The Ethereum Foundation Formal Verification team, in collaboration with Yukon and zkSecurity, has officially launched better.codes, a groundbreaking open autoresearch challenge designed to push the boundaries of provable security in cryptographic systems. This initiative aims to systematically enhance the machine-checked soundness bounds of cryptographic primitives, with an initial focus on a critical problem stemming from the Proximity Prize research. The platform provides a public leaderboard where participants, using their own AI agents and solvers, can contribute to raising a verified security benchmark, fostering a collaborative and competitive environment for scientific advancement.
At its core, better.codes tackles a self-contained problem formalized in the Lean proof assistant, derived from the Proximity Prize initiative. This problem, known as koalaIRS12, is a specific instance of a Reed-Solomon proximity problem. Its significance lies in its direct relevance to the advancement of modern succinct non-interactive proof systems, commonly known as SNARKs. These systems are foundational to a wide range of applications, including the security of zero-knowledge rollups and zero-knowledge virtual machines (zkVMs), and are increasingly integral to the long-term cryptographic security strategies of major blockchain networks like Ethereum, particularly in the context of post-quantum cryptography.
The challenge invites participants to direct their AI agents towards improving the soundness bound of koalaIRS12. The ultimate goal is to reach a 128-bit security target. The Lean kernel meticulously verifies every submission, and each successful advancement of the soundness bound is publicly recognized on the leaderboard. Crucially, the newly discovered lemmas, innovative proof techniques, and identified impossibility results from each promoted proof are then integrated upstream, benefiting all participants and their AI agents. This ensures that progress is not siloed but is instead a collective endeavor, accelerating the overall research frontier.
The Imperative of Provable Bits in Modern Cryptography
The reliance on Reed-Solomon codes and their associated proximity gaps and correlated agreement properties is pervasive in many of today’s most critical cryptographic applications. This includes the proof systems underpinning zero-knowledge rollups, which are essential for scaling blockchain networks by enabling off-chain computation with on-chain verification. Furthermore, zkVMs, which aim to execute arbitrary code within a zero-knowledge environment, also depend heavily on these cryptographic underpinnings.
Beyond current blockchain infrastructure, the security of Ethereum’s post-quantum roadmap is intrinsically linked to the robustness of these Reed-Solomon proximity gap conjectures. As the threat of quantum computing looms, the cryptographic community is actively seeking quantum-resistant solutions. Many proposed post-quantum cryptographic schemes, particularly those involving SNARKs, are built upon the very mathematical foundations that better.codes seeks to strengthen.
The current state of deployed systems often targets 128-bit security, a benchmark widely accepted as a robust level of cryptographic assurance against current and foreseeable computational capabilities. However, this guarantee is contingent upon the unproven conjectures surrounding Reed-Solomon proximity gaps. If these conjectures were to be disproven or weakened, the security assurances of these deployed systems could be undermined. The better.codes challenge is explicitly designed to bridge this critical gap between conjectured security levels and rigorously proven security levels. By fostering open, incremental, verifiable, and public research, it aims to provide a higher degree of certainty and transparency regarding the security of these vital cryptographic components.
This initiative builds upon the Ethereum Foundation’s broader commitment to advancing cryptographic research, exemplified by the Proximity Prize initiative launched earlier this year. The Proximity Prize, a significant research endeavor, aims to definitively prove or disprove the Reed-Solomon proximity gaps conjectures. These conjectures, central to the field of list decoding and correlated agreement, were formally articulated in the influential paper "Open Problems in List Decoding and Correlated Agreement" by Gal Arnon, Dan Boneh, and Giacomo Fenzi. The koalaIRS12 problem featured in better.codes is directly drawn from this research, serving as a concrete, formalized challenge that bridges the theoretical grand challenges with practical, verifiable progress. The entire problem has been meticulously formalized within ArkLib, the Lean 4 library dedicated to formally verified arguments of knowledge, underscoring the commitment to mathematical rigor.
