Core Web Vitals Checklist: Before Launch: Item by Item Review

CWV self-check isn’t just about running Lighthouse once. Start by recording baseline values, then review the four core areas—first screen, resources, interactiv…

Insights:Core Web Vitals Checklist: Before Launch: Item by Item Review

This checklist is suitable for use before redesigns, template changes, adding large amounts of images, or integrating new scripts. The order is: "record baseline first, then check pages, then verify deployment."

I. Record the Baseline First

1. Record field data before changes.

Passing criterion: Know the 75th percentile values for LCP, INP, and CLS; know the respective figures for mobile and desktop; identify which template performs worst.

Without a baseline, post-change assessments become mere guesses like 'it seems faster.'

2. Record lab data before changes.

Passing criterion: With fixed device, network, and cache settings, save at least one Lighthouse or WebPageTest report; note the LCP element and major blocking resources.

II. Above-the-Fold & Loading

3. Identify the LCP element instead of guessing.

Passing criterion: Clearly define what constitutes the largest above-the-fold content—hero image, headline, video thumbnail, or lengthy text—and ensure its DOM position is known.

If the LCP element is misidentified, subsequent optimizations to images or fonts may be wasted effort.

4. Ensure above-the-fold images have explicit dimensions and load on demand.

Passing criterion: width/height or aspect-ratio attributes are set; appropriate sizes are used for mobile devices (avoiding 2000px images squeezed into 375px screens); formats prefer WebP/AVIF; above-the-fold images should not use lazy loading.

5. Keep TTFB within acceptable range.

Passing criterion: Static pages leverage caching or CDN; dynamic pages evaluate server-side processing time and database queries; TTFB cannot rely solely on frontend fixes.

6. Eliminate unnecessary blocking resources on the initial viewport.

Passing criterion: Critical CSS is inlined or prioritized; non-critical JS uses defer/async; third-party scripts avoid synchronous execution during initial view; fonts and icons do not block above-the-fold text rendering.

III. Interaction & Stability

7. Font loading must not cause text to remain invisible or flash for an extended period.

Pass criteria: Properly configured font-display; only load the font weights and subsets actually used; fallback fonts have similar line height to the target font; no double flash of the first-screen font.

8. No significant long tasks on the main thread.

Pass criteria: No long task exceeding 50ms before first-screen interaction; third-party chat, analytics, and ad scripts deferred; input event handlers avoid expensive recalculation.

9. Space is reserved for images, ads, and embedded content.

Pass criteria: Images, iframes, and ad slots with explicit dimensions do not push down subsequent content; dynamically inserted content has a placeholder; CLS is checked on both mobile and desktop.

10. Test on both mobile and desktop.

Pass criteria: Run lab checks on both device types; inspect field data grouped by device; do not use desktop scores as a substitute for mobile.

IV. Release & Regression

11. Run a comparison before release.

Pass criteria: Use identical conditions before and after changes; modify only one variable at a time; record differences in LCP, INP, CLS, and request count; allow release only when no metric regresses.

12. Perform regression testing and prepare rollback procedures after release.

Pass criteria: Check RUM or run lab regression within 24 hours post-release; review field data after 28 days; identify which commit or configuration can be rolled back; never adopt a "set and forget" mindset.

If You Can Only Fix Three Things

Start with the LCP element itself, then address first-screen blocking resources, and finally handle layout shifts. INP typically requires separate attention to third-party scripts and event handling—do not mix it with LCP fixes.

Don't Treat a Score of 100 as the Goal

Field data reaching the 75th percentile is more meaningful than a lab score of 100. Stripping content or delaying critical first-screen resources for the sake of a perfect score often harms conversion rates in the end.

Related Reading: