The AI only translates your idea into precise rules. Once the rules exist, the
backtest is run by a fixed, deterministic simulation engine over real
historical prices — the same rules and the same data always produce the same
numbers. The AI never invents performance figures.
Where the AI stops and the math begins
AI handles interpretation only:- The chat agent that talks to you.
- Translating your English into a structured strategy (entry rules, exit rules, filters, parameters).
- Small language helpers, like turning “since 2020” into a concrete date window or naming a strategy.
- Indicators are real TA-Lib functions (RSI, EMA, MACD, ATR, Bollinger Bands, …) — not AI approximations.
- The simulator replays history one bar at a time, applying your exact rules.
- Trading costs are modeled (see below), and exit reasons are explicit
(
stop,target,trailing stop,time,session close,signal). - Run the same strategy on the same data twice and you get identical results.
What the backtest does — and doesn’t — model
| Modeled | Not (yet) modeled |
|---|---|
| Real historical OHLCV prices | Sub-minute / tick fills |
| Back-adjusted continuous contracts | Order-book depth & partial fills |
| Slippage (default 1 tick per trade) | Overnight margin calls |
| Commission (default 2.50/side full-size) | Real broker fees beyond the modeled commission |
What this means for you
Reproducible
The numbers are an honest replay of your rules over real prices — not a guess
and not a sales pitch.
Yours to verify
Every backtest reports the exact date window it covered, and you can export
the full trade list to check it yourself.