Why good numbers do not hold
After an optimisation the figures look good. Six months later they look worse, and nobody can say when that happened.
The cause is rarely a single event. Content grows, image galleries get longer, an extension receives an update, marketing adds a script for a campaign and nobody removes it again. Each step on its own is unremarkable, and the sum is not.
It can only surface if somebody looks regularly. Nobody does that by hand for months on end.
SiteSentry
We use SiteSentry for it, the tool from aceArt GmbH, the agency t3land belongs to. It is free.
What it does: at an interval you choose between two hours and once a day it calls Google’s official PageSpeed interface and measures your pages. It records the four PageSpeed categories, meaning performance, SEO, accessibility and best practices, along with Core Web Vitals such as the largest contentful paint and layout shift. Mobile and desktop run as separate measurements.
A threshold can be set per category and per device. When it is breached, an email arrives. Every measurement lands in a history with charts, and that history is often the most useful part of the whole thing: it answers the question of since when.
The interface is available in German and English. Setting it up needs a registration with an email address and the addresses of the pages to be measured.
What a lab measurement can and cannot do
So the expectation is right: the PageSpeed interface measures under standardised conditions, with a simulated device and a simulated connection. That is a lab measurement.
Its advantage is comparability. Because the conditions are the same every time, a change in the number means a change in the site. That is exactly the point of watching it.
What it does not answer is whether your visitors notice. For that there is field data from real sessions, and that sits in Search Console. The two together give you a picture; either one alone does not.
Which pages belong under observation
Not all of them, and the home page is rarely the most interesting. What works is a selection by template: one page from each kind there are many of.
On a typical TYPO3 site that comes to four to six addresses: the home page, a standard page with a lot of text, a list view, a detail page from an extension and, if there is one, a page with a form. When a template tips over, every page built on it tips with it, and that is exactly what you never see on a single address.
What we do with it
You can set the measurement up without us; the tool is open and it takes a few minutes. Where we come in is the reading of it.
An alert says a value has dropped. It does not say why. On installations we look after, checking that is part of the ongoing maintenance: we line the moment up against what happened in the same week and tell you whether it was a piece of content, an update or an embedded script. If that turns into work, it is described under page speed optimisation.