← Case Studies

Hellas: Protocol & Fraud-Game Economics


The client

Hellas is a fully decentralized network for verifiable off-chain tensor compute: it connects clients who need AI computation with providers who supply it, replacing trust with economic guarantees. Its core innovation, the Catgrad compiler, makes execution deterministic across environments, so any incorrect computation can be pinpointed and proven on-chain through succinct fraud proofs.

The problem

Providers post collateral as a commitment to correct execution. For an open compute market to work, cheating must carry negative expected value: under which stake, slashing, and dispute conditions is honest execution the equilibrium, and when do clients find it worthwhile to challenge incorrect results?

What we did

We specified the protocol’s economic structure: actors, value flows over the state-channel lifecycle, and fee distribution. We then analyzed the fraud game, deriving the minimum viable provider stake that eliminates the impunity region and characterizing detection as an inspection game with a mixed-strategy equilibrium. Finally, we validated the theory with an agent-based simulation over heterogeneous strategic populations, stress-testing reputation farming, Sybil attacks, collusion, griefing, and censorship.

Deliverables