What we check
Ten checks on every website, watching for around twenty distinct failures — from the address alone. Nothing to install, and nothing to ask your client's developer for.
Nineteen ways a page loses money while still answering 200
The checks above are how often we look. These are what we are looking for — every one implemented, tested, and reported with the evidence behind it.
Critical Wakes somebody up
- The site has been hackedInjected scripts or content that were not there before.
- The page told Google to stay awayA noindex appeared on a page that used to be indexed.
- The page went nearly emptyIt still answers, but the content collapsed.
- A missing page pretends to be fineIt says not found while answering 200, so nothing else notices.
- The secure address stopped being securehttps started sending visitors to http.
- Text you asked us to watch vanishedA phone number, a price, a booking button — whatever you named.
Investigating Visible, waits to be looked at
- The analytics tag stopped being thereEvery report since is based on numbers nothing recorded.
- The page got materially slowerMeasured against how this page normally responds, not a fixed number.
- Visitors bounce through extra hopsEach redirect is a delay everybody pays and credit search engines discard.
- A redirect is temporary that should be permanentSearch engines keep the old address instead of moving.
- The page title vanishedNothing to show in a search result or a browser tab.
- The page title changedSometimes a rewrite, sometimes somebody else's.
- The description wentGoogle writes its own, and usually worse.
- The social preview image wentLinks to the page now share as a blank rectangle.
- The canonical link changedThe page is now telling search engines a different page is the real one.
- Structured data disappearedStars, prices and opening hours drop out of the search result.
- The mobile viewport tag wentThe page renders desktop-width on a phone.
- A new robots directive appearedSomething is now instructing crawlers that was not before.
- Language alternates disappearedVisitors get sent to the wrong language version.
And what a failure looks like when we find one
What was checked, what came back, and how many times before we were willing to call it confirmed. A platform problem on our side is recorded as a platform problem, not as a verdict about your client's website.
The check that could not run was never counted against Harbour Law. Our monitoring breaking is not evidence the site is broken, or that it is fine.
Example check history.
The website itself
That the site loads for visitors, and how quickly.
Domain name
That the domain is not about to expire and take the site down.
The padlock in the address bar
That the certificate keeps working. When it runs out, every visitor is shown a full-page warning that the site is unsafe — and renewal is meant to be automatic, so this warns while there is still time to do it by hand.
Search engine access
That nothing is telling Google to stay away, and that the sitemap works.
Email records
That the domain still receives email, and that its records keep messages out of spam folders.
Speed for real visitors
How the site actually feels to the people who visit it — measured by Google from real Chrome users over 28 days, not from a test run here.
Links that lead nowhere
A page removed or renamed leaves every link to it as a dead end. Nothing on the site looks wrong; the visitor just hits an error and leaves.
Forwarding before the page loads
How many times a visitor is bounced between addresses before the page appears. Each hop is a delay everybody pays, and search engines pass less credit along a chain.
WordPress weaknesses
That the usual ways into a WordPress site are shut — the username list, xmlrpc, and browsable folders.
The page in a real browser
That analytics tags run, images load, and nothing is blocked as insecure — the things only a real browser can see.
And twelve things read from the page itself
These are the ones no uptime monitor will ever catch, because the page loads perfectly while every one of them is true.
How we decide to tell you
A failure has to happen three times in a row before anyone is told, and the site has to be clean twice in a row before we call it fixed. That costs a few minutes of detection time and buys something worth more: an alert you can forward to a client without checking it first.
Of the twenty things we look for on a page, six are allowed to interrupt somebody. The rest are shown to you and sent nowhere.
And when we cannot check something, we say so. A check that could not run is never reported as healthy and never counted towards an uptime figure. “We do not know” is a real answer here, and most monitoring will not give it to you.