell’s earliest blockchain projects. In 2018, our researchers were using Agda, a dependently typed programming language and proof assistant, to prove properties of Snowdrop, a framework for blockchain development. The research covered read-only state access, transaction validation, and block processing. Daniel Rogozin documented some of this work in his three-part series, Constructive and Non-Constructive Proofs in Agda, followed by articles on formal verification, mathematical logic, and type theory.
The same interests shaped our commercial engineering work.
For Tezos, we developed Morley, a set of tools built around the Michelson smart contract language, including its typechecker, interpreter, and testing framework. We also worked with Runtime Verification on the K Framework, improving the performance and implementation of its Haskell backend used for symbolic execution. Our ongoing contributions to GHC, particularly the development of dependent types in Haskell, are another part of this history. They help bring more expressive correctness guarantees into general-purpose programming.
Giving formal verification a dedicated focus
The dedicated team will bring this experience together and make it easier to apply across Serokell’s projects. Its focus is on formal specifications, machine-checked proofs, and the tools needed to apply verification to real software. This includes identifying the properties that matter, expressing them precisely, and choosing appropriate techniques to establish that they hold. Recent advances in AI-assisted theorem proving are making proofs considerably cheaper to produce. As that happens, deciding what to prove and stating it precisely becomes the central part of the work, as we discussed in our recent analysis of Anthropic’s work. The team will pay particular attention to the growing volume of code written with AI agents, and to tooling that brings proofs closer to the languages production software is written in.
Continuing in the open
With a dedicated team, we intend to make this expertise more accessible to projects involving critical infrastructure, financial systems, decentralized protocols, and other software with demanding correctness requirements. We’ll share the team’s research, tools, and results through the Serokell blog. Interested in applying formal verification to your project? Get in touch with Serokell.


