How should L2s outsource proofs without compromising decentralization?
In this talk from @Zk_Hub_Global, @NorbertVadas, Head of Product at @gevulot_labs (@thezkcloud), breaks down the risks of auction-based proving marketplaces and argues for a permissionless, randomized, and redundant approach to proof infrastructure.
A clear case against centralizing trust in ZK rollups 🤝
Also available on the @HouseofZK YouTube: piped.video/watch?v=WjHV2kIJ…
Apr 23, 2025 · 2:32 PM UTC
Key Takeaways - Practical Challenges in Proof Outsourcing by @NorbertVadas of @gevulot_labs/@thezkcloud
Risks of Auction-Based Proof Marketplaces
• Auction-style proof markets risk centralization by favoring resource-rich entities.
• Bidding races lead to wasted compute and create unhealthy dependencies for rollups.
• One-to-one auction matching lacks redundancy - failure requires restarting the process.
Lessons from Ethereum’s PBS Model
• MEV relays show that centralization can re-emerge even in large validator sets: 5 builders account for 90% of PBS blocks.
• Applying similar relay logic to ZK proving could lead to few provers dominating rollup finality.
Importance of Decentralization and Liveness
• Provers must inherit the same censorship resistance, neutrality, and liveness guarantees as the underlying network.
• Randomized prover selection, not auctions, is the most credible neutrality mechanism, akin to validator selection in PoS chains.
Need for Redundancy and Fallbacks
• Protocols must build in fallback options (e.g. secondary provers, open job windows) to ensure liveness and prevent downtime.
• Sequencers should not have to manually restart the proving process in case of failures.
Gevulot’s ZKCloud Architecture
• zkCloud uses permissionless, decentralized prover selection with built-in redundancy.
• Economics are protocol-driven to prevent cost manipulation by dominant provers.
• Designed specifically to meet the reliability and decentralization needs of L2s.
• Firestarter testnet is live for exploration.
Core Thesis
• Proof infrastructure should not reintroduce central points of failure.
• Decentralized L2s require decentralized proving, not trusted middlemen.



