The Lighthouse score lie
A 100/100 Lighthouse score does not mean a fast site. A 75/100 site can convert better than a 100/100 site. A note on what to actually optimize.
Clients ask for a 100/100 Lighthouse score. We push back. The score is a synthetic measurement on a synthetic device. Real users on real connections experience something different.
What Lighthouse tests
A controlled environment with throttled bandwidth, a specific device profile, no third-party scripts firing as they would in the wild, no user interactions. The score is internally consistent but loosely correlated with what a real user feels.
What real users experience
A specific device, on a specific network, with specific browser extensions, possibly mid-action when the page started loading. None of that fits in a single number.
What we optimize for instead
Real-User Monitoring. Field data from actual visitors via the Chrome User Experience Report or a RUM tool. Median LCP. p75 INP. Real CLS over real sessions.
These numbers usually disagree with Lighthouse by 10-30%. The real-user numbers are the ones that move conversion.
What still makes Lighthouse useful
Spotting regressions during development. Comparing two implementations. Finding low-hanging fruit you missed.
It's a diagnostic tool. Not a target.
The next time someone asks for a "perfect Lighthouse score," ask what problem they're trying to solve. Usually it's "the site feels slow" - and that's a problem real-user data answers, not Lighthouse.