Handling OP Stack Upgrades: The Karst Prestate

Handling OP Stack Upgrades: The Karst Prestate

Conduit has been contributing to the OP Stack since early on, and we run upgrades across every chain we operate. That means when a contract release lands wrong, we see it across many chains, have the operational context and tooling to work out why, and can share findings with the broader community.

During the Karst (Upgrade 19) activation in June, that’s what happened. After the upgrade went live, op-challenger on our chains began refusing to participate in new dispute games. We traced the cause to the absolutePrestate on PermissionedDisputeGame, which was set to an invalid placeholder value instead of the registered kona-client prestate. We reported the finding to OP Labs with the contract state and the challenger behavior we observed, and they posted a workaround in their docs within the hour.

What the absolute prestate does

Every fault-proof game on the OP Stack starts from a known commitment to the fault-proof program (op-program-cannon pre-Karst, Kona-client post-Karst ). That commitment is the absolute prestate. The challenger checks it before defending or attacking a claim, and it will not touch a game whose onchain prestate it doesn’t recognize. That check is a safety feature: it stops a challenger from making claims against a program it can’t execute.

It also means the prestate has to be right onchain. If it isn’t, the challenger stands down, and the games it is supposed to resolve sit there unresolved.

What we found

Karst activated on Sepolia Superchain networks including Mode, Metal, Zora operated by Conduit on 17th June but they made the accompanying contracts upgrade sometime on June 15th. Due to our monitoring systems we detected it on testnets and were able to get it patched before Karst on mainnet.

In the days that followed, the challengers on our chains started logging errors on newly created dispute games. We compared the upgraded PermissionedDisputeGame configuration against what the challenger expected locally, and the absolutePrestate didn’t match; it had been reset to an invalid value rather than carried forward.

After Karst activated, op-challenger couldn’t resolve dispute games on Mode, Metal, and Zora, blocking withdrawal finalization. Unlike other chains migrated to kona-client (gameType 8), these remained on Cannon (gameType 1) without prior notice to Conduit. The incident highlights the need to account for chain-specific configurations during upgrades.

What we did

On June 19, our security team raised this directly with OP Labs, including the specific contract state and the challenger logs. OP Labs published a workaround in the OP Stack docs about an hour later, and chains on the upgrade schedule could apply it before more games accumulated.

Why we watch upgrades this closely

We’ve built up a set of post-activation monitors over years of running upgrades on OP Stack chains, maintaining our own forks, and sitting in the upgrade path for every chain we operate. The challenger refusing games is exactly the kind of signal those monitors are tuned for.

An upgrade artifact can be correct for the chains it was built against and still land wrong on a chain with a different proof configuration, a different fork, or a different operator setup. That was the case here: the Karst artifact was valid on its own, and the mismatch only showed against a live challenger holding the registered prestate. Some issues only show up in production, which is why operating many chains matters for spotting them early.

The only way to know is to watch each chain through activation, and when something breaks, share it with OP Labs so the fix reaches everyone.