Performance
Lab data
Also called synthetic.
Lab data is a score from a lab browser with a fixed device and network profile, not from real visitors. It is repeatable and good for debugging.
How it is measured
A tool such as Lighthouse loads the page once with set CPU and network throttling and reports metrics like FCP, LCP, TBT, and CLS. The same page can vary by a few hundred ms between runs, so median several.
It cannot capture real interaction, cookie banners after consent, different locales, or long sessions. INP and the full CLS window are only partly visible.
Worked example
A site gets a Lighthouse performance score of 94 with LCP 1,700 ms on a simulated Moto G throttle. Field p75 for the same page is 3,100 ms. The difference is a hero that is personalized for logged-in users and never appears in the lab.
Running three tests gives 91, 94, and 88, so the team compares medians rather than single runs.
How it differs
Lab data is controlled and repeatable. Field data is real, varied, and slower to change. Lighthouse is one tool that produces lab data. Lab shows why something is slow, field shows who is affected.
Common errors
Treating a lab score as what users get. Comparing runs across different machines. Testing only the home page. Chasing a score instead of the slowest user experience. Running tests on a developer laptop with the cache warm.
In practice
Use the lab to reproduce and fix problems found in field data. Use the same throttling every time and keep the results with dates.