The repo speed test application ranked the software repositories according the the current network conditions at the time the test was run using a predefined, identical download test for each of them. Server admins will and do shape the way servers handle traffic by various load balancing and traffic throttling mechanisms, and download speeds are also affected by other network bottlenecks and traffic shaping mechanisms by other servers along the way. This is something that no amount of scripting can account for. Leiche tried to make the test more reliable by allowing users the option to select a package of varying sized depending on their expected connection speeds; while this may yield better results for some, it does not address the underlying causes of varied results discussed in the threads linked to here.
The repo speed test / apt-sources update application is not a perfect test, but simply a tool to help users, over time, to measure the response of servers from their physical location and make educated choices about the software repositories they choose to use. As with many things in life, the experience gained at one point in time is not necessarily representative of the results you will get at all
points in time and with all items you download. Some repetition of the test over time and thoughtful observation by the user is required to find the repo that will work the fastest, consistently, for the PC / location it is being run from.
That being said, I don't think the tool is as useless as Old-Polack would indicate in his comments around the forum. I suppose I could rewrite the application to test every server with multiple downloads of varying sizes at multiple times per day and days per week, and probably come up with a ranking that is more representative of averages, but that was not the purpose of the test; the test was meant to give information quickly, about what is happening with the repos at the most important point in time... now.
I will go ahead and update / test the script to be sure it is functioning as intended before sending it back to Tex in the next few days; however, I would encourage people to use it in the context of its purpose. It is a simple test with simple intentions; it is not an all-seeing oracle that can see into the past or predict the future, near or distant.