Layla Zulfa
The initiative, accessible at better.codes, represents a significant step forward in the realm of cryptographic security and formal verification. It directly addresses a critical gap in the understanding and provable security of advanced cryptographic primitives, particularly those essential for the robustness of succinct non-interactive proof systems (SNARKs). Built by the Ethereum Foundation’s Formal Verification team, in partnership with cybersecurity experts at Yukon and zkSecurity, better.codes invites researchers, developers, and AI agents worldwide to contribute to a shared objective: to incrementally enhance the machine-checked soundness bound of a specific Reed-Solomon proximity problem known as koalaIRS12. This problem is not merely an academic exercise; its resolution has direct, profound implications for the security guarantees of widely deployed cryptographic systems, including those powering zk-rollups, zkVMs, and the future roadmap of the Ethereum blockchain.
Understanding the Core Challenge: koalaIRS12 and Formal Verification
At its heart, better.codes focuses on a self-contained problem derived from the broader Proximity Prize research, an Ethereum Foundation initiative dedicated to advancing the state of formal verification in cryptography. This particular problem, koalaIRS12, is meticulously formalized in Lean, a powerful proof assistant and functional programming language. Lean is instrumental in this endeavor because it allows for the creation of machine-checkable proofs, ensuring an unprecedented level of rigor and certainty in mathematical and logical assertions. Unlike traditional peer review processes, where human error or oversight can exist, proofs verified by Lean’s kernel offer cryptographic-grade assurance.
Participants in the better.codes challenge are tasked with deploying their own AI agents or custom tools to search for and prove a higher soundness bound for koalaIRS12. The current known soundness bound is displayed on a public leaderboard, which serves as a real-time indicator of progress. Each successful submission, once verified by the Lean kernel, pushes this bound further towards a fixed 128-bit security target. This iterative process is designed to be highly collaborative. Crucially, any new lemmas, proof techniques, or impossibility results discovered and incorporated into a promoted proof are then "upstreamed" into the challenge’s public repository. This open-source approach ensures that all solvers and agents can build upon the collective knowledge, avoiding redundant efforts and accelerating the overall pace of discovery. This shared knowledge base is fundamental to the autoresearch model, fostering a continuous cycle of improvement and innovation.
The Imperative of Provable Bits: Why 128-bit Security Matters
The seemingly abstract concept of "provable bits" carries immense practical significance for the security of digital systems, particularly in the rapidly evolving landscape of blockchain technology. Modern cryptographic systems, including virtually all production hash-based SNARKs, rely heavily on specific mathematical properties such as proximity gaps and correlated agreement for Reed-Solomon codes. These cryptographic primitives are the bedrock for critical applications, ranging from the privacy-preserving zero-knowledge rollups (zk-rollups) that scale Ethereum to the secure execution environments of zero-knowledge Virtual Machines (zkVMs), and even elements central to Ethereum’s long-term post-quantum roadmap.
Currently, there exists a notable discrepancy between the conjectured security benchmarks for these systems and what can be formally proven. While deployed systems commonly target a 128-bit security level – a standard considered robust against most known computational attacks in the foreseeable future – this guarantee often rests on unproven conjectures. If these conjectures were to be disproven or found to be weaker than assumed, the security of these systems could be compromised, potentially exposing sensitive data or financial assets. The better.codes autoresearch challenge directly aims to bridge this critical gap. By fostering open, incremental, verifiable, and public research, it seeks to transform conjectured security benchmarks into definitively proven ones. This endeavor is not merely about theoretical purity; it’s about providing concrete, mathematical assurances that the cryptographic systems securing billions of dollars in digital assets and countless sensitive transactions are as robust as they are claimed to be.
The origins of this focused effort can be traced back to the Ethereum Foundation’s broader Proximity Prize initiative, launched earlier this year. This initiative was specifically designed to either formally prove or disprove the long-standing Reed-Solomon proximity gaps conjectures. The "grand challenges" of this research area were comprehensively outlined in the influential paper "Open Problems in List Decoding and Correlated Agreement" by leading cryptographers Gal Arnon, Dan Boneh, and Giacomo Fenzi. The koalaIRS12 problem, chosen as the inaugural challenge for better.codes, is directly extracted from this seminal paper, establishing a clear link between the practical, hands-on efforts of the challenge and the highest echelons of cryptographic theory. Furthermore, the problem is formalized end-to-end within ArkLib, the Lean 4 library specifically developed for formally verified arguments of knowledge, ensuring that all work within better.codes contributes to a robust and auditable foundation.
The Autoresearch Paradigm: Accelerating Discovery through Open Collaboration
The "autoresearch challenge" model adopted by better.codes represents a paradigm shift in how complex scientific and mathematical problems can be tackled. It moves beyond traditional, often siloed, research efforts by establishing an "always-on" environment for open collaboration. In this model, participants are encouraged to run their own AI models, specialized computational harnesses, and custom software tools in parallel, all working towards a common, formally verified benchmark. The genius of this approach lies in its ability to leverage the diversity of computational strategies and human ingenuity simultaneously.
The premise is straightforward yet powerful: no single agentic setup or research team is likely to possess the optimal strategy for solving an entirely open problem. By enabling numerous independent setups to work concurrently on the same benchmark, the collective frontier of knowledge can be advanced significantly faster than any single team could achieve alone. This distributed, competitive, yet collaborative framework ensures that various approaches – from novel AI algorithms to unique heuristic search methods – are explored efficiently. Each successful, machine-verified submission incrementally raises the floor for progress, establishing new baselines for subsequent efforts.
