A sentence goes in.
A checked system comes out.
Five stages sit between an idea and an order on your broker. Each one can stop the strategy, and the last two are the only ones that can move money.
Describe it in plain English
"Buy EUR/USD when the 20 EMA crosses above the 50 on the 15-minute chart, stop 1.5 ATR below entry, take profit at 2R." That is the whole input. No scripting language, no MQL, no visual node editor to learn.
The model turns that into a strategy configuration: symbol, timeframe, the indicators it needs, the entry and exit conditions, and the risk block. It writes data, never code — nothing it produces is executed as a program.
The configuration is validated before you see it
The config has to match a fixed schema, and then pass a second set of checks: every indicator used in a rule must be declared, the symbol and timeframe must be ones we actually carry, the position size must be under the platform cap, and a stop-loss must be present. A strategy with no stop-loss is rejected outright — it is not an option you can switch off.
If a check fails, the errors go back to the model to repair, at most twice. After that the request fails and you see exactly which rule was wrong rather than a strategy that quietly does something else.
Backtest it bar by bar
Signals are computed on the close of a bar and filled at the open of the next one, so the backtest cannot see a price the strategy would not have known. Spread, commission and slippage are applied per symbol, and position sizing runs against the simulated equity rather than a fixed lot.
You get return, Sharpe, Sortino, maximum drawdown, win rate, profit factor and expectancy, plus the full trade list and equity curve. The drawdown is the number worth staring at.
Run it on paper against live prices
Paper mode is the same execution engine, the same risk checks and the same journal — only the broker is simulated. It is the default for every new strategy, and it is where you find out that a backtest which looked clean fires four times a day in a live spread.
Go live, deliberately
Connect your own MT5 account through MetaApi and switch the strategy to live. StockStudios never holds your funds; orders go to your broker under your account. Going live is a decision you make per strategy, and the server enforces that it has been backtested and paper-run first.
What the AI is not allowed to do
The model that writes your strategy has no path to your broker. It produces a configuration, and a separate deterministic engine decides whether anything happens.
It cannot place an order
The execution engine has no import path to the language model — enforced by a test that fails the build if one is ever added.
It cannot skip a stop-loss
An order intent without one is rejected at the risk check, regardless of what the config or the prompt asked for.
It cannot raise your limits
Position size, open positions, daily loss and drawdown ceilings are yours. Nothing the model writes can widen them.
It cannot clear a halt
When a limit is breached the strategy stops and stays stopped until a human resumes it. There is no automatic retry.
Five checks run before every single order
Stop-loss present
No stop, no order.
Position size
Under your maximum, and under the equity percentage cap.
Open positions
Below your concurrent position limit.
Daily loss
Today's realised and unrealised P&L still inside your floor, or trading halts.
Drawdown
Equity still within your drawdown limit of its high-water mark, or trading halts.
Every check, every fill and every halt is written to an audit log with the rule and the indicator values that triggered it. When a strategy does something you did not expect, the answer to "why" is recorded rather than reconstructed.
Build your first strategy