The landscape of modern cryptography is undergoing a seismic shift. As the industry races toward more scalable, secure, and post-quantum resilient infrastructure, the underlying mathematical foundations of our digital world have become a primary battleground for research. This week, the Ethereum Foundation’s Formal Verification team, in a landmark collaboration with Yukon and zkSecurity, officially launched better.codes—a revolutionary open-source "autoresearch" challenge designed to bridge the gap between theoretical conjecture and proven cryptographic security.
By leveraging the Lean 4 theorem prover and harnessing the collective power of AI agents and human researchers, better.codes seeks to solve the "proximity gap" problem—a critical bottleneck in the performance and security of modern zero-knowledge proof (ZKP) systems.
Main Facts: The Intersection of AI and Formal Proofs
At its core, better.codes is an experimental, public, and incremental research platform. It invites participants to apply their own AI-driven agents, custom toolsets, and human ingenuity to tackle a specific, high-stakes mathematical problem: the soundness bounds of the koalaIRS12 Reed-Solomon proximity problem.
In the world of SNARKs (Succinct Non-Interactive Arguments of Knowledge), "soundness" is the property that ensures a prover cannot convince a verifier of a false statement. Most production-grade hash-based SNARKs—the engines powering zkRollups, zkVMs, and Ethereum’s long-term post-quantum roadmap—rely on assumptions about how closely a received data set approximates a valid Reed-Solomon codeword. Currently, the industry targets a 128-bit security margin, but the formal mathematical proofs for these guarantees often lag behind the practical implementations.
The better.codes platform changes this by:
- Formalizing the Problem: Using Lean, a language for formal verification, the challenge defines the koalaIRS12 problem end-to-end within the ArkLib ecosystem.
- Gamifying Research: Participants push their AI agents to raise the machine-checked soundness bound. Each successful submission is verified by the Lean kernel and promoted to a public leaderboard.
- Upstreaming Knowledge: Unlike traditional academic silos, better.codes makes every submission, lemma, and failed attempt transparent. This "open-diff" approach allows researchers to build upon one another’s work, effectively treating research as a shared, evolving repository.
Chronology: The Road to better.codes
The journey to this launch is the result of a deliberate, multi-year progression in the formal verification and cryptography communities.
The Proximity Prize Initiative
Earlier this year, the Ethereum Foundation launched the Proximity Prize, a formal initiative aimed at settling the Reed-Solomon proximity gap conjectures. This followed the publication of “Open Problems in List Decoding and Correlated Agreement” by researchers Gal Arnon, Dan Boneh, and Giacomo Fenzi. This paper provided the theoretical roadmap for the challenges now hosted on better.codes.
Precedent and Iteration
The development of better.codes did not occur in a vacuum. It draws heavily from the lessons learned by previous successful open-research experiments. Projects such as ecdsa.fail, zk.golf, and snark.fast pioneered the model of using public, competitive, and verifiable benchmarks to push the frontier of specific technical domains. These projects demonstrated that by creating a "always-on" environment, developers could rapidly accelerate progress in areas ranging from post-quantum proving speeds to circuit design optimization.
The Launch Phase
With the infrastructure now live, the project enters its first phase: the koalaIRS12 challenge. This phase focuses specifically on raising the proven lower bound for the code’s proximity properties. The organizers have stated that this is merely the inaugural challenge, with plans to expand the scope to other critical cryptographic problems as the community matures and the tooling improves.
Supporting Data: Why "Provable Bits" Matter
The urgency behind better.codes is driven by a simple, uncomfortable reality: the gap between "conjectured" and "proven" security.
The 128-Bit Standard
In modern cryptography, 128 bits is the "gold standard" for security. Achieving this level of security means that an attacker would theoretically need to perform $2^128$ operations to break the system—a task currently considered computationally infeasible for all existing human technology.
However, many current SNARK implementations rely on "conjectured" security bounds. These are mathematical claims that are widely believed to be true by the research community but have not yet been fully, rigorously proven in a machine-checked environment. If these conjectures are flawed, the entire security of a zkRollup or a zkVM could be compromised.
