Strategy anatomy

Every strategy the AI assembles is a structured object with a fixed set of fields. This page explains what each field means, how the platform uses it at runtime, and what the valid range is.

Understanding these fields is the fastest way to get from a first backtest to a well-calibrated live strategy.

Identity & classification​

Entry conditions​

The entry block defines when the strategy opens a position. All conditions must be true simultaneously on a completed candle close.

Example entry condition set:

RSI(14) above 60
EMA(9) above EMA(21)
Volume above 20-period average

All three must be true on the same candle close for a position to open.

Signal plan structure​

Internally the AI constructs a SignalPlan that the execution engine evaluates:

interface SignalPlanEntry {
  name: string;                          // indicator name, e.g. "RSI", "EMA"
  params: Record<string, unknown>;       // indicator parameters, e.g. { period: 14 }
  signal_type: 'TRIGGER' | 'FILTER';    // TRIGGER fires the trade; FILTER must also pass
  timeframe: string;                     // candle resolution for this indicator
}

interface SignalPlan {
  entry: SignalPlanEntry[];
  entry_condition: string;               // logical expression, e.g. "A AND B"
  exit: SignalPlanEntry[];
  exit_condition: string;
  signals_available: number;             // total indicators available in the KB
  signals_used: string[];                // which signals this strategy uses
  kb_signals_available: number;          // knowledge-base indicator count
  kb_signals_used: string[];             // KB-sourced signals used
}

TRIGGER signals are the primary fire condition. FILTER signals must also be true but alone cannot open a position. A typical strategy has 1-2 TRIGGERs and 0-2 FILTERs.

Exit conditions​

The exit block controls when an open position is closed. The platform checks exit conditions on every candle close and on every price tick for stop-loss and take-profit.

How trailing TP works:

When trailing_tp is true, the take-profit only arms after price has moved arms_at_pct in your direction. Once armed, it trails trail_pct below the peak price seen since entry. The position closes when price falls back to the trailing level.

Example: entry at 100, arms_at = 1.5%, trail = 0.8%.

• TP arms when price hits 101.5
• At that point the trail level is 100.69 (1.5% - 0.8% from entry)
• If price peaks at 102, the trail level rises to 101.18 (102 x (1 - 0.008))
• Position closes when price falls to 101.18

Multi-target take-profit (TP1 / TP2)​

Instead of a single all-or-nothing exit, a strategy can take profit in stages - closing part of the position at a first target (TP1), more at a second (TP2), and so on. This locks in gains early while leaving a portion running for a larger move.

Example - a two-stage exit:

• At +1.5%, half the position is sold - profit banked, risk on the trade halved.
• The rest runs to +3.0% (TP2) or until an exit condition / stop fires first.

Multi-target take-profit combines with the other exit mechanics: the stop_loss_pct still protects the whole position, and trailing_tp can manage the final remaining slice after the fixed targets are hit.

Sizing & risk​

How position size is calculated:

position_size = (capital_allocated × per_trade_risk_pct) / stop_loss_pct

If this would exceed max_position_size_pct, the position is capped at that percentage instead. This means the actual loss on a stop-out could be less than per_trade_risk_pct.

Trading window​

Status fields​

These are set by the platform, not the AI, and reflect the current state of the strategy.

A strategy can be in both deployedPaper and deployedLive simultaneously - those are independent paths.

Next​

• Run a backtest ->
• Deploy to paper or live ->