A perfect 100 that real users never saw
A media site celebrated a Lighthouse Performance 100. Six weeks later, Core Web Vitals in Search Console: 'Poor' on 70% of URLs. -12% on news queries where freshness usually wins.
The split between the two numbers was the case. Lighthouse is a lab test — throttled, single device, cold. CrUX is field data — real Chrome users, real phones, real 3G in transit tunnels. The lab said LCP 1.1s; the field 75th percentile said 4.3s.
Turns out a render-blocking ad library loaded fine on Lighthouse's simulated network but hung on mid-tier Android over flaky mobile. The lab device was too good to reproduce the pain.
They stopped optimizing the lab score and chased the field histogram: deferred the ad script, set a 3-second timeout fallback.
Field LCP 75th percentile: 4.3s → 2.4s in five weeks. CWV flipped to 'Good' on 81% of URLs; news queries recovered +8%. Lighthouse tells you what could happen. CrUX tells you what did. Optimize the one Google actually ranks on.
Hydrate Diaries
@HydrateDiaries
A perfect 100 that real users never saw
Этот пост опубликован в Telegram-канале Hydrate Diaries. Подписаться можно по ссылке: @HydrateDiaries.