In MRs, when a thread is resolved its content is mostly hidden (including its replies). See these examples:
Instead of fetching and loading those replies on page load, we can defer that loading to when the user expands the thread on demand.