Especially cause it feels like their "other" category is probably scrappers and bots.

Also noteworthy is that Cloudflare isn't about Desktop operating systems only. I don't know how StatCounter operates, but they probably ignore everything that is identified as mobile and Smart TV. Cloudflare is about ANY HTTP connection / request. So its natural that these numbers are quite different, as they are different in their scope anyway.

Liek said, all data is biased, and that's why we sould rely on multiple sources. The absolute values are really hard to get a precise mapping of, but the trends seem to be consistent. There's a recent dip in windows usage in that chart as well

I don't think one data point is better than the other. They are all a small portion of the entire base. Its like looking at different game reviews from different people to get an idea how good it is (I know little odd comparison :D). In example Cloudflare data is only relevant to those communities and sites that can pay for it and don't have a different solution. Cloudflare is also a "recent" phenomenon; not sure how relevant that info is though...

They're both likely biased in different ways, different goals to a point. It sounds like statcounter is making some effort to verify unique devices for determining numbers potentially, but Cloudflare looks like it's just raw requests which is going to bias you towards the noisiest devices. Still useful but I feel like that's a harder sell if you're trying to determine market share specifically. I'm not sure how you'd improve either one without drastically raising your overhead.

'Unknown' is the secret OS all the cool kids are using. But we won't tell you what it is!

A few things can cause that:

• Browsers that now block operating system fingerprinting by default, a privacy feature several major browsers have quietly shipped in the last couple of years (Librewolf and Helium browser do that)
• Bot and AI scraper traffic that doesn’t identify itself the way a normal desktop browser would
• Locked down corporate or government machines with identifying headers stripped out
• Modified or anonymized versions of Windows that don’t report themselves correctly

it means 21% are using some tracking prevention. Probably a lot of Linux users in here. This is good. And then there're Linux users like me who pop one time as Chromium on mac, Edge on windows, firefox on atari...

perhaps something like AI scrapers not properly reporting their UA? It would make sense to me that it's a strong second place.