A page-load audit is a good start, but it won’t tell you why a filter feels slow after the page renders. Here’s a DevTools workflow for separating network, rendering and interaction problems.
Get a trace you can compare
Pick one URL, viewport and user state. Note whether you’re testing a fresh or repeat visit. If you use throttling, keep the same settings between runs. Capture both navigation and the interaction in question when the complaint is about a tap, click or keystroke.
Inspect the document before optimising assets
In Network, find the main document and inspect its redirect chain and timing breakdown. Compare connection time with the wait for the response. Inspect cache headers and any CDN cache status your setup exposes. If the document wait dominates, check backend timing or logs before assuming image compression will fix the initial delay.
Break down a late LCP candidate
Identify the element in the performance trace. For an image, separate four questions: When did the document start arriving? When was the image requested? How long did the transfer take? How long after the transfer finished did the element render? A late request points to discovery or loading strategy. A slow transfer points to sizing or delivery. A late render after an early download points you back to styles, scripting or rendering work.
Record the slow interaction
Reproduce the exact click, tap or keystroke and inspect the main-thread activity around it. Check for work already running before the handler starts, expensive handler work and layout or rendering work before the next visible update. Inspect call stacks before blaming a library or vendor. Isolate one suspect in a test environment, then compare the traces.
Find the cause of layout shifts
Inspect layout shift events and compare them with screenshots around the affected frame. The shifted node may be a symptom: check for an image above it without dimensions, a widget expanding into the flow or a font swap changing text size. Reserve space where the layout needs it and verify at more than one viewport.
Patch, compare, and check for regressions
Write down the hypothesis before making a change. Capture another trace in comparable conditions and see whether the suspected delay changed. Then verify dependent features such as navigation, forms, consent and checkout. If you have real-user measurements, use them to check whether the improvement also shows up outside the lab.
Top comments (0)