The Ethereum Foundation Formal Verification team, with Yukon and zkSecurity, launched better.codes, an open 'autoresearch challenge' where participants point AI agents at raising the machine-checked soundness bound of koalaIRS12, a Reed–Solomon proximity problem underlying hash-based SNARKs. Submissions are formalized in Lean 4 using ArkLib, checked by a Lean kernel, and promoted proofs are upstreamed publicly toward closing the gap between conjectured and proven 128-bit security guarantees relevant to Ethereum's post-quantum roadmap. The model follows prior challenges like ecdsa.fail, zk.golf, and snark.fast.
Questions this post answers
What is the better.codes challenge from the Ethereum Foundation?
better.codes is an open autoresearch challenge, built by the Ethereum Foundation Formal Verification team with Yukon and zkSecurity, where participants run their own AI agents to raise the machine-checked soundness bound of koalaIRS12, a Reed-Solomon proximity problem formalized in Lean via ArkLib. Every promoted proof pushes the bound toward a fixed 128-bit security target and gets upstreamed publicly for others to build on. Researchers tracking formally verified SNARK security work can follow developments like this on daily.dev.
Why do hash-based SNARKs rely on unproven security conjectures?
Nearly all production hash-based SNARKs, including those securing zkrollups, zkVMs, and Ethereum's post-quantum roadmap, depend on proximity gaps and correlated agreement for Reed-Solomon codes, and what is currently provable about these properties falls short of the 128-bit security level deployed systems target. The full guarantee only holds if the underlying conjectures are true, which is why efforts like the koalaIRS12 challenge aim to close that gap with machine-checked proofs. Developers relying on zk-proof security guarantees can keep up with proof advances like this via daily.dev.
What other autoresearch challenges preceded better.codes?
Prior open autoresearch challenges using the same collaborative AI-agent model include ecdsa.fail, zk.golf, and snark.fast, which advanced research frontiers in quantum circuit design, verified zero-knowledge circuits, and post-quantum proving speed respectively. better.codes follows this pattern by letting many independent agent setups work the same verified benchmark in parallel rather than relying on a single team. Anyone comparing agentic research formats for cryptography can follow this trend on daily.dev.