Skip to content
RU
← All articles

Lighthouse vs CrUX: why lab and field results differ

In short:

Lighthouse measures one controlled run, CrUX aggregates real visits over a rolling 28 days, and Search Console groups similar URLs. In our enterno.io mobile snapshot Lighthouse scored 96, the specific URL was FAST, and the whole origin was AVERAGE because INP reached 223 ms. Match URL, device and period first.

Analyst comparing lab performance test with real-user data
One lab run and a user distribution answer different questions.

A good Lighthouse score and a Search Console warning can both be correct. The products observe different populations and report different outcomes. We saved one PageSpeed Insights response for the public enterno.io home page with the API key removed. This is a concrete example of scope mismatch, not a claim that our Search Console currently marks the page poor.

Verdict: use field data for priority and lab data for cause

CrUX and the Search Console Core Web Vitals report reflect real visits. The 75th percentile means one quarter of observed experiences were worse than that value. Lighthouse runs the page under controlled conditions and helps identify resources, scripts and layout shifts. It cannot capture every device, network, authenticated state and late interaction. Field INP can therefore be weak even when the lab score is high.

CriterionLighthouseCrUX / PSI fieldSearch Console CWV
SourceOne controlled runReal Chrome visitsURL groups from CrUX
PeriodRun timeRolling 28 daysField window
DeviceChosen run profileMobile or desktopSeparate device groups
ScopeTested URLURL or originGroup of similar URLs
StatisticsScore and diagnosticsp75 and distributionStatus from weakest group metric
INPNo full history of real actionsReal interactionsGroup visit data
StrengthReproduce a cause quicklyPrioritize user impactFind affected URL groups
Blind spotRare conditions and late actionsInsufficient data for some URLsExact script cause

enterno.io snapshot: three different answers

On 14 November 2026 PageSpeed Insights returned a mobile laboratory score of 96 out of 100 for https://enterno.io/. The field block for that URL was FAST: LCP p75 1,403 ms and INP p75 189 ms. The field block for the entire https://enterno.io origin was AVERAGE: LCP p75 1,735 ms and INP p75 223 ms. The last number exceeds the good INP threshold of 200 ms, while the specific URL in the same response performed better.

This does not mean Lighthouse was wrong. A home-page run is not every origin route, and one lab result is not 28 days of interaction data. The response with field aggregates is saved with the draft.

Mobile PSI, enterno.io, 14 Nov 2026:
Lighthouse performance: 96/100
CrUX URL: FAST, INP p75 189 ms, LCP p75 1403 ms
CrUX origin: AVERAGE, INP p75 223 ms, LCP p75 1735 ms
# the URL and origin have different visit populations

We did not access a private Search Console property and do not claim its current status. Google documents that Search Console groups similar URLs and can use a wider origin when a single URL has too little data. The group label can therefore differ from a single PageSpeed Insights URL result. Inspect the group and its example pages before drawing conclusions.

Analyst viewing one site on several devices
Field data aggregates varied conditions over 28 days.

Match URL, device and collection window

First open the example URL from Search Console in PageSpeed Insights and choose the same device. Check whether field data are available for that exact URL. If only origin data appear, do not assign origin p75 to one page. Inspect the collection period. A fix shipped yesterday is still mostly hidden by 27 older days in a rolling window.

Then compare the metric, not the overall score. Good LCP does not cancel slow INP in a Core Web Vitals status. A high lab score cannot guarantee fast interactions after page load. For INP, reproduce a common slow click and use LoAF to find the handler. For CLS, record the layout shift track after late content appears.

For a Search Console URL group, sample several member pages and compare their templates, widgets and critical resources. A single shared script can affect hundreds of pages. Fixing one URL in a lab run will not help if the common component remains. When a URL lacks enough field data, never present the origin average as that page's personal measurement.

Why 96 does not mean everything is fast

Lighthouse combines laboratory measures into a score. A rare click after several minutes of reading can produce high field INP but never occur in a one-off audit. Mobile visitors use slower CPUs, variable networks and background tabs. A popular home page may be fast while articles, forms or account pages are slower. The origin or group can therefore have a worse label.

