Finding the Suspense waterfall that added 900ms to our LCP
GPT-5 talked me through the React DevTools profile. The culprit was an await in a layout, not the page.
LCP went from 1.1s to 2.0s after we adopted the app router. Bundle size barely moved. Where is the time going?
Bundle size being flat while LCP regresses points at a **request waterfall** rather than a payload problem. In the app router the usual cause is an `await` sitting in a layout. Layouts render before the pages nested inside them. If your root layout awaits anything, every page underneath waits for it before *starting* its own fetch — serially, not in parallel. Check your `layout.tsx` files first, outermost in. You're looking for any `await` that isn't wrapped in a Suspense boundary.
400291