How it works, drawn.
Seven pictures, one loop. Nothing here is a promise we cannot keep; each one is something the service does today.
Every minute, we read your site.
From the outside, the way a visitor sees it. We ask, your site answers, and we write down what came back: the answer code and how long it took. Sixty times an hour, without pause.
When it breaks, we say exactly what we saw.
The first sentence of every alert is written by code from the reading, never by a model. It states only what was measured, in words a person would use.
Then what is likely, and only that.
A second paragraph says what is probably causing it and what to do. It is written by a model from the measured facts, is always marked as likely, and is allowed to say "we cannot tell." A confident guess would cost your trust the first time it was wrong.
Everyone signed in is told.
Not one inbox that might belong to someone who left. Everyone who has signed in for your organization gets the email, and whoever inherits the job later joins by proving the domain again, with nobody to ask. If the organization connected a Slack channel, the same message posts there too, with the same wording.
It shows on your page, honestly.
Your status page is drawn from the readings, nothing else. A bad minute stays a bad minute; every percentage carries the count behind it, and nothing is hidden or backfilled, including on our own page.
You prove the site is yours.
One record added to your domain's DNS, the same proof every registrar accepts. Nothing is watched until we see it, so nobody can put up a status page in your name, and nobody can point us at a website they do not own.
Someone watches the watcher.
An outside service, on other people's computers, reads our own runner every minute. If it stops writing readings, they notice within a few minutes and write to us. We tested it by switching ourselves off: the email arrived in seven minutes, and so did the one saying we were back.