CrUX vs Lighthouse: 28-day lag vs single sample

CrUX vs Lighthouse: 28-day lag vs single sample
Field data (CrUX = Chrome User Experience Report, real visitors) and lab data (Lighthouse, one synthetic run) measure different things.
—
Lighthouse: 1 run, no network jitter, no real devices. Variance ±15-20% per run.
—
CrUX p75 INP: rolling 28-day window. A fix lands today, the number moves in ~3-4 weeks.
—
Gap example: lab LCP 1.8s, field p75 4.2s. The difference is your slow 25%, not your test machine.
Use Lighthouse to debug a regression, CrUX to decide if it shipped value.
Takeaway: trust the 28-day p75, not one 1.8s lab run.