Ethereum Researchers Race to Close zkEVM Proof Gap Before December

Quick snapshot: the tug-of-war on koalaIRS12

Think of the current situation as a nerdy tug-of-war. On one side, proof builders are trying to push a safety number higher; on the other, attack-minded researchers are trying to pull a different number down. Right now the contest scoreboard for the benchmark called koalaIRS12 shows a lower certificate at about 63.99 bits and an upper certificate at about 116.13 bits — leaving an awkwardly sized no-man’s-land of roughly 52.14 bits. That gap is where the research drama lives.

KoalaIRS12 is a fixed parameter instance used in reductions for Reed–Solomon proximity checks — basically a focused lab test for part of hash-based SNARK/zkEVM security plumbing. The contest around it makes that test reproducible: every submission must export a theorem, an automated comparator checks it vs. the target, and a formal proof kernel verifies the math before a result is promoted to the public leaderboard.

Why this matters (and what needs to happen by December)

There are two distinct lanes in this contest. The soundness lane raises the lower certificate by proving that the reduction error at a certified radius meets the encoded target. The attack lane lowers the upper certificate by formally exhibiting an unsafe suffix under the benchmark’s density condition. Each successful submission tightens one side of the interval, and because the theorem environment is pinned, successive results remain comparable — think of it as repeated measurements using the same ruler.

The broader roadmap sets a pretty clear bar: a 128-bit provable-security target, a final proof size goal (300 KiB or less), and a formal soundness argument for the recursion architecture. Hitting a 128-bit lower certificate for koalaIRS12 would settle the soundness question for that exact parameter point, but that wouldn’t be the end of the story. Production-grade claims need more than a single benchmark: they require component-level accounting, proof-size audits, a documented recursion topology, clear composition arguments showing how parts combine, and evidence that the formal specs match the actual implementations.

So why the fuss? Because the live 52-bit interval is a concrete, checkable measure of unfinished work for this reduction. Every promoted proof nudges the boundary and improves the public evidence researchers and engineers can use when stitching component analyses into a full-system security case. But for now, these moves are squarely research-facing: they change what we can say about a particular parameter point, not what nodes on the mainnet will accept as consensus validation.

In short: the leaderboard is a useful, machine-checked research playground that helps focus effort toward an early-December deliverable. Whether this leads to production-grade zkEVM security depends on teams tying up the remaining bits, shrinking proofs, documenting recursion, and proving that the theory faithfully maps onto live code. Expect a lot more activity — and a fair number of heated math skirmishes — between now and the deadline.