Signal sites love to publish latency numbers. Sub-10-millisecond, sub-1-millisecond, tick-to-trade, microsecond this, nanosecond that. Most of the numbers are marketing, because they measure the one piece of the pipeline that looks best. Here is how we think about latency and what we actually measure.
The pieces of the pipeline
A signal takes a journey from the market event that triggers it to your broker's order book. The pieces, in order:
- Exchange publishes a market event (a trade, a quote update).
- Our data feed receives it.
- Our strategy engine evaluates the rules and emits a signal.
- The signal is serialized and pushed onto the WebSocket.
- The signal crosses the public internet to your client machine.
- Your Socket Trader client receives it and hands it to NinjaTrader's ATI.
- NinjaTrader constructs an order and sends it to your broker.
- Your broker forwards it to the exchange.
- The exchange matches or rejects the order.
What we control, and what we do not
We control steps 2 through 4 completely. We control step 5 partially: the public internet is the public internet. Steps 6 through 9 are your NinjaTrader install, your connection to the broker, and the broker's own infrastructure. We can instrument them, but we cannot make them faster.
What we measure
Two numbers we track internally:
- Emit-to-push: time from strategy signal emission to the signal being queued on every connected WebSocket. This is our latency, and it is typically under 5 ms. Beyond this, it is network physics.
- Emit-to-acknowledge: time from emission to the client acknowledging receipt. This includes your connection. It is typically 30 to 250 ms depending on where you are and what your network is doing.
What the numbers mean for a retail subscriber
For strategies that hold positions for 22 minutes or 6 hours or overnight, a 100 ms variance in signal arrival is invisible. For a scalping strategy that holds for 90 seconds, a slow connection is real alpha leakage. We do not currently run signals where microsecond precision matters, because retail connections cannot meaningfully exploit them.
Why we do not advertise a single number
Because a single number is always measured on the easy piece. If we told you latency was 3 ms, we would be telling you the truth about step 4 and hiding steps 5 through 9. The honest answer is a distribution, not a headline, and the useful distribution depends on where you live and which broker you use.
The honest number is the emit-to-push distribution: p50, p95, p99 of the segment we actually own. Anything past the WebSocket is your network and your broker, and no headline figure can speak for those.