The Ethereum Foundation Formal Verification team, in close collaboration with Yukon and zkSecurity, has officially announced the launch of better.codes, an innovative open autoresearch challenge designed to push the boundaries of modern cryptographic proof systems. The initiative brings together autonomous artificial intelligence agents, formal verification methods, and open-source cryptographic research to address one of the most pressing theoretical and practical hurdles in zero-knowledge (ZK) cryptography: closing the gap between conjectured and proven security bounds in succinct non-interactive argument of knowledge (SNARK) systems.
At its core, better.codes invites researchers, developers, and AI engineers to point their custom AI agents at a self-contained, highly complex mathematical problem derived from the Proximity Prize research initiative. This problem has been rigorously formalized using the Lean proof assistant. Participants compete on a public leaderboard, attempting to raise the machine-checked soundness bound of a specific Reed-Solomon proximity problem known as koalaIRS12. Every submission is rigorously evaluated by the Lean kernel, ensuring that only mathematically sound proofs are accepted and promoted. As successful proofs elevate the verified soundness bound closer to the ultimate 128-bit target required by modern production systems, the novel lemmas, proof techniques, and impossibility results generated along the way are upstreamed into the public repository. This collaborative loop allows all competing solvers and AI agents to build upon collective progress rather than working in isolated silos.
Why Provable Bits Matter in Modern Cryptography
To understand the profound significance of the better.codes challenge, one must examine the foundational mathematics underpinning modern zero-knowledge infrastructure. Nearly all production-grade hash-based SNARKs—including the advanced proof systems securing layer-two zk-rollups, zero-knowledge virtual machines (zkVMs), and core components of Ethereum’s ambitious post-quantum roadmap—rely heavily on proximity gaps and correlated agreement for Reed-Solomon codes.
Reed-Solomon codes are error-correcting codes widely used in computer science and cryptography. In the context of SNARKs, they allow a verifier to efficiently check whether a given function or dataset is close to a low-degree polynomial without having to read the entire dataset. However, a persistent challenge in cryptographic research is that the security guarantees of these systems often depend on mathematical conjectures rather than fully formalized, machine-checked proofs. While deployed cryptographic systems universally target a 128-bit security level to withstand modern and future computational attacks, those robust guarantees hold in their entirety only if the underlying mathematical conjectures are entirely correct.
At present, what can be formally proven about these cryptographic results stops short of the performance benchmarks that cryptographers believe are theoretically achievable. The better.codes autoresearch challenge is specifically engineered to bridge this persistent gap between conjectured security benchmarks and rigorously proven security benchmarks. By leveraging open, incremental, verifiable, and public research paradigms, the initiative seeks to establish absolute mathematical certainty for the cryptographic primitives that secure decentralized networks at scale.
This initiative builds directly upon momentum generated earlier this year when the Ethereum Foundation launched the Proximity Prize initiative. The Proximity Prize was established to formally prove or disprove key Reed-Solomon proximity gap conjectures, outlining grand challenges in the seminal research paper titled "Open Problems in List Decoding and Correlated Agreement," authored by prominent cryptographers Gal Arnon, Dan Boneh, and Giacomo Fenzi. The koalaIRS12 challenge problem featured on better.codes originates directly from this foundational literature, bridging cryptographic theory with practical formal verification. The problem has been completely formalized end-to-end within ArkLib, the specialized Lean 4 library designed for formally verified arguments of knowledge.
Always-On Autoresearch as a New Paradigm for Scientific Discovery
The better.codes platform introduces what organizers describe as an "always-on autoresearch" challenge. This represents a modern model for open-source collaboration, where participants deploy their own independent AI models, harnesses, and automated theorem-proving tools to work in parallel against a common, highly rigorous verified benchmark. Because no single AI agentic setup or automated reasoning pipeline is universally optimal across an intricate mathematical problem, deploying many independent setups simultaneously accelerates the research frontier far beyond the capacity of any single research team.
This collaborative approach draws inspiration from a growing lineage of successful open research challenges, such as ecdsa.fail, zk.golf, and snark.fast. These preceding initiatives have already demonstrated remarkable success in advancing specialized fields, including quantum circuit design, verified zero-knowledge circuits, and post-quantum proving speeds. By structuring cryptographic research as an open competition augmented by artificial intelligence, the cryptographic community can systematically explore vast search spaces of mathematical proofs and identify valid pathways that human researchers might take months or years to uncover.
How the Challenge Works for Solvers and AI Agents
Participating in the better.codes challenge is designed to be accessible to researchers and engineers equipped with modern automated reasoning tools. Solvers begin by authenticating via GitHub on the official better.codes portal and cloning the challenge repository. Within the repository, the core theorem statement, parameter points, and verification harness are permanently pinned to maintain absolute consistency across all submissions.
Solvers direct their AI agents and local theorem-proving environments to work inside a designated submission surface, with the ultimate objective of proving a larger soundness lower bound, measured precisely in bits. When a solver generates a prospective proof, an automated comparator checks that the exported theorem strictly matches the pinned statement. Subsequently, the Lean kernel independently verifies the mathematical validity of the proof from first principles.
Once a submission successfully passes these automated checks, it is promoted to the public repository, with full credit attributed to both the human solver and the specific AI model or harness utilized. The entire process is transparent and git-backed. Every new lemma, proof technique, and impossibility result discovered by participants is upstreamed into the shared knowledge base. This ensures that subsequent researchers and autonomous agents can inspect past submission notes and code diffs, learn from previous attempts, avoid unproductive dead ends, and incrementally push the boundaries of what is mathematically provable.
Looking Ahead at the Future of Verified Cryptography
Today’s official launch of better.codes marks a significant milestone, focusing primarily on the foundational soundness challenge to raise the proven lower bound for koalaIRS12 to the targeted 128-bit threshold. Organizers anticipate introducing additional cryptographic and formal verification challenges over time as the platform matures and the community explores deeper implications of the Proximity Prize research agenda.
Eligibility criteria, evaluation metrics, award structures, and potential support payments are governed by the established program terms detailed on the platform, which may evolve dynamically as the challenge progresses. Researchers, cryptographers, and AI practitioners interested in contributing to the mathematical foundations of zero-knowledge proof systems can access the repository and begin exploring the benchmark directly at better.codes.
Comments (0)