SquishPixel Blog · Core Web Vitals · Web performance · SEO

Why lab scores and CrUX field data can differ when page loads are incomplete

A DEV Community article looks at why lab tools such as PageSpeed Insights and Lighthouse can show a good result while Chrome User Experience Report (CrUX) field data stays in the amber range for the same URL. CrUX collects results from many real Chrome visits. A lab run is a controlled test that waits for the page to finish loading.

The article uses "partial page loads" as shorthand for real visits that do not match a finished, foreground, full-page load. These include visits abandoned early, pages opened in a background tab, restores from back/forward cache (a browser feature that brings back a page instantly), and single-page apps where the user thinks they moved to a new page. According to the article, CrUX does not publish a metric called partial session. Some of these visits still add data, and others are filtered out.

It gives several examples. Once a user taps, scrolls or presses a key, browsers stop counting later paints as new Largest Contentful Paint (LCP) candidates. Real users often interact early, so field LCP can differ from a lab run that sits idle. Pages loaded entirely in the background follow different LCP rules. Route changes in single-page apps may still be credited to the first document URL in classic CrUX Core Web Vitals reporting.

The article also says that LCP is measured for back/forward cache restores, but Time to First Byte (TTFB) in CrUX is collected on full page loads only. So the two metrics can describe different sets of visits on the same site. It adds that a URL with no CrUX data yet is a separate problem from partial loads.

Why it matters

If you report on site speed, whether to clients or your own team, a lab score and a field score are not the same test run twice. According to the article, treating them that way makes normal differences look like errors, and it can lead to theme or plugin changes aimed at a problem that is not there.

It also affects how you set expectations. The author says a clean lab result after a deploy does not mean field data will improve the same week. Field data lags and reflects a mix of visit types. Labelling which measure you are showing, and why they can differ, makes results easier to explain and defend.

Source: DEV Community. This brief summarises their reporting; read the original for the full detail.