The page-type labels are the part I would use first. We run a job board where the same job page template lives in a dozen language folders, so one slow change to that template shows up as a dozen failing URLs that all need the same fix. Right now we run PageSpeed checks from a small script once a week and eyeball the list. Does the classification group pages by template across /en/, /de/ and the rest, so a regression is counted once per page type rather than once per URL?