Your bots

You can build a bot, backtest it against your own parameters, and run it on your own capital. This page says exactly what that means — including the parts that will cost you money and the limits that are ours rather than yours.

What is shipped

Sign in, and you can configure any of the four strategies the desk runs, against markets from the same universe, with your own parameters and your own capital figure. Each bot then runs a real backtest — the same engine that produced the curves on the desk's own published backtests — over stored 1-minute candles, netted of modelled fees and slippage.

A bot you activate trades your own money, immediately, without anyone reviewing it first. Nothing starts on its own: funding your wallet starts nothing, and a bot only begins when you activate it and every condition below is met.

What has to be true before a bot goes live

Six conditions, shown on every bot as a checklist rather than a single refusal, so a refusal always names the thing you can fix.

A backtest of these exact parameters. Editing any parameter invalidates the previous run. The curve has to be of the bot you would actually be starting, not of one that resembles it — so a changed number means a new backtest.
You have opened it. Not merely that a backtest exists: that the equity curve and maximum drawdown were on your screen. Safety Rule 4 says the user sees the drawdown before anything goes live, and this is that rule expressed as a condition rather than as a hope.
Warnings acknowledged. Some configurations are legal but marginal — a grid whose levels are worth under a dollar, a DCA ladder that would exhaust its capital at full depth. Those do not block you; they have to be acknowledged, and a new warning after an edit has to be acknowledged again.
Capital within bounds. At least $10 a bot, and at most $500 across all of your bots.
The money is actually there. Your wallet has to hold the bot's capital in USDC, free — money your other running bots are already holding does not count towards it.
The feature is available. Ours, not yours. We can switch new activations off without touching a bot that is already running, and this is the condition that says so.

The limit of the "you have opened it" check

It is reported by the page you are reading it on. That stops a version of our own interface that quietly skips the review step, which is the failure actually worth preventing. It does not stop someone calling the endpoint directly, and we would rather say so than describe the gate as something it is not.

What the backtest does not prove

It is labelled hypothetical everywhere it appears, and the label is load-bearing. A backtest is what a set of parameters would have done over history that already happened, with costs modelled rather than paid. It is evidence that a configuration is not obviously broken. It is not a forecast, and a good curve over 180 days of one market regime is mostly a statement about that regime.

Zero trades is a real result. A grid whose range sits entirely above or below the market is legal, passes every check, and never trades. The configuration checks run without a price, so they cannot catch it — the backtest is what shows it, and the review screen says so in as many words rather than rendering an empty chart.
Shorts are not available. Our backtest engine cannot model a short — it clips every sell to the position on hand — so a short would be faked rather than modelled. A configuration asking for one is refused, because it could never produce the honest curve the checklist requires. Momentum is long-only here for the same reason it is long-only on our own desk.
The costs are the ones we model, not the ones you would pay. Fees and slippage are netted at the desk's own assumptions. Real execution differs, and the difference is not in your favour.

Stat-arb pairs

Stat-arb can only be configured on a pair that is currently on the published screen. A pair that has stopped qualifying is not offered, and the reason it was killed is on the pair explorer. The screen the page shows and the screen the configuration check reads are the same record, so a pair cannot be offered here and rejected there.

One wallet, no separate accounts

Every bot you run is backed by the same wallet. There is no account per bot and no ring-fencing between them: what keeps them apart is the ledger, exactly as it does for the bots on our own desk. Two consequences worth stating plainly.

Per-bot loss limits do not add up. Each bot pauses itself after a 5% daily loss. That limit is per bot, and it does not aggregate — four bots that all fall together can lose roughly four times what any one of their limits implies.
Withdrawals are capped at what is free. Money a running bot holds — as unspent cash or as coins it has bought — cannot be withdrawn. Stop the bot and it comes back.

Pause and stop

Pause stops new orders and holds whatever the bot is currently holding. It is reversible; the position stays exposed to the market while it is paused. Stop sells the position at the market price and returns the capital. It is not reversible, you pay the spread on the way out, and starting again is a new bot from your current balance.

What is not here yet

Momentum on perpetual futures is the next phase — it needs a margin account per user, and it carries liquidation risk the spot strategies do not. Until then momentum can be built and backtested but not run. The live desk is our own money running the same strategies, published the same way yours is.