Layla Zulfa
A pivotal open autoresearch challenge, better.codes, has officially launched, marking a significant collaborative effort by the Ethereum Foundation Formal Verification team alongside prominent entities Yukon and zkSecurity. This initiative aims to tackle a critical problem in the realm of modern cryptography, specifically focusing on the foundational security of succinct non-interactive proof systems (SNARKs) through a novel, open-source, and AI-driven approach to formal verification.
The better.codes platform centers on a highly technical, self-contained problem derived from the Proximity Prize research, formalized meticulously in Lean, a powerful interactive theorem prover. At its core, the challenge invites researchers and AI developers globally to contribute to a public leaderboard by incrementally improving the machine-checked soundness bound of ‘koalaIRS12.’ This particular problem pertains to Reed-Solomon proximity, a fundamental component underpinning the cryptographic guarantees of many SNARK constructions. The ultimate goal is to push the proven security bounds closer to the industry-standard 128-bit target, thereby closing the gap between theoretical conjectures and formally verified security.
The Crucial Role of Provable Bits in Cryptographic Security
The drive behind better.codes stems from a pressing need for absolute certainty in the security claims of cutting-edge cryptographic systems. Modern hash-based SNARKs are indispensable to the burgeoning blockchain ecosystem, serving as the cryptographic bedrock for innovations such as zero-knowledge rollups (zk-rollups), which scale Ethereum by processing transactions off-chain, and zero-knowledge Virtual Machines (zkVMs), which enable private and verifiable computation. Furthermore, these proof systems are integral to Ethereum’s long-term post-quantum roadmap, designed to future-proof the network against the theoretical threats posed by quantum computing.
A common thread across nearly all these deployed and planned systems is their reliance on the intricate properties of proximity gaps and correlated agreement in Reed-Solomon codes. While researchers have made significant progress in understanding these properties, the current state of formally proven results often falls short of what experts believe the benchmarks to be. Existing deployed systems, targeting a robust 128-bit security level, operate under the assumption that certain conjectures about these proximity gaps hold true. If these conjectures are not formally proven, a subtle yet critical vulnerability could theoretically exist, undermining the very security guarantees upon which these systems are built.
The better.codes autoresearch challenge directly confronts this discrepancy. It offers an open, transparent, and verifiable framework to bridge the chasm between conjectured and proven security benchmarks. By fostering incremental and publicly accessible research, the initiative seeks to eliminate any lingering doubts about the cryptographic strength of these foundational elements.
Tracing the Genesis: The Proximity Prize Initiative
The launch of better.codes is not an isolated event but rather a direct extension of a broader strategic effort initiated by the Ethereum Foundation earlier this year: the Proximity Prize. This ambitious initiative was specifically conceived to either rigorously prove or definitively disprove the long-standing conjectures surrounding Reed-Solomon proximity gaps. The grand challenges and the intellectual framework for this endeavor were comprehensively laid out in the influential paper, "Open Problems in List Decoding and Correlated Agreement," authored by leading cryptographers Gal Arnon, Dan Boneh, and Giacomo Fenzi.
The ‘koalaIRS12’ problem, which forms the current focus of the better.codes challenge, is directly sourced from this seminal paper. It has been meticulously formalized end-to-end within ArkLib, a Lean 4 library specifically developed for formally verified arguments of knowledge. This formalization ensures that all submissions and proofs are machine-checkable and adhere to the highest standards of mathematical rigor, a crucial aspect for establishing cryptographic soundness.
The Paradigm Shift: Always-On Autoresearch
The methodological innovation of better.codes lies in its adoption of the "autoresearch challenge" model. This represents a new paradigm for open, collaborative research, moving beyond traditional contests or isolated academic pursuits. In an autoresearch challenge, participants deploy their own AI models, computational harnesses, and specialized tools to continuously work against a common, formally verified benchmark. Crucially, every successful, promoted submission to the platform not only advances the specific proof but also raises the baseline for all subsequent progress.
The underlying philosophy is that no single agentic setup or research team holds a monopoly on optimality when tackling complex open problems. By encouraging a diverse array of independent setups to simultaneously address the same benchmark, the collective research frontier can be pushed forward at an accelerated pace, far exceeding what any individual team or methodology could achieve in isolation. This collaborative competition model leverages the strengths of distributed intelligence and automated reasoning.
