What Your Users Actually Experience
You are the least representative visitor your site will ever have. You are on a fast machine, on a good connection, geographically close to wherever it is hosted, with every asset already cached and the fonts already resolved. Everyone else is on something else. Real user monitoring is the measurement taken from their side — what the page actually cost the people who loaded it, on the hardware and networks they happen to own.
Lab and field are different measurements, not competing ones
A performance test you run is lab data: controlled, repeatable, available on demand, and produced from a machine you chose. Real user monitoring is field data: uncontrolled, noisy, only covering pages people actually loaded, and the only measurement that reflects your real mix of devices and networks. Google assesses Core Web Vitals from field data, which is what turns this from an interesting metric into a consequential one.
When the two disagree, neither is broken. A lab score of 95 alongside a failing field assessment is an entirely normal outcome, and the gap is the finding — it says the cost is in real devices and real networks rather than in anything your test machine experiences.
What the three metrics actually tell a developer
Largest Contentful Paint is usually about one element — a hero image, a background, a heading waiting on a web font — and the fix is usually about how early the browser learns it needs that resource. Cumulative Layout Shift is about reserved space, which mostly means dimensions on images and room for anything that arrives late, and it accumulates across the whole visit rather than just during load. Interaction to Next Paint is about the main thread being busy, which makes it the hardest to fix, because the answer is usually to do less JavaScript rather than to change a setting.
Field data is what tells you which of the three is actually your problem. Optimising all three at once is a project; optimising the one that is failing at the 75th percentile on mobile is an afternoon.
Why the percentile matters more than the number
These are reported at the 75th percentile rather than as an average, and that is deliberate. An average is dominated by the fast visits — the desktop loads with a warm cache — and will look healthy while a quarter of your visitors are having a genuinely bad time. The 75th percentile answers a more honest question: was this good for at least three out of four visits. It is also what determines the assessment that affects your search ranking, so it is the number worth watching regardless.
What it will not do
It needs traffic, so a page nobody visits produces nothing and a low-volume page produces numbers too noisy to act on. It is not error tracking — a page that renders instantly and fails to submit scores perfectly. And it tells you what is slow, not why: attribution narrows it to an element, and a lab run with a full waterfall is what confirms the cause. Running both against the same pages is what makes each one more useful than it is alone.