Condition Evaluator

The C++ engine evaluates conditions the same way in a backtest and on the live engine. Live, it appends each closed bar to the shared feed and runs the same bar step. A difference between a backtest and live results therefore comes from the data, the timing or the costs, never from the conditions being evaluated differently. The engine's own tests run every DSL bot both ways and compare the exits.

When a strategy is loaded the engine collects every operand it uses (indicator, parameters, output, timeframe, candle type) and computes each distinct series once over the whole history. Two conditions on EMA(20) share one series; several strategies on the same symbol and timeframe share them too. Evaluating a condition on a bar is then a few array reads and one comparison.

Both sides of a cross are read on both bars, so crossing a moving line such as Close crossesAbove EMA(20) works. An operand with an offset or a higher timeframe is read the same way: the previous bar of the signal's timeframe.

A cross compares this bar with the previous one and is true only on the transition. That one bar of memory is the whole difference between crossesAbove and >, and it decides how often a strategy trades.

crossesAbove — what "once" means
────────────────────────────────────────────────────────

  Bar N-1:  left = 28   right = 30    left <= right   (at or below)
  Bar N:    left = 32   right = 30    left >  right   (above)
  ⇒ fires on bar N

  Bar N+1:  left = 35   right = 30    still above
  ⇒ does NOT fire; there was no fresh crossing


RSI(14) crossesAbove 30 — a real sequence
────────────────────────────────────────────────────────

  bar  97    RSI 27.3    below 30
  bar  98    RSI 29.1    below 30
  bar  99    RSI 31.8    crossed        ← fires
  bar 100    RSI 34.2    still above
  bar 101    RSI 28.5    back below     (arms the next cross)
  bar 102    RSI 32.1    crossed        ← fires again

• The previous bar must be at or below the level and the current bar strictly above it. Landing exactly on the level does not count as crossed; starting from exactly the level does.
• A cross cannot fire twice without going back. If the value stays above the level for forty bars, that is one signal, not forty.
• On the first bar, and on the first bar after an indicator's warm-up, there is no valid previous value, so no cross can fire.

• Entry signals keep being evaluated while a position is open. With the default maxPositions of 1 an entry signal while holding does nothing. With a higher limit each entry signal that acts opens another position, at most one per bar; re-entry rules (cooldown, grid, DCA, pyramid, a maximum count) restrict when those adds are allowed.
• Stops and targets are checked on every bar against the bar's high and low, separately from the exit signals. A stop still protects a position whose exit condition never becomes true.
• A position opened on a bar is first checked against its stop and targets on the next bar. Exits run before entries, so a strategy can exit and re-enter on the same bar.

An indicator has no value until it has enough bars. During warm-up its value is NaN (no value, not zero), and any condition reading it is false. The same applies to an offset that reaches before the first bar and to a higher-timeframe operand before its first closed bar. The strategy simply does not trade until every series it needs has a value.