What is real user monitoring?

A lab measurement runs under fixed conditions: a simulated device, a throttled connection, one load. It is repeatable — the same page twice gives almost the same number. That is exactly why it suits comparing two versions.

A field measurement collects what happened to real visitors: a seven-year-old phone on a train, a machine with thirty tabs open, a fast line at three in the morning. It is not repeatable and is still the only one describing reality.

Both are correct. They just answer different questions: is my change better? versus how do people experience the site?

For Core Web Vitals, field data counts. Google draws it from the Chrome User Experience Report, CrUX for short — anonymised measurements from Chrome users who opted in.

Two properties of it explain almost every confusion. First the value is a 28-day window: an improvement made today only becomes visible over weeks. Second it is the 75th percentile — not the average but the value three out of four loads stay under. So a slow quarter drags the grade along, and a very fast median does not rescue it.

The everyday case: Lighthouse shows 98, Search Console reports "needs improvement". Not an error but four differences at once.

• Device — the lab simulates a mid-range phone; visitors bring everything.
• State — the lab loads cold; real visitors often have a warm cache.
• Interaction — INP only exists once someone clicks. Nobody clicks in the lab.
• Time — the lab measures now, the field the last 28 days.

Which is why the order matters when diagnosing: the field tells you whether there is a problem. The lab helps find why. Measuring only in the lab risks optimising something no visitor ever meets.

CrUX costs nothing and is the source Google draws its assessment from. It has limits: Chrome only, only pages with enough traffic, and values arrive as a window rather than an individual case.

Your own RUM collects in your own code — usually through the web-vitals library — and delivers what CrUX cannot: all browsers, immediately rather than after weeks, and the ability to join the value to your own attributes: page type, country, signed in or not.

The practical split: CrUX is the grade that counts. Your own RUM is the tool for understanding it and for checking quickly whether a change worked.

Checking immediately after a deploy. The field window is 28 days; yesterday's improvement is barely visible in it. Judging the day after mostly measures the old state.

Looking at the average. The 75th percentile is what is graded. A good mean can sit alongside a bad grade when a quarter of loads fall far behind.

Taking one page for the site. CrUX reports per URL and, where data is too thin, per group of similar pages. The homepage does not stand for the catalogue, and one measured page does not stand for all.