This isn’t the first foray into such a model for critical cryptographic and computational problems. Previous successful autoresearch challenges have already demonstrated the efficacy of this approach in various complex domains. Notable examples include:
- ecdsa.fail: An initiative focused on quantum circuit design, pushing the boundaries of what’s possible in post-quantum cryptography.
- zk.golf: A challenge dedicated to verified zero-knowledge circuits, aiming to optimize and secure the underlying mathematics of ZK proofs.
- snark.fast: An effort to enhance post-quantum proving speed, critical for the practical deployment of future-proof cryptographic systems.
These prior successes underscore the transformative potential of autoresearch to rapidly iterate on solutions, discover novel techniques, and collectively advance the state of the art across challenging scientific and engineering frontiers.
Mechanism of Engagement: How the Challenge Works
Participation in the better.codes challenge is designed to be streamlined and accessible to the global research community. Interested solvers begin by signing in with their GitHub accounts at better.codes, which grants them access to clone the challenge repository. This repository contains the immutable elements of the challenge: the precise theorem statement, the specific parameter point for the koalaIRS12 problem, and the robust verification harness.
Solvers operate within a designated submission surface, where their primary task is to construct and prove a larger soundness lower bound for koalaIRS12. The progress is quantifiably scored in "bits," reflecting the depth and certainty of the cryptographic guarantee achieved.
A rigorous verification process is central to the integrity of the challenge. Before any submission is considered, a comparator tool meticulously checks that the exported theorem from the solver’s proof exactly matches the pinned statement in the challenge repository, preventing any alteration of the problem itself. Following this, the Lean kernel, the computational engine of the Lean interactive theorem prover, performs a full, automated check of the submitted proof. Only proofs that are formally validated by the Lean kernel are accepted and subsequently promoted.
Upon acceptance, successful results are integrated into the public repository. Each promoted proof is fully credited to the submitting solver and, importantly, to the specific AI model or methodology employed. This transparency is key to the autoresearch model. All submissions are git-backed, meaning a complete, auditable history of progress is maintained. New lemmas, innovative proof techniques, and even impossibility results discovered during the challenge are upstreamed and made publicly available. This open sharing allows any solver to review past diffs and submission notes, build upon previous work, avoid redundant efforts, and sidestep unproductive "dead ends," thereby fostering a truly collaborative and incrementally advancing research environment for all participants and their AI agents.
Broader Implications and Future Trajectory
The launch of better.codes carries profound implications not just for the immediate security of SNARKs but for the wider fields of blockchain technology, formal verification, and the nascent integration of artificial intelligence into scientific discovery. By rigorously proving the soundness bounds of Reed-Solomon proximity problems, better.codes directly contributes to the foundational security of decentralized applications, enhancing trust in zk-rollups that handle billions in transaction volume and bolstering the cryptographic resilience of future blockchain architectures. This move towards fully verifiable security guarantees aligns perfectly with the ethos of public ledgers, where transparency and immutability are paramount.
Furthermore, this initiative highlights the increasing sophistication and utility of formal verification methods, particularly Lean, in tackling real-world cryptographic challenges. It demonstrates how these tools can move beyond academic curiosities to become essential components in the security assurance pipeline of critical infrastructure.
The autoresearch model itself stands as a testament to the power of open collaboration and automated reasoning. It represents a scalable and efficient way to explore complex problem spaces, potentially accelerating discovery in areas far beyond cryptography. The success of better.codes could pave the way for similar challenges in other scientific and mathematical domains where formal proof and incremental progress are vital.
For the Ethereum ecosystem, solidifying the cryptographic underpinnings of SNARKs is a strategic imperative. As the network continues to scale and evolve, relying on the strongest possible security assurances is non-negotiable. better.codes is a proactive step to ensure that the theoretical security claims of these advanced cryptographic primitives are matched by concrete, machine-checked proofs.
Today’s launch focuses on the specific soundness challenge of raising the proven lower bound for koalaIRS12 to the critical 128-bit target. However, the organizers have expressed an intention to expand the scope of better.codes over time, introducing further challenges that address other fundamental open problems in cryptography and formal verification. The eligibility criteria, evaluation metrics, awards, and payment structures for participants are governed by the program terms, which may be adjusted as the challenge progresses to best serve its evolving objectives.
The better.codes initiative represents a bold step forward in the quest for provably secure cryptographic systems, harnessing the collective intelligence of the global research community and the formidable capabilities of AI-driven formal verification. Those interested in contributing to this cutting-edge research are invited to begin their journey at better.codes.
