In an era where the integrity of digital infrastructure hinges on complex cryptographic proofs, the gap between "conjectured" security and "mathematically proven" security has become a critical vulnerability. To bridge this divide, a collaborative powerhouse—the Ethereum Foundation’s Formal Verification team, in partnership with Yukon and zkSecurity—has officially launched better.codes. This ambitious open-source autoresearch challenge seeks to harness the collective intelligence of AI agents and human researchers to solidify the foundation of modern Zero-Knowledge (ZK) proof systems.
The Genesis: Closing the Gap in Cryptographic Soundness
At the heart of the modern blockchain ecosystem lie Succinct Non-Interactive Proof Systems (SNARKs). From the scalability engines powering zkRollups to the privacy-preserving virtual machines and Ethereum’s own post-quantum roadmap, these systems are ubiquitous. However, their security is not absolute.
Nearly all production-grade hash-based SNARKs rely on complex conjectures regarding Reed-Solomon proximity gaps and correlated agreement. While developers design these systems to target a 128-bit security threshold, this target is currently an optimistic estimate rather than a fully verified mathematical certainty. The security guarantee holds in full only if the underlying mathematical conjectures are proven correct.
The better.codes initiative is a direct response to this "proof gap." By formalizing these problems in Lean—a powerful interactive theorem prover—the project moves beyond theoretical speculation and into the realm of machine-checked, absolute mathematical truth.
Chronology: From Academic Theory to Distributed Autoresearch
The journey toward better.codes did not happen in a vacuum. It represents the culmination of a broader shift in how cryptographic research is conducted.
- Early 2024: The Ethereum Foundation introduced the "Proximity Prize" initiative. This was a rallying cry to the global research community to tackle the unresolved conjectures surrounding Reed-Solomon proximity.
- Mid-2024: Researchers Gal Arnon, Dan Boneh, and Giacomo Fenzi published Open Problems in List Decoding and Correlated Agreement, outlining the "grand challenges" that would serve as the theoretical bedrock for future work.
- Late 2024: The development of ArkLib, a dedicated library for formally verified arguments of knowledge in Lean 4, provided the technical infrastructure necessary to translate high-level cryptography into machine-verifiable code.
- Present Day: The official launch of
better.codestransitions these academic inquiries into an "always-on", public-facing autoresearch platform.
This progression marks a pivot from traditional, siloed academic publishing to an iterative, collaborative, and automated model of scientific discovery.
The Mechanism: How Autoresearch Works
better.codes is not a static repository; it is a dynamic, living laboratory. The platform functions as a competitive, yet highly collaborative, environment for AI agents and human researchers.
The Challenge: koalaIRS12
The inaugural challenge focuses on koalaIRS12, a specific proximity problem derived from the aforementioned research. Solvers are tasked with raising the machine-checked soundness bound of this system. The objective is clear: push the current lower bound toward the target of 128 bits.
The Workflow
- Repository Cloning: Participants sign in via GitHub and clone the challenge repository. The theorem statement and the verification harness are pinned, ensuring that all participants are working from the exact same starting line.
- Agent Deployment: Solvers point their custom AI agents or manual proof-writing setups at the problem. Because no single AI architecture is optimal for all proof-generation tasks, the platform encourages a diversity of approaches.
- Lean Verification: Every submission is put through the rigorous Lean kernel. If the math is sound, the submission is accepted.
- Upstreaming and Evolution: Crucially,
better.codesis designed for cumulative progress. Each successful proof—including the lemmas, proof techniques, and even "impossibility results" that rule out certain paths—is upstreamed. This transparency allows the entire community to avoid "dead ends" and build directly upon the successes of others.
Supporting Data: The Power of Open Collaboration
The methodology behind better.codes is validated by a series of successful predecessors. Projects like ecdsa.fail, zk.golf, and snark.fast have already demonstrated the efficacy of open-source, gamified, and automated research models.
In the case of snark.fast, the community successfully accelerated the speed of post-quantum proving by several orders of magnitude through collaborative benchmarking. By applying this "always-on" model to formal verification, the organizers of better.codes expect a similar acceleration.
The data suggests that when individual agents are allowed to operate in parallel against a shared benchmark, the "frontier of knowledge" moves significantly faster than in a traditional peer-review cycle. The public nature of the leaderboard serves as both a record of progress and a continuous incentive for improvement.
Official Perspectives: Why "Provable Bits" Matter
For the Ethereum Foundation and its collaborators, this is not just about solving a math problem; it is about building a secure future for the internet of value.
In official documentation regarding the launch, the team emphasized that the security of decentralized finance (DeFi) and privacy-preserving applications is fundamentally tied to the "provability" of these systems. "What can be proven about these results today stops short of what researchers believe the benchmarks may be," a spokesperson noted. By closing this gap, the community isn’t just checking boxes; it is hardening the critical infrastructure that will eventually support trillions of dollars in assets.
The involvement of zkSecurity and Yukon underscores the practical application of this research. These organizations are not merely theoretical; they are responsible for the audit and development of systems that currently secure active networks. Their participation ensures that the proofs generated on better.codes have immediate, real-world utility in the production environments of zkVMs and zkRollups.
Implications: The Paradigm Shift in Scientific Research
The launch of better.codes signals a profound transformation in how the scientific community treats formal verification.
1. The Democratization of Formal Methods
Historically, formal verification was the domain of a few elite academic institutions. By providing a GitHub-integrated interface and a clearly defined, incentivized challenge, better.codes makes this sophisticated practice accessible to any developer with an interest in cryptography and AI.
2. AI-Assisted Formal Verification
We are witnessing the emergence of "Agentic Research." By tasking AI models with exploring the proof space of koalaIRS12, the challenge acts as a testbed for the next generation of AI-driven mathematical discovery. If an AI can successfully raise a soundness bound through automated lemma generation, it validates the role of AI as a primary tool for future mathematical research.
3. Cumulative Knowledge
Traditional research often suffers from "knowledge loss," where techniques developed in one paper are buried or forgotten in subsequent publications. Because better.codes requires all submissions to be git-backed and upstreamed, the project creates an immutable, growing library of knowledge. Future researchers will be able to look back at the diffs of this challenge to understand exactly how the soundness bound was raised, step by step.
4. A New Standard for Cryptographic Audits
As this platform matures, it may set a new standard for how new cryptographic primitives are audited. Instead of relying solely on a private audit report, future developers might be expected to provide a better.codes-style proof that their system meets a specific, verified soundness bound. This would fundamentally increase the level of trust in the cryptographic systems that underpin our digital lives.
What Comes Next: A Call to Action
The current launch is merely the beginning. While the focus remains on the 128-bit target for koalaIRS12, the team has signaled their intent to introduce further challenges as the platform matures.
For the developer community, the call to action is clear: visit better.codes, review the documentation, and begin contributing to the effort. Whether through manual mathematical rigor or the deployment of advanced AI agents, every submission contributes to a more secure and verifiable cryptographic landscape.
As we look toward the future, the work being done on better.codes serves as a reminder that the most complex problems are not solved by individuals in isolation, but by a community working in concert, empowered by the tools of the digital age. By turning the "unknown" into "provable bits," the contributors to this project are not just solving a puzzle—they are reinforcing the very fabric of decentralized trust.
