Better.codes: An Open Autoresearch Challenge to Strengthen the Cryptographic Foundations of SNARKs and Blockchain Scalability
The launch of better.codes marks a significant shift in how the cryptographic community approaches the formal verification of proof systems, introducing an open, "always-on" autoresearch challenge aimed at bridging the gap between conjectured and mathematically proven security bounds. Developed by the Ethereum Foundation’s Formal Verification team in strategic partnership with Yukon and zkSecurity, the platform provides a collaborative environment for researchers and AI agents to address foundational security vulnerabilities in modern succinct non-interactive proof systems (SNARKs). By utilizing the Lean theorem prover to host a public leaderboard, the initiative seeks to push the soundness bound of the koalaIRS12 Reed-Solomon proximity problem toward a 128-bit security target, a standard required for the robust operation of zero-knowledge rollups and decentralized virtual machines.
The Cryptographic Urgency of Proximity Gaps
At the heart of the modern cryptographic stack—particularly within Ethereum’s post-quantum roadmap and the infrastructure of zero-knowledge rollups—lies a reliance on hash-based SNARKs. These systems are essential for verifying the integrity of computations without revealing the underlying data. However, their security is not absolute; it is frequently contingent upon mathematical conjectures regarding Reed-Solomon codes. These codes are used to encode data in a way that allows for efficient error correction and proof generation.
The challenge, as identified by researchers, is that while these systems are designed to offer 128-bit security—a level of protection considered virtually unbreakable by classical computing standards—that guarantee is only as strong as the underlying mathematical assumptions. Currently, the "proximity gaps" in these codes remain a subject of active inquiry. If these gaps are larger than conjectured, or if the proof of their bounds is insufficient, the integrity of entire blockchain networks could be theoretically compromised. The better.codes initiative serves as a public repository for researchers to formally prove these bounds, moving away from theoretical conjecture toward machine-checked, immutable certainty.
Chronology of the Proximity Prize Initiative
The origins of better.codes can be traced back to the broader Proximity Prize initiative, which was launched by the Ethereum Foundation earlier this year. The initiative was designed to address the lack of rigorous, formal proofs for the conjectures surrounding list decoding and correlated agreement, which are fundamental to the security of Reed-Solomon proximity tests.
In early 2024, the publication of the seminal paper Open Problems in List Decoding and Correlated Agreement by Gal Arnon, Dan Boneh, and Giacomo Fenzi served as a rallying cry for the cryptography community. The paper outlined the specific mathematical challenges that, if solved, would provide a more solid foundation for the next generation of cryptographic proofs. The better.codes platform is the practical, executable extension of these research goals. By formalizing the koalaIRS12 problem within ArkLib—a specialized Lean 4 library for formally verified arguments of knowledge—the team has created a structured, reproducible environment where solvers can submit improvements that are validated by the Lean kernel.
The Mechanics of Autoresearch
Better.codes introduces a novel collaborative model dubbed "autoresearch." Unlike traditional academic research, which often functions in silos, this model treats the challenge as a competitive yet cooperative optimization problem. Participants—which may include both human researchers and autonomous AI agents—clone the challenge repository, work on the soundness lower bounds, and submit their proofs for verification.
The process is inherently transparent and incremental. Every submission is recorded on a git-backed ledger, ensuring that the history of progress, including failed attempts and successful breakthroughs, is available to the entire community. When a solver pushes the soundness bound higher, that new knowledge becomes the baseline for all subsequent participants. This "ratcheting" effect ensures that the community does not duplicate effort or repeat historical mistakes. By allowing AI agents to work in parallel against a common benchmark, the platform effectively accelerates the rate of scientific discovery, as the collective intelligence of various agents can identify optimization paths that a single human team might overlook.
Data-Driven Security Benchmarks
The specific objective of the initial challenge, koalaIRS12, is to reach a 128-bit soundness bound. In the context of SNARKs, this represents the probability of a prover successfully forging a fake proof. A 128-bit target implies that an attacker would have to perform roughly 2^128 operations to find a collision or a vulnerability—a computational task that is effectively impossible with existing technology.
Current research suggests that while many systems claim this security level, the formal proofs supporting them often rely on heuristic assumptions. By moving these proofs into the Lean theorem prover, better.codes mandates that every claim be mathematically verified by the machine. This transition from "paper-and-pencil" mathematics to "machine-checked" proofs is a vital development for decentralized finance, where a single bug in a proof system could lead to the loss of billions of dollars in locked assets.
Industry Implications and Broader Impact
The launch of better.codes is not an isolated experiment; it is part of a larger trend of gamifying and decentralizing hard mathematical problems. Similar projects like ecdsa.fail, zk.golf, and snark.fast have previously demonstrated that creating an open, competitive environment can lead to rapid breakthroughs in areas like quantum circuit design and ZK circuit optimization.
The implications for the broader tech sector are profound. First, by standardizing the verification process, the industry can create a shared library of "proven" components, reducing the security overhead for new blockchain projects. Second, the reliance on AI-driven research agents marks a transition in the scientific method. As AI models become more adept at symbolic logic and formal verification, they are increasingly capable of finding proofs that are too complex or tedious for human mathematicians.
From an economic perspective, the initiative also introduces a meritocratic structure for scientific funding. By providing a platform where contributions are explicitly credited—whether to an individual, a lab, or an AI model—the Ethereum Foundation is setting a precedent for how public goods research can be incentivized and rewarded in an open-source ecosystem.
Future Trajectories
While the current launch focuses on the koalaIRS12 problem, the roadmap for better.codes suggests an expanding scope. The organizers have indicated that further challenges will be added as the current one matures. The program terms, which govern eligibility and potential rewards for successful contributors, are designed to be flexible, reflecting the iterative nature of the research.
As the platform gains traction, it is expected that the competitive nature of the leaderboard will attract a diverse range of participants, from academic cryptographers to machine learning engineers specialized in automated theorem proving. The success of this endeavor will likely be measured not just by the attainment of the 128-bit goal, but by the creation of a permanent, verified knowledge base that serves as a foundation for all future SNARK-based systems.
By prioritizing formal verification, the Ethereum Foundation is signaling a pivot toward "security-first" development. In an era where blockchain systems are increasingly complex and prone to sophisticated exploits, the ability to mathematically guarantee the soundness of cryptographic primitives is no longer a luxury—it is a requirement for the long-term viability of the decentralized web. As the community converges on better.codes, the focus remains clear: the objective is not just to build faster systems, but to build systems that are, by their very nature, provably secure.
