ทำความเข้าใจการให้คะแนน Lighthouse

The Lighthouse Performance score is a 0-100 summary of several lab metrics. It is a weighted blend, not a single measurement, so improving one metric does not always move the score the same amount.

The score is built from a handful of metrics, each with a weight. The biggest contributors are typically Largest Contentful Paint, Total Blocking Time, and Cumulative Layout Shift.

In recent Lighthouse versions the approximate weights are: TBT 30%, LCP 25%, CLS 25%, FCP 10%, Speed Index 10%. Heavier-weighted metrics move the score most.

Every raw metric value is first mapped to its own 0-100 sub-score using a log-normal scoring curve derived from real-world data. The weighted average of these sub-scores is the final number.

The curve is non-linear. Near the fast end, shaving milliseconds barely helps; in the middle, the same improvement can jump your score several points. This is why mid-range pages see big gains from small fixes.

• 0-49 red (poor)
• 50-89 orange (needs improvement)
• 90-100 green (good)

Aim for green, but treat the bands as guidance, not a finish line.

Lighthouse runs in a lab: a single simulated load on throttled hardware. Real users vary by device and network, so lab scores are a controlled proxy, not exact field data.

Variance comes from network jitter, background CPU, A/B tests, and third-party scripts. Run audits multiple times and look at the median rather than a single run.

Lighthouse reports link to a scoring calculator. Paste your metric values to see how each contributes and simulate the impact of a fix before you build it.

A perfect 100 is rarely worth the effort. Focus on getting heavy-weighted metrics into the green and on real user experience over the vanity number.

• Target TBT and LCP first (highest weight).
• Run 3-5 audits, use the median.
• Use the calculator to prioritize fixes.
• Validate against field data when possible.

You learned the Lighthouse Performance score is a weighted blend of metric sub-scores, mapped through a non-linear curve, run in a lab environment. Focus on heavy-weighted metrics, use medians, and prioritize with the scoring calculator.