This autoresearch model has already demonstrated its effectiveness in other critical areas of cryptographic research. Notable examples include:
- ecdsa.fail: This challenge focused on advancing the design of quantum circuits. By crowdsourcing efforts to optimize quantum algorithms related to elliptic curve digital signatures, it pushed the boundaries of what was thought possible in post-quantum cryptography research.
- zk.golf: This initiative aimed at verifying zero-knowledge (ZK) circuits. It facilitated the development of more efficient and formally proven ZK circuits, which are crucial for the scalability and privacy of blockchain applications.
- snark.fast: Dedicated to improving the proving speed of SNARKs in a post-quantum context, this challenge brought together diverse computational approaches to make these complex cryptographic proofs more practical and performant.
The successes of these preceding challenges underscore the potential of better.codes to make similar breakthroughs in the formal verification of Reed-Solomon proximity problems, a field critical for the long-term security and viability of decentralized systems.
Participation and Operational Transparency: A Call to Action for Solvers
Engaging with the better.codes challenge has been designed to be as straightforward and transparent as possible, welcoming a broad spectrum of participants from academic researchers to independent developers and AI enthusiasts. The initial step for any interested solver is to sign in with their GitHub account at better.codes. This grants access to the challenge repository, which contains all necessary resources.
Within this repository, the core elements of the challenge are clearly defined and "pinned": the precise theorem statement to be proven, the specific parameter point for the koalaIRS12 problem, and the verification harness. Solvers operate within a designated "submission surface," which essentially provides the environment and tools needed to construct their proofs. Their objective is to prove a larger soundness lower bound for koalaIRS12, with their progress being quantitatively scored in "bits" on the public leaderboard.
The verification process is robust and automated, ensuring the integrity of all submissions. When a solver submits a proof, a comparator first checks that the exported theorem statement from their submission precisely matches the pinned, official statement. This ensures that solvers are indeed tackling the defined problem and not a subtly different one. Following this, the Lean kernel, the heart of the formal verification system, rigorously checks the submitted proof. Only proofs that are fully validated by the Lean kernel are accepted and promoted to the public repository. Upon acceptance, the achievement is publicly credited to the solver and, significantly, to the specific AI model or methodology they employed. This attribution not only recognizes individual contributions but also provides valuable data on the effectiveness of different AI strategies.
A cornerstone of the better.codes philosophy is its commitment to transparency and collaborative progress. All submissions are git-backed, meaning every change, every new lemma, and every proof attempt is recorded and made publicly accessible. This allows the entire community to read past "diffs" (differences between versions) and submission notes, to analyze successful and unsuccessful approaches, and to build directly on prior work. By transparently sharing new lemmas, innovative proof techniques, and even impossibility results (which inform future directions by indicating dead ends), the challenge ensures that knowledge is shared, collective intelligence is amplified, and progress for all solvers and AI agents is incrementally advanced. This creates a powerful feedback loop where every contribution, whether directly raising the bound or merely informing the community, accelerates the overall research trajectory.
Broader Implications and The Road Ahead
The launch of better.codes transcends the immediate goal of proving a 128-bit soundness bound for koalaIRS12. It embodies a larger vision for the future of cryptographic security, formal verification, and decentralized scientific research. The successful resolution of the koalaIRS12 challenge will directly strengthen the foundational cryptographic primitives used in next-generation blockchain technologies. By converting cryptographic conjectures into undeniable mathematical proofs, it will enhance the trustworthiness and long-term resilience of systems that manage vast amounts of value and critical data. This enhanced provable security is paramount for fostering greater adoption and confidence in the decentralized web.
Beyond its immediate impact on SNARKs and blockchain, better.codes serves as a powerful experiment in the application of formal methods to complex, open-ended mathematical problems. It demonstrates how advanced tools like Lean, combined with AI, can revolutionize the pace and reliability of mathematical discovery. The autoresearch model itself has profound implications for how scientific research might be conducted in the future, offering a blueprint for democratized, accelerated, and highly verifiable progress across various domains. By lowering the barriers to entry for contribution and fostering a truly global, collaborative environment, it could unlock solutions to problems that have long eluded traditional research structures.
While today’s launch focuses on the soundness challenge for koalaIRS12 and the critical 128-bit target, the organizers have indicated a clear ambition to expand the scope of better.codes. The statement that "We hope to add further challenges over time" suggests a long-term commitment to this autoresearch model, potentially tackling other critical open problems in cryptography and formal verification. The eligibility criteria, evaluation processes, and any potential awards or payments associated with the challenge are governed by specific program terms, which the organizers note may be adjusted as the initiative progresses to ensure fairness and adaptability.
In conclusion, better.codes is more than just a competition; it is a pioneering initiative at the intersection of blockchain, artificial intelligence, and formal mathematics. It represents a bold step towards a future where the security of our digital infrastructure is not merely assumed but rigorously and demonstrably proven. As the global community of researchers and AI agents converges on this critical problem, the insights and breakthroughs generated are poised to have a lasting impact on the security, scalability, and trustworthiness of the decentralized technologies shaping our future.
Individuals interested in contributing to this groundbreaking research are encouraged to visit better.codes to begin their journey.
