How far we've come that near-real-time multidimensional dynamic views onto multi-GB data stores is considered 'a pure document' by someone. Why, just because it uses links and URLs and instead of buttons and fetch?
No, you've assumed the question. "Applications are much harder to cache and start up fast" because you're defining "applications" to intentionally exclude architectures which are cachable and fast to start.
If I shipped a git repo viewer as an executable, nobody would be asking why it's not rather a txt/doc/PDF. It's unquestionably an application, not a document.