Core Web Vitals are lab numbers

When a crawl finishes, this product takes a lab performance run of up to 5 of its pages, once. These are synthetic measurements taken by a machine on its own connection — not what your visitors experienced.

What is on the tab

• Largest contentful paint: how long the biggest thing above the fold took to appear.
• Cumulative layout shift: how much the page moved about while it was loading.
• Total blocking time: how long the page was busy enough to ignore a tap.

Each is banded against Google’s own published thresholds, which are a published model and not a measurement of your pages. The bands are labelled as such on the tab.

There is no interaction to next paint, and that is not an omission

Interaction to next paint is a field metric: it needs a real person on a real device doing something to the page. A lab run has nobody to measure, so there is no interaction to next paint here and nothing on this tab claims one. Total blocking time is the lab stand-in for the same question, and the column says so rather than being labelled as the field metric it is not.

Why the crawl layout figure is not mixed in

A crawl’s own fetch of a page also produces a layout shift figure, free, and it is deliberately left off this tab. It is a different measurement taken by a different method, and one chart holding two instruments leaves a reader with no way to see that two columns came from different calls.

Which pages, and what a blank one means

Which pages are sampled

The crawled pages with the most Search Console clicks, with ties broken by address so a second run picks the same pages, and never a page the crawl could not fetch cleanly. Without a Search Console connection every page ties at nothing, so the sample falls back to address order — which is arbitrary rather than representative, and worth knowing before reading a sample as a verdict on the site.

What a blank row means

One page’s measurement can fail while the others land. The tab tells apart a crawl nobody has sampled yet, a sample where every call failed, and a sample where some did — three states a blank table would make one, and only the first of them means the work has not been done.