Performance

Open Performance for measurements collected from visitors’ browsers.

Metrics

Choose a metric in the top strip to update the chart and breakdowns. Timing measurements and CLS use different units.

Percentiles and attention

Percentile views describe the sample distribution. A p75 value means 75% of the samples are at or below that value. Needs attention orders paths by the selected metric’s p75, worst first, and requires at least three samples per path.

Percentiles use the nearest-rank rule: sort the observed values and take position ceil(percentile × sample count), counting from one. They do not interpolate between samples. For four sorted measurements, p75 is the third value. The calculation preserves the stored measurement precision; the displayed value may be rounded for readability.

The chart’s daily points are calendar days in the display time zone (hourly points for ranges up to two days). See one time zone for every report. Use p50, p75 and p95 above the chart to show or hide their lines. These controls change the chart display; the KPI strip and breakdown tables still use p75. Time metrics are displayed as durations; CLS is a unitless score. Each KPI shows its own sample count, so measurements need not have equal denominators.

Breakdowns

Inspect device, country and browser groups, each with p75 and sample count (three cards across on wide screens, stacked below). Groups sharing the same p75 are ordered by their key in descending order, whichever way the analysis ran. Browser and device rows use the same icons and fallback badges as Overview. Small samples can change sharply; compare equivalent periods before drawing conclusions.

Measurement limits

The tracker uses the Web Vitals library and browser-supported observations. Missing metrics are not zero scores. These are field measurements, not a Lighthouse test, and single-page navigation does not necessarily create a new document measurement. Web Vitals have a separate usage allowance.

When the tracker sends a newer update for the same measurement identifier, metric and page, it replaces that measurement’s value. Repeated or older updates do not add another sample. This remains true for delayed updates after a period of inactivity. Measurements without an identifier are stored separately. If an identity cannot be checked during a storage outage, the request can be rejected temporarily; it is not counted as a new sample. See storage and retries. A cached report keeps its captured results until a new analysis runs.

Loading an analysis

The selected period is calculated from its captured sample set. The analysis runs in the background and shows a loading state until the exact result is ready. Accepted measurements that are still becoming available can delay preparation. No sampling or approximate percentile estimator is used. Calculations save final results without retaining per-sample working copies; source measurements and saved results are preserved. The result is not a truncated sample. An empty metric means no measurements are available for that metric in the selected period. A finished result is reused for thirty minutes and the period’s end moves in half-hour steps, so reopening the page or switching metrics back does not recompute the same period; the notice above the page shows when it was computed.

Choose the correct website and date range before comparing results. No measurements and a dash mean that metric has no samples in the scope. An empty Needs attention table can mean paths have fewer than three samples even when the metric has data overall. Interacting with a page may be required before the browser can report some metrics, and browser support differs.

If preparation fails, read the analysis notice and retry when offered. A loading or failed analysis is not a zero score. Check tracker setup and the separate Web Vitals allowance if new samples stop arriving.

Something unclear? Contact support.