Methodology

There are 3 types of providers on the website: with reviews, with overviews and with none of these. First two are actually tested, but to a different degree, for them data is collected from logging in, exploring, buying and using proxies. For 3rd type data gathered only from surface, landing pages and whatnot. One thing worth mentioning - price tables are available right now only in reviews, so filtering by price may be wonky (it still says "from some price", but be aware). Prices for providers without reviews are listed for minimal amount you can buy, not cheapest per provider. When available - I'm trying to use trial, but it's often very limited, so I end up buying plans anyway, especially for re-tests.

Each proxy type is being tested differently. For rotating plans it's 1,000 requests and 100 download tests. For static - either available number of proxies, or, if this number is 1, - 10-100 requests and 10 downloads. All testing is done with US targeting, because it's popular, it's available and it's America, Hell Yeah. 1,000 sequential requests for rotating plans seems to be just enough to have a grasp on proxy success rate and latency. I've tried to do more and it ends up with roughly the same numbers.

Speed is a bit trickier. If you go to speedtest.net or fast.com - your speed is high, matter of fact - it's way higher than numbers on this website. This is because these websites aren't only measuring speed with multiple threads, their tests also have enough time to reach this peak speed of your connection. When using proxy you rarely encounter situations when you can sit and wait for a 100 Mb file to download. That's why speed in my tests is being measured by downloading 1Mb file, think of a web page or an image. It's including connection speed, time to first byte and full file download time. I've tried A LOT of different methods and this seems to appropriately describe proxy speed without eating too much traffic, which, not gonna lie, is nice to have when your budget is negative. On a review page speed described as one of three statuses: bad (under 2.5 mbps), alright (2.5-4 mbps) and good (over 4 mbps). This is done so people won't lose their minds over 30 mbps (very good) of datacenter proxy. You still can get values by hovering speed text value though.

Risk Score is being measured as median value of Scamalytics risk score on selected sample pool with 0 being not flagged by any database and 100 being flagged by them all. Everything under 10 score is considered good, 10-20 is decent and higher is bad. While it's not very definitive for rotating residential and mobile proxies - it's very important for datacenter/ISP proxies where proxies are often have years of same IP. Sample pool often differs from number of requests because of success rate and number of unique IPs in these requests. IPv6 support is being tested by accessing ipv6.icanhazip.com. However it's only available for IPv4 backconnect proxy, right now I have no way to test IPv6 proxy, I have plans for it. When testing UDP - I'm connecting to my own UDP echo server at first and just to be sure - to publicly available servers (there was 1 occurence of a proxy that didn't work on my server, but worked on a public host).

Right now I'm gathering IPs with the same targeting (US) as long as I keep getting unique IPs. To be precise - until rate of unique IPs drops below 5%. That's why all providers in overlap table have different amount of IPs. After gathering I just calculate how many IPs and subnets 2 providers are sharing. Right now only IP overlap is available in the table. Subnet overlap becomes meaningful over a longer time frame, as providers' IPs changing and I haven't been running this long enough yet to make that data reliable. Keep in mind that real number of shared IPs is higher in every case, but how higher is defined by pool size and each service rotation policies. I've tried to estimate pool size and real overlap number with Chao-Sørensen index, but not ready to share estimates for now, especially when raw overlap data is enough to make a decision in many cases. Right now these hidden from table numbers are helping me with another tool - list of proven IP pools, which will be available rather sooner than later.