A green Lighthouse score with failing Core Web Vitals is not a contradiction. One is a simulation on a fixed device; the other is what happened to your actual visitors. Only one of them is in Search Console.
Every performance conversation eventually reaches the same impasse: the developer has a Lighthouse score of 98 and the SEO report says the site fails Core Web Vitals. Both are correct.
What each one is
Lab data — Lighthouse, PageSpeed Insights' lower section, WebPageTest — is a simulation. One page, one simulated device, one throttled connection, no human interacting with anything. It is repeatable, which makes it perfect for catching regressions in a build pipeline.
Field data — the Chrome UX Report, and any real-user monitoring you run yourself — is a record of what happened to real people on their own devices and networks over a rolling 28-day window. It is what Search Console reports, and it is what page experience signals are based on.
Why they disagree
- Interaction to Next Paint cannot be measured in the lab at all. It requires a real interaction, so Lighthouse simply does not report it. This alone explains most disagreements.
- Your users' devices are slower than the simulation, and their networks are less consistent.
- Field data is a distribution, reported at the 75th percentile. Your median visitor may be having exactly the experience Lighthouse describes while a quarter of visits are far worse.
- Lab tests run one URL; field data groups many. A fast home page and a heavy product template can average into a failing group.
- Logged-in and returning users carry different caching, different scripts and different behaviour from the cold anonymous load a lab test simulates.
How to use both properly
- Diagnose in the field. Search Console tells you which URL group fails and on which metric. Real-user monitoring with attribution tells you which element and which phase.
- Reproduce in the lab, with CPU throttling at 4x or more. Without throttling you are testing your laptop.
- Fix, then wait. CrUX reports on a rolling 28-day basis, so a change shipped today will not show fully for about a month.
- Guard with lab tests in CI, so the fix does not quietly reverse in three sprints.
The rule of thumb: never celebrate a lab score, and never debug from field data alone.
Looking for more? Browse all resources.