What the latency test actually measures
The lightning icon on the Proxies page checks every node in the current group, and it measures one thing: how long a handshake with the test address takes. Bandwidth, packet loss, whether the line falls apart at nine in the evening. It knows none of that.
Which is why a node reporting 40 ms can stall on a 4K stream while a 200 ms node plays it end to end. Low latency means the line answers quickly. It says nothing about how wide the pipe is.
The default test target is a foreign connectivity-check endpoint. If your network interferes with that one address, every node times out while all of them actually work.
Test against something you actually open
The Test page lets you add your own targets. Point it at the service you use daily and the numbers start describing your experience.
Three numbers that are not the same number
- 1
Latency decides how fast a page starts painting
The gap between clicking a link and seeing content is mostly latency. For browsing, chat and search the ranked list is genuinely useful, and anything under a hundred milliseconds feels the same.
- 2
Bandwidth decides video quality and download speed
You cannot read bandwidth off the list. Take two or three nodes with different latency, open the same clip at top quality on each, and see which one holds it. Ranking and experience are separate things.
- 3
Stability decides whether the stream survives
Retest the same node fifteen minutes later. A line that swings between 60 and 300 feels very different at night than at noon. Check again between eight and eleven, when the route is under real load.
Pick by what you actually do
Decide what the node is for before you pick it. For a fixed daily use, one node chosen by hand is less trouble and easier to diagnose: when it slows down or drops, you know which line is responsible. Automatic groups earn their place when nodes are many, uneven, and some die regularly.
- Browsing and chat: sort by latency, take one of the top few
- Video: ignore the ranking, run ten real minutes at full quality
- Test at midday and again at peak, then drop the ones that swing
Common questions
Every node shows a timeout, but web pages still load. Is something broken?
Usually the test address is unreachable, not the nodes. On the Test page, change the target to a site you know you can open and run it again. If the numbers come back normal, only the default endpoint is being interfered with.
Latency is a few tens of milliseconds and video still buffers. What should I switch to?
A node with more bandwidth, not one with lower latency. A quick handshake tells you nothing about throughput. Within the same subscription the less crowded lines have more headroom, so at peak hours try a few nodes in unpopular regions.