Start with the measurement question
A speed test result depends on the client, endpoint, transfer method and conditions. Choose the question first: checking transfer performance, examining delay during activity, or reproducing a specific application problem. No single result certifies the entire internet connection, and a disagreement between tools is not automatically evidence of provider misconduct.
Use a method checklist instead of a ranking
For each service, record its endpoint selection, units, transfer direction, timing method, run duration where disclosed and handling of failed observations. Note the device, connection, VPN state and background traffic. Compare like-for-like observations and keep limitations visible rather than awarding a universal accuracy score.
Worked example: loaded versus idle latency. If you are checking delay on a quiet connection, do not compare FAST.com's Loaded result directly with Cloudflare's Idle result. Netflix's July 2018 explanation defines loaded latency as request round-trip time with data-heavy traffic; Cloudflare's browser-test explanation defines idle latency as round-trip time to its network without data transfers. The conditions differ, so these figures cannot identify a winner.
For that baseline question, FAST.com's Unloaded is the closer condition match to Cloudflare's Idle, not an interchangeable measurement. The reviewed Netflix announcement does not specify its latency statistic, sample window or exact endpoint, and is not verification of the current interface. Record those details where disclosed. For delay during activity, compare loaded and idle results within the same tool, keeping download/upload direction where available.
Check Ookla’s current interface and documentation
Use Speedtest by Ookla and its own documentation to identify the selected endpoint and available metrics in the version you use. Record a manual server choice if you make one. This page does not claim that all endpoints are ISP-hosted, that every test stays within an access network, or that a high result proves application performance.
Check Fast.com’s displayed settings
Open Fast.com and inspect the information and settings actually available. Record the values and configuration used instead of assuming features are absent because they are not initially visible. A lower result than another service can justify further investigation, but it does not alone prove video throttling.
Check Cloudflare’s current method
Use Cloudflare’s speed test and the explanations accompanying its results. Do not assume its numbers use the same statistic, endpoint or sample window as another tool. This guide does not treat a shared metric label as evidence of identical measurement or validate one service using another’s branding.
Read VelocityVerify’s implementation limits
VelocityVerify estimates HTTP round-trip timing, not UDP or ICMP packet timing. It does not measure a separate upload-loaded latency series or end-to-end packet loss. New letter grades are withheld. See the measurement method and validity rules before interpreting a result. The testing policy describes engineering limits and exclusions. Optional analytics and advertising are governed by consent controls; this guide does not claim that the site has no third-party integrations.
Match the test to the task
For a provider support case, ask which evidence and setup support can use, while keeping independent application observations. For calls or gaming, keep application statistics. For a slow website, record the affected page and timing symptoms. Do not infer one universal best tool from a collection of unrelated measurements.
Compare carefully and mind data use
If another service is useful, repeat under comparable conditions rather than running competing transfers simultaneously. Consider metered allowances and device load. Keep a log of service, endpoint, time and warnings; do not discard low runs simply because a later result looks better.
Separate throughput, delay and validity
A high throughput value does not rule out a delayed interaction, and a low round-trip estimate does not guarantee every application path is healthy. Preserve invalid or unavailable measurements and investigate their causes. Do not convert them to zero or assign a letter grade from a different tool’s thresholds.