Field vs. lab Core Web Vitals: read the difference
Diagnose conflicting Core Web Vitals results by separating field coverage from controlled lab tests, then reproduce a slow-device interaction before changing code.
Why can field and lab results disagree?
Field and lab results answer different questions. Real visitors bring varying devices, networks, caches and interaction patterns. A lab test controls some of those conditions to make diagnosis repeatable. Before treating a difference as an error, record what each result actually measured and whether the scopes match.
Google’s web.dev identifies LCP, INP and CLS as the current Core Web Vitals and recommends judging their targets at the 75th percentile, segmented across mobile and desktop. Its lab guidance also explains that a simulated load without user input cannot directly measure INP. [Google web.dev: Web Vitals]
A field-versus-lab decision table
| Observed situation | What it means | Next diagnostic action |
|---|---|---|
| No URL-level field data | Coverage is unavailable or insufficient for that view | Check whether origin data is shown; do not report a pass |
| Fast lab load; slow field loading | Conditions or visitor mix may differ | Reproduce slower network, device and cache conditions |
| Good load test; slow interaction | The test may not exercise the actual interaction | Record a real menu, filter or calculator action |
| Layout stable initially; shifts later | Late content or interactions may matter | Observe the full flow and delayed elements |
| Result changed after release | Window or mix may still include older visits | Record release timing and compare matching scopes |
Do not merge URL and origin results into one page verdict. The origin can include routes with different assets and behavior. Similarly, a desktop result does not establish the mobile experience. Save the exact reported coverage so the next reviewer can reproduce the interpretation.
A reproducible slow-device check
- Choose the affected public route and a representative task, such as opening a comparison and calculating a result.
- Record browser version, viewport, device or emulation, network profile and cache state.
- Load the page, wait for delayed content and perform the same interactions in the same order.
- Capture the relevant trace and note the element or event associated with the delay or shift.
- Change one suspected cause, then repeat the same fixture before comparing outcomes.
A useful fixture includes the real long title, image, table and form state. Testing an empty shell can miss the asset or interaction causing the problem. If the issue appears only after consent or a third-party widget loads, record that state explicitly instead of silently comparing different page conditions.
Repeat enough runs to see whether the local observation is stable, and preserve the range rather than selecting the fastest run. The purpose is to diagnose a cause. It is not to manufacture a lab score that resembles the field report.
Which problem should be fixed first?
Prioritize a reproducible issue that affects the primary task and has a plausible causal trace. A large image delaying the main content, a long calculation blocking input or a late banner moving the submit control can each matter differently. Tie the change to the observed mechanism and retest that mechanism.
Do not equate a lab proxy with the field metric it helps diagnose. A loading audit without user input may report blocking-related measures, but that does not supply a measured interaction distribution. Keep the metric names and test conditions in the report.
A useful performance handoff
Record the affected route, field scope, local fixture, suspected cause, change and observed retest. Include what remains unknown. If field data is unavailable, say so and retain the controlled test as technical evidence. That gives the next check a clear baseline without inventing a user-experience result.
Sources
AI-assisted writing and an AI-generated editorial cover. Source links and checked dates are listed below; illustrative examples are labeled in the text.
- Google web.dev: Web Vitals — checked 2026-10-18