Where this started

The opening symptom was blunt: a failing Cumulative Layout Shift score on the homepage, with the rest of the site not much better. From there it turned into a genuine investigation — one that kept surfacing new, more specific culprits every time the previous fix landed.

The chain of fixes

IssueFix
CarouFredSel row-wrap on initial paint

white-space: nowrap / overflow: hidden on the <ul>, fixed height + display: inline-block on the <li>

Missing image dimensions sitewide

{image:width} / {image:height} via EEHarbor Channel Images, rolled out template by template

Fixed-height, variable-width carousel thumbnails (mobile-only)Space reservation moved to the container: fixed height + overflow: hidden on .image-overlay, object-fit: cover on the image
RevSlider transition jank on the homepageTransitions and captions switched to fade, min-height added to banner containers
Blank LCP reading in DevToolsExplicit width: 274px added to carousel <li> items, eliminating the runtime inline-width injection that was corrupting LCP attribution

Each of these looked, on the surface, like a variation of the same problem — "carousel causes layout instability." A couple of them turned out to be genuinely different failure modes wearing the same disguise, which is exactly why this took more than one pass to fully resolve.

The part that's easy to forget: the 28-day wait

Search Console's Core Web Vitals report doesn't show live data — it shows a 28-day rolling average pulled from the CrUX dataset, built from real visitor traffic. That means every fix in the table above had a built-in lag before it showed up as a change in the report. PageSpeed Insights lab data and the Search Console trend graph both served as earlier proxies during that gap — useful for confirming a fix was working before the field data caught up to confirm it officially.

Why this matters for client conversations: If a client checks Search Console the day after a fix ships and sees no change, that's expected — not a sign the fix didn't work. Setting that expectation up front avoids a lot of "did this actually do anything?" conversations.

What "all good" actually confirms

Getting every URL into "Good" on both device types means the fixes weren't just theoretically correct — they held up across real visitor sessions, real devices, and real network conditions, averaged over a full month. That's a meaningfully higher bar than a clean Lighthouse run in a controlled test environment, and it's the number that actually factors into how Google evaluates page experience.