The real questions (where and for how long do you cache, what triggers are fetch, how to avoid waterfalls) are something that the library itself is better positioned to answer and figure out. We don't have a particular suggestion there (although the docs mention our recommendation).