How this share is counted
An article counts as open only where an open licence was actually found on it — a Creative Commons or public-domain statement read from the article itself. Everything else counts as not open, including articles whose licence could not be read at all, which across the corpus is roughly two thirds of the peer-reviewed record.
So this figure is a floor, not an estimate. The true share is somewhat higher, by an unknown margin that differs between publishers: an article is missing a licence here because the platform could not reach or parse one, and that is more common for older work, for publishers whose pages the parser handles poorly, and for articles held only as a bibliographic record.
It is counted this way on purpose. Restricting the denominator to articles whose licence could be read looks more complete but is far worse: a subscription journal where only its handful of optional-open articles carry any licence at all would then read as 100% open. Undercounting can be stated and reasoned about; that cannot.
