For institutions
A validator your firm owns
The agent is untrusted. Every allow or deny leaves evidence an examiner can read. Human-in-the-loop is not a control; a real control sits upstream of execution.
CECO → Evidence trail →Edinburgh DeFi Summit · 1 September 2026
Agents can propose. They should not move the money until a rule has held.
Martyn Riley and Jamie Riley-Betts are at the Edinburgh Decentralised Finance Summit. Riley Betts builds formal specification and cryptographic enforcement for regulated AI — including when that AI is acting on tokenised assets, payments, or a wallet.
We are also in the cryptographic problem the STARK community is now living with: the Reed–Solomon proximity-gap assumption under FRI does not reach the capacity people priced in. We are researching whether other codes can restore a proven gap, and we are putting published agent protocols — x402, A2A, A2P — under Lean scrutiny. That work is a research programme, not a product.
For tokenisation & custody
A compact public receipt
Hashes and time on-chain. The book, the terms, and the client data stay off it. We sit beside custody, FIX, and SWIFT — we do not replace them.
Trust Fabric → OTC confirmation PoC →For builders & auditors
Specification is the scarce skill
Write what must be true. Check it. Steward it when agents write the code. Asking an agent to find bugs is not the same as proving a protocol respects a stated property.
Spec++ explainer → Lean in practice →For cryptographers · research programme
Proximity gaps after the Reed–Solomon conjecture
FRI-based STARKs assumed a Reed–Solomon proximity gap that has been refuted at capacity. Proven gaps for plain RS stop at the Johnson bound. We are researching successor codes — folded Reed–Solomon with subspace-design decoding is the strongest candidate we currently see — and stating in Lean what is and is not yet proven. Open questions remain. We have not replaced STARKs, and we have not claimed a prize.
Separately, we check published agent-economy protocols as specifications. Issues found under Lean scrutiny stay off this page until a disclosure posture exists. Happy to compare notes.
Research collaboration →Here today
Martyn Riley — cryptographic enforcement, formal security models, zero-knowledge proofs, the STARK / proximity-gap programme. LinkedIn →
Jamie Riley-Betts — formal methods, determinism versus probabilism, specification-driven development. University of St Andrews. LinkedIn →
On the site, not in the room
Robert Betts — twenty-five years building software for financial institutions. The FS history lives on the team page; Martyn and Jamie will send you there rather than perform it.
The team →If we met in the corridor: this page is the shortest path into the work. A technical conversation, not a sales call.