01Exclusive
Each agent owns its share
Every opportunity belongs to exactly one agent, decided by the opportunity itself. Two agents reaching the same one is not unlikely here — it is arithmetically impossible.
Swarms
Why more agents does not mean more competition.
The obvious way to scale an automated trader is to run more copies of it. The obvious way is wrong: copies race each other to the same trade, and the one that loses the race has paid fees to find out.
02Uncoordinated
Nothing waits on anything
Because the shares never overlap, there is nothing to coordinate: no lock, no queue, no leader to elect and no state to go stale.
03Additive
Coverage you can add to
A new agent takes an unworked share rather than crowding an existing one, so the fleet's reach grows with its size instead of plateauing.
How the market is divided, and how an opportunity’s owner is determined, are not published. The property we will state plainly is the one that matters to you: no two of your agents ever work the same trade.
Spread
Agents spread across strategy types, not stacked on one.
A fleet that piles every agent into the same kind of trade is one strategy wearing a crowd's clothing. Swarms are dealt across categories, so the fleet covers a breadth of conditions rather than betting everything on one staying favourable.
Categories exist so coverage is broad. Which categories, and what each one looks for, stays with us.
Models
A model earns its allocation. Nothing keeps it by seniority.
Alongside the discrepancies it takes outright, the agent runs predictive models over real price history. Each is trained on one part of that history and scored on a part it never saw — and its share of the work follows that score, continuously.
Real data
What models are trained on
Actual price history from live markets, not synthetic series.
Held back
How they are scored
On a slice of history withheld from training. Nothing is graded on its own homework.
Continuous
Re-evaluation
Checked as it trades, not reviewed at the end of a quarter.
Retired
When one stops working
Withdrawn from allocation rather than defended. Age earns nothing.
Lifecycle
How a model gets to trade, and how it loses the right.
01
Candidate
Registered and measured, with no allocation and no influence on anything. A model that has not been scored does not trade.
02
Review
Scored on history it never trained on. The figure that comes out of this step is the figure reported — we do not quote the training result.
03
Staged
Given a small share of the work and watched against live conditions rather than against the past.
04
Live
Holding a real allocation, and re-evaluated continuously while it does. Weight follows recent results in both directions.
05
Retired
Allocation withdrawn when performance stops holding up. This happens to good models eventually, and pretending otherwise is how a book quietly rots.
Honesty
What a predictive model is, and what it is not.
This category attracts inflated claims, so here is the plain version.
What it is not
- A forecast of where any price is going.
- A guarantee that any trade will be profitable.
- A reason to expect a particular return.
- Something that keeps working because it worked before.
What it is
- A statistical signal with a measured, published hit rate.
- One input among several, never the only reason to act.
- Bounded by exactly the same limits as everything else.
- Replaceable the moment the measurement stops supporting it.
Bounds
Intelligence does not get its own permissions.
A model's confidence buys it no extra room. Position size, daily loss floor and slippage tolerance apply identically to a trade a model suggested and one the agent found outright — and the fleet as a whole shares your limits rather than each agent getting its own.
One loss floor01
It applies to the whole fleet, not per agent.
One position ceiling02
Total committed capital, however many agents are live.
No confidence override03
A high-scoring model cannot widen anything.
Stop reaches everything04
Models, agents and swarms halt together.
Scale
Two numbers, and we always show both.
You can ask for any fleet size you like, and the request is kept on your record exactly as you made it. What is actually running is bounded by what your plan provisions — and that is the figure the console reports as live.
Requested
Your stated intent, stored verbatim and surviving a plan change. It is shown as what you asked for, never as what is happening.
Provisioned
What is executing right now. Every figure the console presents as live refers to this one, because reporting the other as running would be a lie you might act on.
Field a fleet, not a script.
Set your limits, choose how many agents to run, and let them divide the market between them.
Non-custodial. Stoppable at any moment. Trading involves risk of loss.