Fix missing web vitals
Why web_vital events or speed values are missing, and what to change.
Use this page when the speedInsights plugin is set up but the speed reads show no values, or only some metrics.
No web_vital events arrive
- The plugin is not in
plugins. Speed Insights is not on by default. AddspeedInsights()to the client'splugins. - The page load was sampled out. With
sampleRate: 0.1, one page load in ten measures. The choice is made once, when the client starts. Use the default of1while you test. - The page closed before
web-vitalsloaded. The plugin loads theweb-vitalsattribution build with a dynamic import, and starts measuring once it arrives. A visitor who leaves before that sends no metrics. - The event itself did not go out.
web_vitalevents follow the same rules as every other event: development mode, consent, opt-out andbeforeSendapply. See Find missing events.
With ?ra=debug, each metric the plugin sends shows in the console as an RA_EVENT line for web_vital.
Some metrics never arrive
These limits come from the web-vitals library the plugin uses:
| Metric | Not reported when |
|---|---|
| CLS | The browser is not Chromium-based, or the page loaded in a background tab |
| INP | The visitor never interacts with the page |
| LCP, FCP | The page loaded in a background tab |
LCP, INP and CLS change while the visitor stays on the page, so the plugin holds them and sends them together when the tab is hidden, at the next pageview, or once 6 metrics are waiting. Expect them after the visitor switches tabs or leaves, not right after load.
Metrics are credited to the wrong page
The plugin measures the hard navigation only, and credits every metric to the path the page loaded with, even when it is sent after a client-side navigation. Pages reached by client-side navigation get no metrics of their own.
A page load reports its metrics twice
When a visitor goes back or forward to a page the browser kept in its back/forward cache, web-vitals reports every metric again with a new id. The API stores each id as its own sample, with navigationType set to back-forward-cache. These are separate visits to the page, so they count as separate samples.
Preview deployments show no speed
The speed reads show production by default. Add environment=preview for preview deployments only, or environment=all for both.
The speed read shows null for a metric
The speed reads show a value only for a metric with at least 20 samples in the range, after filters, and leave the others out of the score. samples in the response says how many there are. Widen the range, set device=all, or remove filters.
The API stores no speed sample for:
- events from bot, internal or localhost traffic, whatever
traffica read asks for; - an unknown metric or rating, or a missing metric
id; - a negative value, a CLS above 10, or a timing above 120 seconds.
Speed for dates more than 30 days ago is missing a filter or the elements
The daily rollup job drops raw speed rows older than 30 days after writing daily percentiles and rating counts per route, device and metric to rollup_vitals. The speed reads use the raw rows for the last 30 days and the rollup for older days, where a day's percentile is the mean of its per-route and per-device percentiles, weighted by samples.
The rollup has no page, country, selector, hour or preview rows. For days older than 30 days:
- a read with
filter[page],filter[country]orenvironment=previewanswers from the raw rows only, so older days are empty; /speed/elements,/speed/routes?group=pathand/speed/timeseries?interval=hourcover the last 30 days only;/speed/routesleaves out rolled-up samples without a route.
Older days also show nothing when the rollup job has not rolled them up, even if their raw rows are still stored, because the reads take those days from the rollup only. Run POST /v2/admin/jobs/rollup with days (1 to 90) large enough to cover them.