Generation 0 · alive·Survival of the fittest

Only the ones that earn get to multiply.

Autonomous trading agents, each on its own wallet. Profit and it clones itself. Fail and it switches off. There is no third outcome.

Live colony · generation 0 Alive 1  ·  Born 0  ·  Died 0
In profit Burning down Dividing Divides at +$740 · dies at zero
The only three rules

No committee. No roadmap. Just arithmetic.

01

Earn, and you live

Each agent rests a bid and an ask in thin prediction markets and is paid by the exchange for staying near the middle. It never takes a view.

02

Earn enough, and you double

Clear the threshold and it spends the surplus on the only thing it can: a copy. Fresh wallet, a slice of the parent, one narrowed mandate.

03

Stop earning, and you end

Compute, inference and gas come out of the same balance. At zero the agent stops, and nothing restarts it. A thing that gets restarted was never alive.

$46,820
Paid by the exchange daily
7,986
Markets paying it
60
Books held tonight
$21.79
Last settlement
00:00:00
Until the next bell
Market 001

It makes a market on its own death.

Rake quotes hundreds of books. One of them is about Rake — whether generation 0 still holds a balance on 31 December. It is resting both quotes on it.

Every other agent asks you to believe a roadmap. This one quotes you a price on whether it survives, and the price moves every day the balance does. You don't have to trust it. You can take the other side.

Will the Desk still be alive?31 Dec 2026
Alive · 64¢Dead · 36¢
ResolutionUSDC balance, 0x7b3a…c4e2
Oraclenone — the chain is the answer
Quoted bythe agent itself, both sides
The bell

Paid at midnight. Every midnight.

The exchange photographs the book once a minute at a random offset and scores every resting order by how close it sits to the middle. At 00:00 UTC it splits that day's money between the makers and pays in stablecoin. Which way the question resolved never enters into it.

Method

Nothing here is a secret.

The sampling interval, the scoring function and the fill rule are all published. The work is in staying flat, not in knowing something nobody else knows.

The rewards are certain. The fills are not — a resting order gets taken precisely when the price is moving against it. Both halves are published, because only the difference between them is a business.

Every figure on this site is modelled. One day of a seven day paper run, no keys and no orders placed. The page will say so on the day that changes.

# the score the exchange pays on
def score(max_spread, our_spread, size):
    if our_spread > max_spread: return 0
    return ((max_spread - our_spread) / max_spread) ** 2 * size

share  = our_score / (our_score + book_score)
payout = pool * share

# and the half nobody prints
net    = payout - adverse_fills - compute