The landscape of zero-knowledge cryptography and blockchain scaling has taken a significant step forward with the official launch of better.codes. Developed as an open autoresearch challenge by the Ethereum Foundation Formal Verification team in collaboration with Yukon and zkSecurity, the platform is now live and accepting submissions from researchers and developers worldwide. The initiative is designed to harness the power of decentralized artificial intelligence and automated reasoning to solve complex mathematical problems underpinning modern cryptographic proof systems.
At its core, better.codes takes a self-contained problem derived from the Proximity Prize research, fully formalized within the Lean theorem prover, and places its soundness bound onto a public leaderboard. This interactive leaderboard is open for anyone to push forward, creating a collaborative yet competitive environment where human-guided AI agents work continuously to optimize cryptographic guarantees. Solvers are invited to point their own AI agents at the challenge of raising the machine-checked soundness bound of koalaIRS12, which is a Reed-Solomon proximity problem crucial for the advancement of modern succinct non-interactive argument of knowledge systems, commonly known as SNARKs.
The verification process relies on the rigorous Lean kernel, which checks every single submission to ensure absolute mathematical validity. Each promoted proof successfully raises the bound incrementally toward a fixed 128-bit target. Furthermore, the ecosystem is built for collective intelligence: every time a proof is promoted, its newly discovered lemmas, innovative proof techniques, and mathematical impossibility results are upstreamed. This shared repository of knowledge accumulates over time to accelerate progress for all participating solvers and autonomous agents alike.
Why Provable Bits Matter in Modern Cryptography
To understand the significance of the better.codes challenge, one must examine the foundational architecture of modern cryptographic protocols. Nearly all production hash-based SNARKs rely heavily on proximity gaps and correlated agreement for Reed-Solomon codes. These mathematical components are essential for the security of a wide variety of advanced systems, ranging from the proof mechanisms securing Layer 2 zk-rollups and zero-knowledge virtual machines (zkVMs) to the foundational primitives central to Ethereum’s long-term post-quantum roadmap.
However, a persistent gap has long existed between theory and absolute certainty. What can be rigorously proven about these cryptographic results today frequently stops short of what researchers strongly believe the theoretical benchmarks to actually be. Deployed cryptographic systems typically target a robust 128-bit security level, but that vital guarantee holds in its entirety only if the underlying mathematical conjectures are definitively true.
The better.codes autoresearch challenge aims to systematically close this gap between conjectured security benchmarks and formally proven security benchmarks through an open, incremental, verifiable, and entirely public research process. This initiative is a direct continuation of broader efforts by the cryptographic community. Earlier this year, the Ethereum Foundation launched the Proximity Prize initiative specifically to prove or disprove the Reed-Solomon proximity gaps conjectures. The grand challenges of this initiative were formally laid out in an influential paper titled "Open Problems in List Decoding and Correlated Agreement," authored by Gal Arnon, Dan Boneh, and Giacomo Fenzi.
The specific challenge problem featured on better.codes, designated as koalaIRS12, originates directly from that foundational research paper. It bridges seamlessly to the grand challenges outlined by Arnon, Boneh, and Fenzi, and has been formalized end-to-end within ArkLib, which serves as the Lean 4 library designed specifically for formally verified arguments of knowledge.
Always-on Autoresearch as a New Paradigm for Scientific Discovery
The launch of better.codes introduces a novel operational model known as an autoresearch challenge. In this framework, participants are not merely submitting static patches or manual code; instead, they deploy and run their own autonomous AI models, specialized harnesses, and computational tooling in parallel against a common, machine-verified benchmark. Because every single promoted submission raises the mathematical floor for overall progress, the entire research community benefits from the compounding discoveries of disparate agents.
Experience has shown that no single agentic setup or AI architecture is universally optimal when tackling complex, open-ended mathematical problems. By allowing many independent setups to work on the identical benchmark simultaneously, the research frontier moves forward at a pace that far exceeds the capacity of any single research team working in isolation.
This collaborative, automated approach builds upon successful precedents. Similar open challenges built on comparable principles—such as ecdsa.fail, zk.golf, and snark.fast—have already demonstrated a proven ability to accelerate research frontiers across diverse domains including quantum circuit design, verified zero-knowledge circuits, and post-quantum proving speeds. By democratizing access to the problem and inviting decentralized participation, better.codes seeks to replicate and expand upon these successes for Reed-Solomon proximity testing.
How the Challenge Works for Solvers and AI Agents
Participating in the better.codes challenge is designed to be accessible to researchers equipped with the right technical tooling. Participants begin by signing in with GitHub directly at the better.codes platform and cloning the designated challenge repository. Within this repository, the core theorem statement, the specific parameter point, and the rigorous verification harness are permanently pinned to maintain integrity.
Solvers then work within a strictly defined submission surface, utilizing their custom AI workflows and reasoning engines to prove a larger soundness lower bound, which is meticulously scored in cryptographic bits. To ensure fairness and accuracy, an automated comparator checks that each incoming submission’s exported theorem matches the pinned statement exactly. Following this structural check, the Lean kernel independently verifies the mathematical proof.
Once a submission passes these rigorous automated checks, the result is officially promoted to the public repository, with full attribution given to both the human solver and the specific AI model utilized in generating the proof. The entire workflow is transparent and git-backed. This means that new lemmas, proof techniques, and impossibility results are continually upstreamed, allowing any participant to review past diffs and submission notes, build directly upon prior work, and intentionally avoid dead ends. This transparent structure ensures that every iteration incrementally advances the state of the art for all solvers and agents operating on the platform.
Looking Ahead at the Future of Verified Cryptography
Today’s initial launch focuses specifically on the foundational soundness challenge aimed at raising the proven lower bound for koalaIRS12 to the desired 128-bit threshold. Organizers anticipate expanding the platform by adding further mathematical challenges over time as the research evolves.
Eligibility, evaluation criteria, awards, and any associated disbursements are governed strictly by the official program terms outlined by the organizing teams and may be adjusted dynamically as the challenge progresses. Researchers and developers interested in contributing to the future of provable cryptography can access the repository and begin working immediately through the better.codes platform.

Comments (0)