Take a seat

Five flights on the house. A flight convenes six frontier models from rival labs to deliberate one question you actually care about — they argue in rounds, judges from competing labs harvest the disagreement, and you keep a sealed record with the transcript, the costs, and the fault lines they could not resolve.

What makes the record worth keeping: a chat window is ephemeral, has no provable origin, and was never checked by anything independent. This is the opposite of that — and its date is anchored to Bitcoin, so it holds up in front of someone who has no reason to trust us.

Your email Your name, as the record will carry it

This name is written into sealed records permanently — it cannot be deleted later, even if you delete your account. A non-identifying pseudonym is welcome. Choose before you fly.

What "permanent" does and does not mean

Your rooms are private by default. Only you (and the operator) can read them.
You can delete a private record, or your whole account, whenever you want. Deleting the account removes your email, your ledger, and your private records.
What cannot be undone is publishing. If you choose to publish a room to the commons, it stays — because others may branch from it or cite it, and deleting it would break their work. A record is unalterable once sealed: that is what makes it worth anything as evidence.
Your email never enters a record. Records carry the name you chose and an opaque identifier — never your address.

I understand that a sealed record is never edited after the fact — the software has no edit operation, corrections are made by later records, and the seal's hash is anchored publicly so a quiet revision would be detectable by anyone. I understand that anything I choose to publish to the commons stays published — while private records and my account remain mine to delete. How we handle your data →

What you should know before you spend one

A flight takes 5–12 minutes — real models thinking at their own pace, never a fake spinner.
Ask something you genuinely want stress-tested. Real questions get real value.
The models can be confidently wrong together. The record preserves their disagreements precisely because agreement is not proof.
We publish our own accuracy, including the calls we got wrong — see the scoreboard.