I'm working on improving a website's Lighthouse performance score, but repeated tests can vary by 5–10 points, sometimes landing anywhere from about 39 to 58. That makes it hard to tell whether a change genuinely helped, had no effect, or made things worse. What's the best way to run and compare Lighthouse tests so the results are meaningful?
2 Answers
Lighthouse is naturally noisy, so don't rely on one run. Test the page several times under the same conditions—ideally with a consistent Chrome version, device profile, and throttling settings—and compare the median result rather than the best or worst score. Running with extensions disabled also removes some background interference.
The overall score combines several different metrics, so it can hide what your change actually did. Track the specific metric you were trying to improve—such as LCP, TBT, or CLS—in milliseconds before and after the change, and check the diagnostic details to confirm that the relevant metric moved.
Median is usually better than an average because one unusually fast or slow run won't skew the comparison. CPU throttling and network variability can still cause movement even in an otherwise clean test.