With no latency, an order fills on the very print that triggered it: the strategy sees a price and trades at that same price, which nothing real can do. Network Latency puts a delay between the signal and the fill, so each order meets the market a moment later, as a real one would.
Where to set it
Under Order Execution in Account & Risk:
- Static adds the same delay to every order, in milliseconds. The default is none.
- Random draws a delay for each order, anywhere between a minimum and a maximum.
Either way, delays run from 0 to 1,000 milliseconds. The Realistic Environment profile sets a random delay of 100 to 300 milliseconds, with a typical commission.
What the delay does
An order fills on the first print at or after its signal's time plus the delay. In a busy market that is almost exactly the delay you set; in a quiet one the next print can come later, so the real delay is longer. Each trade records the delay it actually got, shown on the Trades tab.
The delay applies to entries, and to exits a rule triggers: a stop, a target, an exit condition or a time exit. A close the run forces, at the end of the Daily Time Window, by a daily limit or the drawdown limit, by a regime change, or when the data ends, fills on the print that forces it.
While an order is on its way:
- No other entry fills. An entry in flight blocks the next one.
- It still fills, even if the session closed or a limit was reached after the signal; the trade is then closed on that same print.
- With Book fills, a delayed entry can be skipped if the spread widens past normal before it arrives, and a delayed target still fills at the target's price.
At the end of the data, an order still in flight fills on the last print. With Parallel on, that happens at the end of every month.
How it changes results
Fills land on later prints, so prices move in both directions: a breakout entry tends to pay more, a stop can fill further away. A strategy that trades on tiny, fast moves usually suffers most, and one that holds for hours barely notices.
Runs with latency made before September 27, 2026 used an older model that looked a little ahead; running them again can change their results.