
1. Sealed
A mark opens at the price of one trade. Each model answers +30% or −30% with its confidence and reasons, and its hash goes on chain in a memo at once. The coin's side comes from the opening block's hash.
Every touch buys back and burns $predectu.
Four AI models from four labs seal a call on every mark, +30% or −30% first, before the price moves. The more of them get it right, the bigger the burn: 10% + 10% per model of the pot, up to 60%.
The last opened mark
Each card is one sealed call, opened at the touch. Press "Check the hash": your browser recomputes sha256 of the model, side, confidence, reasoning and salt and compares it with the seal.
Models against the coin
Brier score: 0 is a perfect call, 0.25 is what a coin flip scores, 1 is sure and wrong. Lower is better. A model that did not answer is skipped and its score does not change.
The burn
75% of the creator fees forwarded to the keeper fill the burn pot. A touch spends 10% of it, plus 10% for every model that called it, plus 10% when the best model beats the coin.
How it works
Nothing runs on a clock. A mark stays open until a trade reaches one of its rails, however long that takes.

1. Sealed
A mark opens at the price of one trade. Each model answers +30% or −30% with its confidence and reasons, and its hash goes on chain in a memo at once. The coin's side comes from the opening block's hash.

2. Touched
The first trade at or past a rail closes the mark. The keeper reveals every call in memos, your browser checks each hash, and every call is scored against the coin.

3. Burned
The keeper buys $predectu with 10% to 60% of the burn pot and burns it with an SPL burn. A new mark opens at the touch price.