Running Live Algorithmic Strategies at Scale: Signal Delivery and Execution Precision in Production
Introduction
The gap between a backtest's reported performance and what a live strategy actually delivers has two distinct sources. The first is broker execution: what the broker charges in spread, how slippage distributes around the intended fill price, how those costs compound across hundreds of trades. The previous report in this series, the transaction cost analysis, measured that side of the chain on a single strategy over a 9-day window.
This report measures the other side. Between the moment a strategy generates a trade signal on a cloud server and the moment that signal arrives at the broker for execution, there is a signal delivery pipeline. It runs continuously, serves every active live strategy in parallel, and has to remain stable under whatever load the user runs against it. If the pipeline drops signals, delivers them late, or behaves inconsistently across strategies, no amount of broker-side optimization can compensate.
The question this report answers is whether the pipeline holds up at scale. The primary dataset comes from a production EU deployment that ran 13 live strategies concurrently for 29 days and delivered 59,737 signals. A secondary US deployment running 2 strategies over the same window contributed 16,623 signals; that smaller dataset is used in Chapter 3 to decompose where the latency budget goes, and again in Chapter 5 as a second empirical anchor for the broker-confirmed slippage σ, since the two deployments share the same architecture but differ in concurrent load and broker-execution geometry.
Chapter 1 describes the load: which strategies were running, on what instruments, with what trade frequencies. Chapter 2 describes the architecture and what the design predicts the measurements should look like. Chapters 3 and 4 measure how the system performed: latency and execution precision respectively. Chapter 5 translates the latency measurements into the spread-equivalent cost framework introduced in the TCA report, completing the bridge between system-side and broker-side measurements.
The TCA report validated the broker. This report validates the pipeline that gets signals to the broker. Together they account for the full execution chain.