They say they use publicly available information about sites, plus scraping metadata from the sites, but do not reveal the blend between those sources.

If they're relying primarily on scraping (which I have to think they are, given the scale of the operation otherwise), they're really measuring "what proportion of sites make use of languages which we can easily fingerprint in their stacks".

What I think this is telling us is that WordPress is popular, that WordPress is written in PHP, and that PHP is profligate in its use of headers that announce its usage. That doesn't seem very interesting.