Insights:Core Web Vitals Measurement: How Field Data, Lab Data, and CrUX Work Together
The most common debate around Core Web Vitals is: “My Lighthouse score is clearly 95, so why does Search Console say I’m failing?”
This isn’t a tool conflict — the three data sources are answering different questions. Mixing them up leads to repeatedly fixing the wrong things.
What Each Data Source Answers
Field Data: Data collected from real users on real devices over real network conditions. It answers “How do users actually experience this page?”
Lab Data: Data gathered under controlled conditions with fixed devices, networks, and page states. It answers “How fast can this page perform in an ideal environment?”
CrUX: The Chrome User Experience Report is a public source of field data. It aggregates by real users, per URL or origin, typically using a 28-day rolling window and the 75th percentile.
These three don’t replace each other — they serve different purposes.
Field Data: Start Here Because It Reflects Reality
The value of field data is that it doesn’t filter for environment. Whatever phone, city, or network the user has is all included.
When reviewing field data, keep four points in mind:
- The 75th percentile, not the average. CWV thresholds are calculated at the 75th percentile, meaning at least 75% of visits must meet the benchmark. A good average doesn’t guarantee most users have a positive experience.
- 28-day rolling window. Changes you make today won’t appear in CrUX tomorrow; the window needs to roll forward.
- Group by device and template. Mobile and desktop, article pages and listing pages often have different issues.
- Don’t force-read small samples. Low-traffic pages may lack CrUX data or show high volatility; supplement with lab data and your own RUM.
Field data is best suited to answer: which template performs worst, whether mobile or desktop is lagging, and whether changes actually improved after going live.
Laboratory Data: Use It to Pinpoint Issues, Not to Draw Conclusions
Lighthouse, the lab section of PageSpeed Insights, and WebPageTest are all forms of laboratory data. Their advantage is reproducibility: run tests under the same conditions twice and you can compare the results.
Suitable uses for laboratory data:
- Reproduce which step causes a slow LCP on a specific template.
- Compare render-blocking resources before and after changes.
- Evaluate how a third-party script impacts the main thread.
- Run regression testing before release to prevent new components from degrading metrics.
Not suitable for:
- Determining whether "real users have acceptable scores."
- Using a single score of 100 as a launch approval.
- Inferencing mobile performance from desktop scores.
Laboratory data requires fixed conditions: device, network, cache state, login status, and page state. If these aren't consistent, comparing results over time is meaningless.
CrUX: Field Data, But Not the Whole Picture
CrUX provides real-user distributions at the URL or origin level, making it useful for quickly spotting trends. It does have limitations:
- It only covers pages with sufficient traffic; long-tail pages may have no data.
- As aggregated data, it won't tell you which specific element or line of code is slow.
- There's a delay, so it can't be used for same-day acceptance.
- Statistics by URL and by origin cover different scopes—don't compare the two numbers directly.
So CrUX is suited for a "checkup," not for "surgery." When a template fails, you still need to go back to laboratory data and RUM to pinpoint the issue.
Recommended Measurement Order
Step 1: Use field data to find problematic templates.
In Search Console's CWV report or CrUX, group by device and template, then identify the group with the highest failure rate. Don't start with the homepage—it's often specially optimized and doesn't represent the whole site.
Step 2: Use laboratory data to reproduce.
Take a set of similar pages and run Lighthouse or WebPageTest under consistent mobile conditions. Check what the LCP element is, what the TTFB is, how many render-blocking requests there are, and whether the main thread has long tasks.
Step 3: Change one variable and run again.
Only adjust one of images, caching, or scripts at a time. Otherwise you won't know which change made the difference.
Step 4: Wait for field data to roll up.
After deployment, wait at least one 28-day window before checking CrUX or Search Console. Use RUM or lab data for quick regression checks during this period, but don't roll back a correct change just because the field metrics haven't moved the next day.
A Common Pitfall: Treating Scores as the Goal
CWV is one page experience signal, not the only ranking factor. Trying to push a Lighthouse score from 92 to 100 by removing useful content or deferring critical above-the-fold resources often does more harm than good.
A more sensible approach is:
- First ensure the 75th percentile hits target.
- Then address experience factors that affect conversions, such as first interactive state, form feedback, and layout shift on images.
- Finally, aim for clean lab scores.
When measurement methodology is unified, optimization stops going in circles.
Related reading: