What a Lighthouse score does not tell you
A perfect lab score and a slow site are entirely compatible. Here is how we instrument real-user performance instead.
Marcus Ainsley
Principal Engineer
Lighthouse runs on a simulated connection, on a machine that is not your customer's, once. Real-user monitoring runs on every device that visits you, forever. Only one of those tells you whether the site is fast.
The gap usually shows up in two places: third-party scripts loaded after the audit, and images sized for the developer's laptop rather than a 390px viewport.
We ship a performance budget with every build and wire it into CI. A pull request that pushes JavaScript past the budget fails, which turns performance from a launch-week push into a constraint the team designs within.
Keep the lab score. It catches regressions early and cheaply. Just do not confuse it with the experience your customers are having.