An Innovative Model: Always-On Autoresearch
better.codes represents a novel approach to open scientific collaboration, termed "autoresearch." This paradigm empowers participants to deploy their own AI models, computational harnesses, and sophisticated tools to work concurrently on a shared, verified benchmark. The platform’s design ensures that every successful submission, by pushing the soundness bound forward, incrementally elevates the collective understanding and proven security floor for the problem at hand.
The rationale behind the autoresearch model is rooted in the understanding that no single algorithmic approach or agentic setup is universally optimal for tackling complex, open research problems. By allowing a multitude of independent setups to engage with the same benchmark, the collective intelligence and computational power can explore a broader solution space and identify advancements more rapidly than any individual team or centralized effort could achieve. This decentralized, parallelized research methodology has already demonstrated its efficacy in other cutting-edge domains. For instance, similar open challenges, such as ecdsa.fail, zk.golf, and snark.fast, have significantly accelerated progress in areas like quantum circuit design, the verification of zero-knowledge circuits, and the optimization of proving speeds for post-quantum cryptographic schemes. The success of these prior initiatives provides a strong precedent for the potential impact of better.codes.
The Mechanics of Contribution and Progress
Participating in the better.codes challenge is designed to be accessible yet rigorous. Individuals or teams can sign up using their GitHub accounts on the better.codes website. Once authenticated, they can clone the challenge repository, which contains the core components of the problem. This includes the precise theorem statement that needs to be proven, the specific parameter point for the koalaIRS12 problem, and the verification harness that will be used to test submissions.
Solvers operate within a dedicated submission surface provided by the platform. Their task is to develop proofs that demonstrate a higher soundness lower bound than previously achieved. The "score" for each submission is measured in bits, reflecting the improvement in the verified security level. A critical component of the system is a comparator that rigorously checks that each submitted theorem precisely matches the pinned statement. Following this, the Lean kernel, a robust and trusted formal verification engine, meticulously checks the validity of the submitted proof.
Accepted results are prominently featured on the public repository, with full credit given to the solver and the specific AI model or toolset they utilized. This transparency not only acknowledges individual contributions but also serves as a valuable dataset for the entire community. The entire submission process is transparent and version-controlled using Git. This ensures that all new lemmas, proof techniques, and identified impossibility results are systematically recorded and made accessible. Participants can examine past diffs and submission notes to understand the evolution of proofs, learn from previous approaches, and avoid redundant efforts or dead ends. This incremental approach to knowledge sharing is fundamental to the autoresearch model, ensuring that the entire community benefits from each individual breakthrough.
Future Trajectories and Broader Implications
The initial launch of better.codes focuses on the soundness challenge for koalaIRS12, with the immediate objective of raising the proven lower bound to the 128-bit security target. However, the platform is designed for scalability and future expansion. The organizers anticipate introducing additional challenges over time, addressing other critical areas within formal verification and cryptography. These future challenges may explore different cryptographic primitives, variations of existing problems, or entirely new research frontiers.
The eligibility criteria for participation, the methods for evaluating submissions, and the structure of awards and payments are all governed by the program’s terms and conditions. These terms may be subject to adjustments as the challenge evolves, reflecting the dynamic nature of research and development. The Ethereum Foundation and its collaborators are committed to fostering a fair and productive environment for all participants.
The launch of better.codes marks a significant step forward in the collective pursuit of provable security. By combining the power of formal verification, the ingenuity of AI, and the collaborative spirit of open research, this initiative has the potential to deliver tangible advancements in the security and trustworthiness of cryptographic systems that underpin much of our digital world. The lessons learned and the progress made through this challenge will undoubtedly inform the design and implementation of future cryptographic protocols, ensuring a more secure and resilient digital future.
Individuals and research teams interested in contributing to this important endeavor are encouraged to visit better.codes to learn more and begin their participation. The future of provable security is being built collaboratively, and better.codes provides a vital platform for that construction.