The Lean Advantage
Lean 4 is a functional programming language and theorem prover that allows mathematicians and computer scientists to write proofs that are verified by a computer. Unlike a peer-reviewed paper, which relies on the fallible human eye for verification, a Lean proof is "unshakeable"—if the kernel verifies the code, the proof is mathematically sound.
By pushing the soundness bound of koalaIRS12 through the Lean kernel, the better.codes community is essentially "hardening" the foundation of the Ethereum ecosystem. Every bit added to the proven soundness bound is a concrete step toward eliminating the risk of catastrophic implementation flaws.
Official Perspectives: The Collaborative Effort
The partnership behind better.codes represents a unique alignment of interests between the Ethereum Foundation and specialized security firms.
The Ethereum Foundation (Formal Verification Team): By prioritizing this research, the Foundation is signaling that the future of Ethereum is not just about throughput or UX, but about absolute mathematical certainty. The Foundation’s involvement provides the necessary resources and long-term vision to ensure that the research does not stall.
Yukon and zkSecurity: These firms bring deep technical expertise in zero-knowledge systems. Their role in building the infrastructure—from the Lean formalization in ArkLib to the user-friendly GitHub-based submission portal—is critical.
In a recent communication, the organizers emphasized that this is not a traditional bounty program where the prize goes to the person who finishes first. Instead, it is an autoresearch challenge. The "prize" is the advancement of the field. By requiring that every submission, including new lemmas and proof techniques, be upstreamed into the public repository, they are ensuring that the entire industry benefits, rather than a single winning team. This philosophy of "collective intelligence" is central to the project’s design.
Implications: The Future of "Autoresearch"
The launch of better.codes may signal a fundamental shift in how complex mathematical and cryptographic research is conducted.
1. The Rise of the Agentic Researcher
Traditionally, formal verification was the domain of highly specialized PhDs working in isolation for months or years. The better.codes model assumes that AI agents, working in parallel and guided by human overseers, can outperform monolithic, slow-moving research projects. By providing a standardized interface for AI models to interact with, the platform effectively turns the community into a massive, decentralized research lab.
2. A New Standard for Infrastructure Security
If better.codes succeeds in raising the soundness bounds for core cryptographic primitives, it will set a new standard for what it means to "deploy" a ZK system. In the future, we may see a world where no production-grade cryptographic system is considered "safe" unless it has been subjected to a similar, publicly verified autoresearch challenge. This could become a prerequisite for institutional adoption of zero-knowledge technologies.
3. Open Source as a Research Engine
The project underscores the power of open-source methodology applied to deep science. By utilizing Git-backed, transparent workflows, the community avoids "reinventing the wheel." Researchers can see exactly where a prior attempt failed, pick up from a specific lemma, and skip the dead-ends that consumed previous agents. This incremental, compounding progress is the hallmark of effective open-source development, and its application to formal mathematics is a major breakthrough.
4. Navigating the Unknown
Despite the optimism, the challenge ahead is significant. The koalaIRS12 problem is notoriously difficult. The researchers involved are essentially trying to solve a puzzle that has resisted formal proof for years. Whether or not they reach the 128-bit goal in the near term is secondary to the process they have established. Even if the goal remains elusive, the journey will yield new techniques in formal verification that will ripple through the entire computer science community.
Conclusion: How to Participate
The challenge is now open to the public. To participate, researchers and developers should:
- Visit better.codes to understand the current state of the leaderboard.
- Sign in with GitHub to access the challenge repository.
- Clone the environment and review the pinned theorem statements and the ArkLib formalization.
- Submit your work. Whether you are using a sophisticated AI agent or a manual proof, your submission will be evaluated by the Lean kernel and, if successful, will be merged into the project’s history.
The Ethereum Foundation, Yukon, and zkSecurity have provided the platform and the incentive, but the outcome rests with the community. As we look toward a future where ZK proofs secure the backbone of the internet, the work happening on better.codes represents the front line of digital trust. By turning proof-writing into a collaborative, high-stakes game, they are not just solving a mathematical riddle—they are securing the future of the decentralized web, one bit at a time.