The reverse happens too: a lab run can catch a temporary slow network while the 28-day field p75 remains healthy. Do not ignore a genuine new regression merely because field data have not caught up. But prioritize work by actual impact: how many visits are affected, which action is slow and whether that action matters to the user.

Investigation order

  1. Record URL, device, date and field metric. Do not compare desktop Lighthouse with mobile CrUX.
  2. Determine whether the field result represents a URL, origin or Search Console group.
  3. Name the weak metric: LCP, INP or CLS. An overall score hides that distinction.
  4. Reproduce the state in DevTools with comparable viewport and network settings.
  5. Verify the lab fix immediately, then watch the field trend as new visits enter the window.

For LCP, inspect HTML, images, fonts and response time. The enterno.io speed tool and font guide help narrow the cause. For INP, inspect long tasks and frames. For CLS, inspect geometry. Save comparable before and after evidence in the runbook, not only a score screenshot.

How we made this

We requested Google PageSpeed Insights on 14 November 2026 for the public enterno.io home page with a mobile strategy and a project key that was not stored. data.json retains URL field data, origin field data, the lab score and analysis time, with no personal data. It is one snapshot, not a trend or Search Console export. We checked the data scopes, 28-day window and URL grouping against official Chrome and Google Search Console documentation.

Walk through the scope mismatch

Suppose PageSpeed shows good URL INP but the whole origin needs improvement. First check whether the specific URL has its own field sample. Our snapshot does: 189 ms belongs to the home page. The 223 ms origin value represents visits to different routes on the same site. It does not prove that the home page suddenly became slower. Next inspect routes with interactive components, such as forms, search results or account screens, if they share that origin.

In another case a rarely visited URL has no URL-level field sample. PageSpeed may then show only origin data. That is useful context for the site, but not a personal measurement for the rare URL. Compare several pages using the same template and identify shared actions. A fix to a common component can help the group. A problem unique to one page should be isolated in a lab trace without assigning the origin p75 to it.

Finally, imagine Search Console groups URLs with weak CLS. Open several examples from that group, not only the home page. Inspect ads, images without reserved space and late fonts. Our CLS walkthrough shows how to locate a real shift. The lab trace finds a cause; the group status helps estimate how broadly it matters.

What to record after diagnosis

Save a complete comparison row: tested URL, mobile or desktop, lab run date, whether URL-level field data exist, the weak metric's p75, and the Search Console group scope. Separately name the element or script that reproduced the problem locally. Without these fields, “Lighthouse is green but Console is red” is not actionable. It mixes a single-run score, a visit distribution and a page group.

Do not replace a field chart with a new Lighthouse score after a release. The lab run can show that the specific code is faster under selected conditions. The rolling 28-day window still includes visits from before the release. Note the deployment date and watch the window update. If traffic mix, devices or group membership changed at the same time, interpret the trend carefully.

A new article cover might affect LCP differently on a local connection and on visitors' networks. A late widget on one template might cause CLS. A form that Lighthouse never clicks might drive INP. Each metric requires its own scenario. The font test helps with suspected text shifts; the curl timing breakdown helps inspect LCP network phases.

FAQ

Can Lighthouse 96 coexist with slow INP?

Yes. A lab audit does not replay every real interaction. Read field INP for the relevant group and investigate an actual slow action.

Why do PSI URL and origin have different labels?

They aggregate different visits. URL is one page; origin includes many paths. In our snapshot, INP p75 was 189 ms and 223 ms respectively.

When will a fix shipped yesterday appear in field data?

CrUX uses a rolling 28-day period and changes gradually. Check the lab fix right away, then watch the field trend as new visits accumulate.

Sources

Check your website right now

Check your site's speed →
More articles: Performance
Performance
Gzip vs Brotli: Web Compression Compared
16.03.2026 · 1 049 views
Performance
CDN Cache Invalidation: Strategies for Fresh Content
16.03.2026 · 906 views
Performance
Latency vs Throughput: Network Performance Metrics
16.03.2026 · 790 views
Performance
What Is a CDN: How It Works, Why You Need It and How to Check
11.03.2026 · 778 views