The methodology is designed to accurately measure the maximum available bandwidth of an Internet connection. To avoid per-socket limitations, the test uses multiple parallel data streams (threads/sockets) within a defined time window. The transferred payload consists of high-entropy, randomly generated data to prevent network compression, and validation checks are performed to ensure compression is negligible. Test parameters (variables and constants) are set before or during execution, within defined ranges depending on implementation and real-time network conditions.

For server selection, major ISPs/MNOs use dedicated on-net servers with sufficient capacity, while smaller operators are measured against a neutral in-country server if their on-net server is unavailable. If multiple servers exist, the test selects the server with the lowest ping prior to each test; servers that time out are excluded.

The SDK-based speed test (Android 7+ / iOS 9.3+) performs downloads over HTTP, dynamically selecting 2 or 10 threads depending on connection capability and using large files to saturate the link for the entire test. Performance is continuously monitored, with results sampled every 100 ms. After completion, the average speed is computed both from samples and from raw measurements, and the higher of the two averages is reported as the final download speed.