The Ethereum Foundation’s 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 cryptographic security. This initiative aims to significantly enhance the provable security of modern proof systems, which are foundational to a wide array of critical technologies, including zero-knowledge rollups, zero-knowledge virtual machines (zkVMs), and Ethereum’s own post-quantum cryptography roadmap. The platform is now live, inviting researchers, developers, and AI agents to contribute to a collective effort in fortifying cryptographic guarantees.
At its core, better.codes tackles a specific, self-contained problem derived from the Proximity Prize research. This problem, formalized in the Lean proof assistant, presents a challenge to advance the machine-checked soundness bound of "koalaIRS12," a Reed-Solomon proximity problem. The ultimate goal is to achieve a fixed 128-bit security target, a standard widely adopted for high-assurance cryptographic systems. Participants are encouraged to direct their AI agents and computational resources towards this objective, with every successful submission that raises the soundness bound being publicly recognized on a dedicated leaderboard.
The Imperative of Provable Security in Modern Cryptography
The significance of better.codes stems directly from the increasing reliance on sophisticated cryptographic techniques for securing digital infrastructure. Nearly all production hash-based Succinct Non-Interactive Arguments of Knowledge (SNARKs) – the cryptographic primitives underpinning many blockchain scaling solutions and advanced privacy technologies – depend on Reed-Solomon codes and their associated proximity gaps and correlated agreement properties. These mathematical underpinnings are what provide the assurance of security, often measured in bits of equivalent cryptographic strength.
However, a critical gap exists between the conjectured security levels of these systems and the provable security levels that can be rigorously demonstrated. Deployed systems often target 128-bit security, a benchmark offering a very high degree of confidence against sophisticated attacks. This guarantee, however, is contingent upon the truth of certain mathematical conjectures. When these conjectures remain unproven, the full 128-bit security guarantee is not yet rigorously established.
The better.codes challenge is engineered to systematically close this gap. By fostering an environment of open, incremental, verifiable, and public research, it aims to transform theoretical conjectures into proven theorems. This collaborative approach leverages the power of distributed effort, allowing a multitude of independent solvers and their AI agents to work in parallel on a common, formally verified benchmark.
A Timeline of Progress: From Proximity Prize to Autoresearch
The genesis of better.codes can be traced back to earlier this year with the Ethereum Foundation’s launch of the Proximity Prize initiative. This initiative was established with the ambitious goal of either proving or disproving the fundamental conjectures surrounding Reed-Solomon proximity gaps and correlated agreement. These grand challenges were formally laid out in the seminal paper, "Open Problems in List Decoding and Correlated Agreement," authored by Gal Arnon, Dan Boneh, and Giacomo Fenzi.
The "koalaIRS12" problem, now central to the better.codes challenge, is directly drawn from this influential paper. It serves as a concrete, formalized instantiation of the broader research questions posed by the Proximity Prize. The formalization of koalaIRS12 has been meticulously carried out using ArkLib, the Lean 4 library specifically designed for formally verified arguments of knowledge, ensuring a high degree of rigor and verifiability from the outset.
The launch of better.codes represents the next evolutionary step in this research effort. It transitions from a prize-based competition focused on solving specific open problems to an "always-on" autoresearch challenge. This new paradigm is designed to foster continuous, iterative progress.
The Autoresearch Model: Harnessing Collective Intelligence
better.codes exemplifies a novel approach to open scientific collaboration: the autoresearch challenge. In this model, participants are not limited to a single optimal strategy or a centralized research team. Instead, they are empowered to deploy their own diverse AI agents, computational harnesses, and algorithmic tools. These independent efforts are directed towards a shared, formally verified benchmark, creating a dynamic and competitive environment.
The rationale behind this model is that no single agentic setup is likely to be universally optimal for tackling complex, open-ended research problems. By encouraging a multitude of diverse approaches to run in parallel against the same benchmark, the collective progress frontier is pushed forward far more rapidly than any single team could achieve.
