The measured cost and gain of keeping content off the boot critical path.
This site's desktop shell does not wait on a CMS read, and the reason is a measurement rather than a preference. With a CMS that accepts connections and never answers, the first byte arrived after 5014 ms while the reads sat inside the shell; moved behind a Suspense boundary, 14 ms. In the first case the visitor stares at a blank page for five seconds; in the second the boot animation starts at once and the content streams in while it plays.

Re-measured in every phase
What the boundary is worth is not one figure but the gap between those two columns. It has been re-measured under the same conditions in every phase that added a window body, and the median stayed between 8.9 and 11.5 ms: six window bodies, a mobile shell and a set of SEO routes have all left the first byte alone. The measuring taught its own lesson too. Machine load can exceed the difference being measured, so single-run numbers are never compared with each other; pooled medians are read against raw ranges.
Tags
- Next.js
- React