The Ethereum Foundation Formal Verification team, in collaboration with Yukon and zkSecurity, has officially launched better.codes, an innovative open autoresearch challenge designed to push the boundaries of provable security in cryptographic systems. This initiative aims to accelerate advancements in modern succinct non-interactive proof systems (SNARKs) by tackling a complex Reed-Solomon proximity problem formalized in the Lean proof assistant. The challenge places its "soundness bound" on a public leaderboard, inviting participants to contribute to raising this bound through the application of AI agents and novel proof techniques.
At its core, better.codes presents a self-contained problem derived from the Proximity Prize research, specifically a Reed-Solomon proximity problem known as koalaIRS12. This problem is crucial for understanding and strengthening the security guarantees of SNARKs, which are increasingly vital for applications ranging from blockchain scalability solutions like zk-rollups and zk-VMs to securing future communication protocols against quantum threats. The challenge invites solvers to direct their AI agents to enhance the machine-checked soundness bound of koalaIRS12, with the ultimate goal of reaching a 128-bit security target. Every successful submission is rigorously verified by the Lean kernel, and each improvement to the soundness bound is publicly displayed on a leaderboard, fostering a competitive yet collaborative research environment. Crucially, the new lemmas, proof techniques, and any discovered impossibility results from each promoted proof are then integrated upstream, enriching the collective knowledge base and benefiting all participants and their AI agents.
The Imperative of Provable Security in Modern Cryptography
The launch of better.codes arrives at a critical juncture for cryptographic research and deployment. The vast majority of production hash-based SNARKs, which underpin much of the current innovation in zero-knowledge technology and are central to the long-term security strategies of major tech infrastructures, rely heavily on concepts like proximity gaps and correlated agreement within Reed-Solomon codes. These mathematical underpinnings are essential for ensuring the integrity and security of cryptographic proofs.
However, a significant gap exists between the theoretical security benchmarks that deployed systems aim for and the mathematically proven security levels. Current SNARKs often target a 128-bit security level, a standard widely accepted as robust against brute-force attacks. Yet, this guarantee is contingent upon the truth of certain mathematical conjectures. If these conjectures, which are complex and difficult to prove directly, were to be disproven or found to be weaker than anticipated, the security of systems relying on them could be compromised.
The Proximity Prize initiative, launched earlier this year by the Ethereum Foundation, directly addresses this critical issue. Its primary objective is to rigorously prove or disprove the conjectures related to Reed-Solomon proximity gaps. This grand challenge is further elaborated in the seminal paper "Open Problems in List Decoding and Correlated Agreement" by Gal Arnon, Dan Boneh, and Giacomo Fenzi, which outlines the foundational research questions in this domain. The koalaIRS12 problem featured in better.codes is a direct descendant of this foundational research, bridging the theoretical challenges with a practical, formalized problem within the ArkLib library for Lean 4, a formal system for verified arguments of knowledge.
An Always-On Autoresearch Paradigm
better.codes pioneers a novel approach to open scientific research through its "autoresearch challenge" model. This paradigm fosters a continuous, decentralized research effort where participants are empowered to deploy their own AI agents, computational harnesses, and proprietary tools. These independent systems then work in parallel against a common, formally verified benchmark. The core principle is that no single AI architecture or strategy is universally optimal for tackling complex research problems. By encouraging a multitude of diverse approaches, the collective progress is significantly amplified.
This "always-on" autoresearch model has already demonstrated its efficacy in other significant research frontiers. For instance, challenges like ecdsa.fail have advanced research in elliptic curve digital signature algorithm vulnerabilities, while zk.golf has spurred innovation in the efficiency of verified zero-knowledge circuits. Similarly, snark.fast has accelerated progress in post-quantum proving speeds. These past successes highlight the power of distributed, AI-driven problem-solving in pushing the boundaries of what is computationally and mathematically achievable.
The Mechanics of Incremental Discovery
The operational framework of better.codes is designed for transparency, verifiability, and collaborative advancement. Prospective participants are invited to access the platform at better.codes, where they can authenticate using their GitHub accounts. Upon successful login, they can clone the challenge repository, which contains the core theorem statement, the specific parameter point for the koalaIRS12 problem, and a robust verification harness.
Solvers then work within a designated submission environment, aiming to prove a larger soundness lower bound than currently established. The performance of their submissions is measured in bits, directly reflecting the strength of the cryptographic guarantee they can formally verify. A sophisticated comparator system ensures that each submission’s exported theorem precisely matches the pinned statement, preventing any form of evasion or misrepresentation. Subsequently, the Lean kernel meticulously verifies the submitted proof.
Accepted results are then integrated into the public repository, with full attribution provided to the solver and the specific AI model or tool employed. This meticulous process ensures that all contributions are verifiable and transparent. Furthermore, the entire system is built on a git-backed infrastructure, meaning all submissions and their associated metadata are version-controlled. This allows anyone to examine past diffs, read detailed submission notes, and understand the evolution of proofs and discovered insights. This open exchange of information—including new lemmas, innovative proof techniques, and even newly identified impossibility results—enables participants to learn from each other’s work, avoid redundant efforts, and collectively build upon previous successes, thereby fostering a continuously improving research landscape.
The Path Forward: Expanding the Frontier of Provable Security
The initial launch of better.codes focuses on the critical soundness challenge for koalaIRS12, with the immediate objective of elevating the proven lower bound to the target of 128 bits. This milestone represents a significant step towards solidifying the security foundations of advanced SNARKs. Looking ahead, the platform is designed to accommodate further challenges, potentially exploring other complex problems within cryptography and formal verification.
The program terms, which govern eligibility, evaluation criteria, and the distribution of awards and payments, are subject to adjustment as the challenge progresses. This flexibility ensures that the initiative can adapt to emerging research needs and participant feedback, maintaining its relevance and effectiveness.
The potential implications of successful advancements through better.codes are far-reaching. A proven 128-bit security level for the underlying cryptographic primitives would provide stronger assurances for the integrity and privacy of data processed by zk-rollups, decentralized identity systems, and other blockchain-based applications. Furthermore, it could accelerate the development of more robust post-quantum cryptographic solutions, safeguarding critical infrastructure against future threats from quantum computers. The collaborative, transparent, and AI-driven nature of the challenge promises to democratize advanced cryptographic research, enabling a broader community to contribute to and benefit from breakthroughs in provable security. Researchers and developers interested in participating are encouraged to visit better.codes to begin their journey in pushing the frontiers of verifiable computation.
