Why your Core Web Vitals take four weeks to move
Field data is a rolling 28-day window. Ship the fix, then wait — and stop judging a rebuild by launch week.
- rolling field window
- 28 drolling field window
- the visit Google scores you on
- p75the visit Google scores you on
Two numbers, two different questions
The lab score answers "how does this page behave on one throttled phone, right now". Field data answers "what did real visitors experience over the last 28 days". Only the second one is a ranking signal, and it is an average of a month that mostly happened before you shipped.
So the morning after a rebuild, the lab number jumps and the field number barely moves. Nothing is wrong. You are looking at 27 days of the old site and one of the new one.
The fix is instant. The evidence is a month behind it.
What to watch in the meantime
- 01Your own analytics, not the dashboardReal user timings move the day you ship. That is where a rebuild proves itself in week one.
- 02The p75 visit, not the averageGoogle scores the 75th-percentile visit. An average hides exactly the slow quarter it is judging you on.
Plan the calendar around it
Ship performance work at least a month before the season you need it for. A rebuild that lands the week of Black Friday gets graded on the traffic it was built to survive, using data from before it existed.