Core Web Vitals & performance
Monitor real-user performance: LCP, FCP, CLS, INP and TTFB, broken down by page, device and browser.
The Performance hub reports Core Web Vitals from real visitors (field data), not lab tests.
The metrics
| Metric | Measures | Good | Needs work |
|---|---|---|---|
| LCP | Largest Contentful Paint — load speed | ≤ 2.5s | ≤ 4.0s |
| FCP | First Contentful Paint | ≤ 1.8s | ≤ 3.0s |
| CLS | Cumulative Layout Shift — visual stability | ≤ 0.1 | ≤ 0.25 |
| INP | Interaction to Next Paint — responsiveness | ≤ 200ms | ≤ 500ms |
| TTFB | Time to First Byte — server response | ≤ 800ms | ≤ 1.8s |
These thresholds are defined once and used everywhere Spectry rates performance — the Performance hub, page detail, AI insights and alerts all agree on what "Good" means.
Breakdowns
Switch tabs to view vitals overall, or split by page, device or browser — useful for finding, say, a slow LCP that only affects mobile. Each row shows its sample count alongside the value.
Low confidence
A 75th percentile calculated from a handful of measurements is noise dressed up as a number. Any metric or breakdown row built from fewer than 100 samples is flagged low confidence and marked with a warning. Treat those values as a hint that data is still accumulating, not as a verdict on the page.
How measurement works
- CLS is measured over the whole session window, not just the initial load — layout shifts caused by lazy-loaded content or late-arriving ads are counted.
- INP is reported at the 98th percentile of a visit's interactions, so one unlucky interaction doesn't define the page but a consistently sluggish control does.
- Measurements are flushed when the page is hidden or unloaded, which catches visitors who leave before a normal page transition.
Values are reported at the 75th percentile (P75) and pre-classified as Good / Needs improvement / Poor, matching Google's thresholds — no need to interpret raw numbers yourself. Field data reflects your real visitors on their real devices and networks, so it will usually look worse than a lab test run from a fast connection. That is the point.