This autoresearch methodology has already demonstrated its efficacy in other cutting-edge fields. Initiatives such as ecdsa.fail, zk.golf, and snark.fast have pioneered this approach, leading to significant advancements in areas like quantum circuit design, the verification of complex zero-knowledge circuits, and optimizing the speed of post-quantum proving systems. These past successes provide a strong precedent for the potential impact of better.codes.
How the better.codes Challenge Operates
Participation in the better.codes challenge is designed to be accessible and transparent. Individuals can sign in using their GitHub accounts to access the platform. Once authenticated, they can clone the challenge repository, which contains the core components of the problem. This includes the precise theorem statement to be proven, the specific parameter point being investigated, and the verification harness that will automatically assess submissions.
Solvers are expected to work within a designated submission surface. Their objective is to prove a soundness lower bound that exceeds the current best known bound for koalaIRS12, measured in bits of security. Each submitted proof is rigorously checked by the Lean kernel, the core engine of the Lean proof assistant, ensuring its mathematical validity. Furthermore, a comparator mechanism verifies that the exported theorem from each submission precisely matches the pinned statement, preventing any form of manipulation or misrepresentation.
Accepted and verified submissions are then promoted to the public repository. This promotion not only acknowledges the solver’s contribution but also credits the specific AI model or approach used. This transparency is crucial for fostering a learning environment.
All submissions are managed via Git, providing a clear, auditable, and version-controlled history of the challenge’s progress. This means that new lemmas, innovative proof techniques, and identified impossibility results are systematically upstreamed into the main challenge repository. This open sharing of discoveries allows anyone participating in the challenge to review past diffs, study submission notes, learn from prior breakthroughs, and avoid repeating unsuccessful avenues of research. This incremental and collaborative advancement ensures that progress benefits all solvers and their agents collectively.
The Broader Implications for Cryptographic Security and Beyond
The launch of better.codes carries significant implications for the future of cryptography and the development of secure digital systems. By focusing on the formal verification of Reed-Solomon proximity gaps, the challenge directly addresses a fundamental weakness in the security guarantees of many contemporary proof systems.
For technologies like zk-rollups, which are vital for scaling the Ethereum network and enabling wider adoption of decentralized applications, the underlying cryptographic primitives must be exceptionally robust. Any doubt about their security, even if theoretical, can undermine confidence and adoption. The work undertaken through better.codes directly contributes to solidifying these assurances.
Furthermore, as the world transitions towards a post-quantum era, where current cryptographic standards may be vulnerable to attacks by powerful quantum computers, the development of quantum-resistant algorithms is paramount. Many of these post-quantum cryptographic roadmaps, including those being explored by organizations like the National Institute of Standards and Technology (NIST), also rely on principles related to coding theory and error correction, making the research facilitated by better.codes highly relevant.
The platform’s success hinges on the collective effort of its participants. The more diverse and numerous the AI agents and human solvers, the faster the 128-bit security target can be reached and potentially surpassed. The upstreaming of new findings ensures that the collective knowledge base grows with each submission, creating a virtuous cycle of innovation.
Looking Ahead: Expanding the Frontier of Verified Research
The initial phase of the better.codes challenge is dedicated to achieving the 128-bit soundness target for the "koalaIRS12" problem. This is a significant undertaking that, once accomplished, will provide a concrete, verifiable foundation for the security claims of numerous cryptographic systems.
However, the vision for better.codes extends beyond this initial goal. The platform is designed to be adaptable and scalable, with plans to introduce further challenges over time. These future challenges could explore other critical areas of formal verification or tackle different aspects of cryptographic security, building on the established autoresearch framework.
Eligibility criteria, evaluation processes, and the distribution of awards and payments are all governed by the program’s terms and conditions. These terms are subject to adjustment as the challenge progresses, reflecting the dynamic and evolving nature of the research endeavor.
The call to action is clear: researchers, cryptographers, AI developers, and anyone interested in the future of secure computation are invited to join the better.codes challenge. By contributing to this open, collaborative, and verifiable research effort, participants can play a direct role in shaping the future of cryptographic security and ensuring the robust protection of our increasingly digital world. The journey begins at better.codes, where the pursuit of provable security is now a collective, ongoing mission.
