Jul 2026
One honest re-run beats a bribed majority
Every consensus system on earth settles disputes by headcount. That's exactly backwards.
Here's the assumption almost every decentralized system quietly runs on: truth is what most of the money agrees on. Get 51% of the stake, the hashpower, the votes, and you own reality. It sounds democratic. It's actually the whole attack surface. If truth is a vote, then truth has a price, and the only question left is whether corrupting the majority costs more than the prize.
Now watch what happens when the thing in dispute is a computation. Someone claims their machine ran your program and got this answer. How do you check? The dumb, boring, correct move is to run it again yourself. If the program is deterministic — same input, same bytes, every time, on every honest machine — then the re-run isn't an opinion. It's a fact. And a fact doesn't care how many people voted against it.
That single move flips the security model inside out. In a vote, you need a majority of honest participants and you're doomed the moment someone buys enough of them. In a re-run, you need one. One honest checker re-executes the job, gets a different answer than the liar signed, and the liar is caught — with a signed receipt, no race to be first, no reorg, nobody to bribe. Flood the network with ten thousand colluding fakes and it changes nothing, because none of them can make a wrong answer reproduce. The researchers who formalized this — Canetti and friends back in 2011, then the whole optimistic-rollup and Truebit lineage after — kept arriving at the same line: the system is safe as long as a single honest party is watching. That is a wildly stronger guarantee than "as long as we out-number the attacker," and it's sitting right there for anyone willing to make their computation reproducible.
So we built the network around the re-run instead of the vote. A provider runs your capsule, signs a receipt pinning exactly what went in and what came out, and gets paid. Any node can pull that receipt, run the same bytes, and check. Cheating isn't caught by consensus — it's caught by arithmetic. And because you only need one skeptic instead of a majority, verification is cheap: sample the honest providers occasionally, check the sketchy ones every time, back it with a bond so lying loses money on average. Proof-of-work burns a country's worth of electricity to buy an ordering nobody asked for. We just run the thing twice.
There's an honest edge here and I'm not going to pretend there isn't — the whole point of this network is that we don't. A re-run only settles a fight if someone actually re-runs, and if the checkers are a small trusted set today, then you're trusting that set until the day the bonds and the referee and the open-verifier market are all wired and paying for themselves. We know exactly where that line is. But the primitive underneath is the part worth stealing: the moment your dispute is a deterministic computation, you don't need a bigger army than the liar. You need one honest re-run, and the truth stops being for sale.
Voting asks who's on your side. A re-run asks what actually happened. Only one of those has a price.