CodeParkLab data comes from a test you control. Field data comes from real users. They often disagree, because real users have very different conditions.
Lab data comes from a controlled test. Examples are Lighthouse, or the DevTools Performance panel, loading a page on a fixed device, network and location. You can repeat it and get similar results. It is good for debugging and for catching regressions before release.
Field data is also called RUM (real user monitoring). It is collected from real visitors, for example by the Chrome UX Report or your own web-vitals script. It reflects real devices, networks, caches, locations and user behaviour. Google uses field data to assess Core Web Vitals, at the 75th percentile.
Why they disagree:
- A lab test is one load of one page, under one set of conditions.
- It usually has a cold (empty) cache.
- Nobody interacts with the page.
So Lighthouse can't measure INP (Interaction to Next Paint: how fast the page responds to input). Instead it uses TBT (Total Blocking Time) as a stand-in.
Use field data to find problems, and lab tools to diagnose them.
- Lab data is controlled and repeatable, so it is good for debugging.
- Field data comes from real users, and it is what Google assesses.
- Real users have different devices, networks, caches and behaviour.
- Lighthouse uses TBT as a lab stand-in for responsiveness, because it can't measure INP.