the split between documents and apps explains a lot of the pain here. if a page mostly presents content, hijacking selection, scroll, or copy should be treated as a regression