Lab speed tests

The Lab card sits beside Speed on a site’s board. It shows performance, accessibility, best practices and SEO scores; lab LCP, CLS and blocking time; a score history; and the top five opportunities with estimated savings and affected resources. Savings overlap and should not be added together.

Lighthouse uses a controlled browser and simulated conditions. Your visitors use many devices, networks and pages, so lab values can differ from real visitor p75. Lighthouse does not measure field INP; total blocking time is a lab diagnostic.

Checked pages are the active page checks on the account’s plan. Tests use the site’s saved public HTTPS host, without custom check headers, sign-in credentials, queries or fragments. A page that requires a sign-in may fail its lab test.

Paid accounts can queue one extra homepage test per UTC day, shared across their sites. On Pro and Ultra, a new deploy marker also queues homepage tests. These are queued requests; a shared quota or Google outage can delay them. Paid runs take priority. Tests retry later on failure and keep the last successful score.

A performance score drop of at least 10 points against a pre-deploy run becomes a lab regression. LCP crossing 4 seconds or CLS crossing 0.25 from a better previous reading also raises a regression. Comparisons use the same page and device. Without a comparable earlier run, PageHum waits for a baseline.

A regression appears in the Lab card and the owner’s notification inbox, with a Copy prompt containing the page, opportunities and resources. Lab scores can vary between runs: confirm a change before treating it as a visitor impact.

Choose a page and device, then expand score history. Summaries follow the plan’s history retention. Pro and Ultra can download the original JSON for 30 days; downloads require access to the site and use a short-lived private link.

Google runs the public page test. Its full report can include public resource URLs and page diagnostics from that test browser. These are separate from the visitor snippet’s aggregate readings.

GET /v1/sites/{key}/lab returns homepage mobile tests by default. Use path, strategy=mobile|desktop, and before (the previous response’s nextBefore) for other pages/devices and older history. Reads start no external test. raw=RUN_ID returns a short-lived full-report download link when available.

The MCP tool is get_lab, with the same site, path, strategy and before arguments. The response includes whether tests are configured, queued or failed, alongside saved scores and regression prompts.