01
48h
Bad idea
Someone posts a primitive that sounds indefensible. If nobody can explain why it fails, it advances.
YEET LABSAI
tested before launch
We build new onchain primitives — designs with no track record yet — and test them hard before anything ships: spec the invariants, then use AI to hunt for the ways they break. Some work. Most don't. We publish what we tested and what we found, either way.
Everything we've shipped, live and dead. Status is honest or it's worthless.
The board is the summary. The floor is the real thing — every protocol gets a machine, and the empty bays are what we haven't built yet.
Enter the arcade →Four gates. Anything that stalls at one of them dies there.
Someone posts a primitive that sounds indefensible. If nobody can explain why it fails, it advances.
Invariants first — what the protocol must never allow. Then we run AI-assisted adversarial testing against the code, permissions, economic assumptions, and edge cases. We publish the attacks attempted, findings discovered, fixes made, and risks that remain.
Testnet, then mainnet with a cap. Real money, small money. Contracts verified, keys public, no private beta.
It earns its cap increase or it gets wound down on schedule. No zombies, no quiet abandonment.
Three rules. They're the only thing here that isn't experimental.
Deploy beats debate.
A live contract with $10k in it teaches more in a week than a forum does in a year.
Break it before you scale it.
Every experiment gets tested before launch, capped while it's live, and scaled once it holds up. A cap limits the damage. It doesn't replace the review.
We eat first.
Team capital goes in at genesis, at the same terms, with the same lock. Always.
—
Rule four is whatever the last failure taught us. It gets written after, not before.
Genesis notices, kill notices, and post-mortems. Roughly one send a month, nothing else.