PageSpeed & Core Web Vitals

Pulse runs Lighthouse audits against your site on its own self-hosted runner and tracks performance over time. Results are stored so you can see whether changes you ship improve or hurt your scores.

Enable PageSpeed checks

Go to your site Settings and toggle PageSpeed Monitoring on. Then choose a check frequency:

You can also trigger a check on demand using the Run Check button. A 5-minute cooldown applies between manual runs.

PageSpeed dashboard

Results are at /sites/{id}/pagespeed.

Lighthouse gauges

Four circular gauges show scores from 0–100:

• Performance
• Accessibility
• Best Practices
• SEO

Each gauge is colour-coded: green (90–100), orange (50–89), red (0–49).

Core Web Vitals

Individual metric cards with a Good / Needs Improvement / Poor rating for each:

Page load filmstrip

A frame-by-frame screenshot sequence showing how the page renders over time. Useful for identifying when content first appears and whether there are layout shifts visible.

Performance trend chart

A historical chart of your Lighthouse Performance score over all recorded checks. See at a glance whether your score is improving, declining, or stable.

Diagnostics

A list of actionable findings grouped by category (Performance, Accessibility, Best Practices, SEO), each with a severity level and estimated savings where applicable.

Mobile and desktop

Use the Mobile / Desktop toggle to switch between the two audit types. Scores and metrics differ significantly between modes — mobile is typically lower due to simulated throttled network and CPU.

Historical navigation

Use the check history navigation to step back through previous audits and compare them. The filmstrip, metrics, and diagnostics all update to reflect the selected historical check.

What each number means

Performance score

Lighthouse's 0–100 score for the page, from Pulse's own Lighthouse runner rather than Google PageSpeed Insights, which this instrument replaced on 14 August 2026. It follows the same colour bands as the gauge above: red 0–49, orange 50–89, green 90–100.

Accessibility score

Lighthouse's 0–100 accessibility score from the same run — automated checks only, such as contrast, labels, and ARIA usage. Checks Lighthouse cannot verify automatically are listed separately in Diagnostics as items to check by hand, and aren't folded into the score. A 100 doesn't mean the page is accessible.

Best Practices score

Lighthouse's 0–100 best-practices score covers browser and web-platform hygiene — HTTPS, console errors, deprecated APIs, security headers, and similar checks. It says nothing about load speed.

SEO score

Lighthouse's 0–100 SEO score covers on-page crawlability — meta tags, crawler directives, link text, canonical and hreflang, and similar checks. It measures whether a crawler can read the page, not how the page ranks or what traffic it gets.

First Contentful Paint (FCP)

Time until the first text or image appeared on screen. Under 1.8s is good.

Largest Contentful Paint (LCP)

Time until the largest visible element finished rendering. Under 2.5s is good.

Total Blocking Time (TBT)

How long the main thread was blocked between First Contentful Paint and the page becoming interactive, counting only the time past 50ms in each long task — three 60ms tasks are 30ms of TBT, not 180ms. Under 200ms is good.

Cumulative Layout Shift (CLS)

How much visible content moved unexpectedly during load, on Lighthouse's unitless scale starting at 0 — not a time. Under 0.100 is good.

Speed Index

How quickly the visible page filled in, scored by Lighthouse across the whole load rather than from one element's timestamp. Under 3.4s is good.

Time to Interactive (TTI)

Time until the page could reliably respond to input. Under 3.8s is good. It carries no weight in the Performance score — it is shown for diagnosis, so a poor TTI beside a green Performance gauge is not a contradiction.

How a check is produced

Where a check comes from

Checks run on Pulse's own Lighthouse runner, not Google's PageSpeed Insights API, which was retired here on 14 August 2026. The engine is pinned to Lighthouse 13.4.1 — the version PSI was serving at the cutover — and the version is recorded on every check, so a score change can be read against a fixed instrument instead of one that may have changed underneath the numbers.

Median of 3

Every stored check is up to three full Lighthouse runs — the budget can end a sweep early, and only successful runs are candidates. The single run whose performance score is the median is kept whole — not a metric-by-metric median, which would describe a page load that never happened. With two candidates there is no true median and the lower of the pair wins rather than the more flattering one.

A failed check

A failed run is stored as its own row carrying the cause, rather than dropped — the dashboard keeps showing the last successful numbers under a visible "check failed" line, so a failure is never mistaken for a fresh result. Every measured field on a failed row, and any metric a successful run couldn't capture, is empty, shown as an em dash, never as zero.

Screenshot and filmstrip retention

The screenshot and filmstrip are kept only on the newest successful check per site and device — older checks keep their scores and metrics but lose their imagery. An older check showing correct numbers and no picture is the retention rule, not a loading failure.

Reading the trend chart

The provenance boundary

The dashed vertical line on the performance trend chart marks the switch to Pulse's own runner. Left of it is a single PageSpeed Insights run on an unknown Lighthouse version; right of it is a median of three on a pinned version. A step across that line is a change of instrument, not necessarily a change in the page.

The trend line

The line itself is a trailing median over the last 7 scored checks including the current one, not the raw score — with an even count it takes the lower of the two middle values, the pessimistic choice. The dots are the individual checks. Both are drawn because the line alone hides run-to-run spread, and the dots alone hide the trend.