Some of you got woken up over the past few weeks for sites that were fine. You'd open the laptop at 2am, load the page, and there it was, up. After that happens three or four times you stop trusting the alerts, and an alert you don't trust is worth nothing.

What we assumed, and what the data said

Moonitor checks your site from Helsinki. Before it opens an incident it asks a second machine in Los Angeles to look too. The idea is that if our Helsinki box is having a bad day, Los Angeles disagrees and we keep quiet.

So we assumed Los Angeles had gone flaky and was calling things down that weren't.

It hadn't. We pulled the records, and every incident from the previous week had been independently confirmed by Los Angeles. The box was healthy and barely working. It wasn't inventing anything.

The problem was ours, and it was subtler than a broken box. Both machines were checking at the same instant. So when a site hiccuped for a second, both of them genuinely saw the hiccup, agreed with each other, and we treated that agreement as proof of an outage.

On top of that, we opened an incident on the very first failed check. One unlucky sample was enough to wake you up.

One customer's site made this painfully obvious. It sits behind a web server that returns an error on roughly one request in five, at random. We tested from three continents and got the same thing everywhere.

What changed

A fault now has to clear two bars before we'll alert you.

The first is agreement. A majority of the checking locations that answer have to see the same thing. One machine's opinion isn't enough.

The second is the one that actually matters: it has to still be broken a few seconds later. When a check fails we wait, then check again from every location. Site's back? It was a blip. We log it and leave you alone. Still down? You get the alert.

The same rule runs in reverse now, which fixes a separate complaint we've had: a site that flickers back up for one check no longer gets marked recovered, then down again ten minutes later. No more alert storms out of one wobbly server.

A third continent

We've added a checking location in Johannesburg, alongside Helsinki and Los Angeles. Different provider, different network.

Part of that is tie-breaking. With two locations they could only agree or deadlock, and a deadlock meant we froze and did nothing at all. On one monitor that happened 23 times in a month, and we were blind to it the whole time. A third location settles the argument.

A fourth continent is being added at the moment.

Your uptime number stays honest

When the locations disagree, or when a fault vanishes on the second look, we record the check and show it in your history. It does not count against your uptime percentage. If we can't demonstrate a fault happened, we're not putting it on your report card.

Hey Sumo-lings,

Quick update on what we've been building. A lot of it came straight out of your emails and questions, so — genuinely — thank you.

We now double-check every outage multiple countries.

This is the one we're most proud of. The honest truth about uptime monitoring is that a lot of "downtime" alerts aren't downtime at all. A bad network route, a firewall that mistook our checker for a bot, one dodgy DNS answer — from a single location, every one of those looks exactly like your site being down. And you're the one who gets woken up for it.

So we stopped watching from a single location. Moonitor now checks from multiple countries, on separate networks and different continents. Before we open an incident, we re-run the failing check somewhere else entirely. If the second country can still reach you, we don't call it an outage, and we don't call you. It doesn't count against your uptime either — we're not going to dent your numbers over something we can't prove.

And it never works the other way. If one of our checkers goes unreachable, we still alert you. We'd rather send an alert we're unsure about than sit quietly on a real outage.

Behind Cloudflare? It just works now.

You can set custom request headers (including a browser user agent) and pick exactly which status codes count as healthy. If a WAF was challenging our checks and giving you phantom downtime, that's sorted.

DNS monitoring that actually catches changes (Team & Max)

We remember the last record we resolved and alert you the moment it moves — you don't have to tell us what to expect. Good for catching a hijack, or someone quietly re-pointing a record.

Status pages got a lot better

- Your own domain, with SSL set up automatically

- Your logo, your colours — proper theming

- White-label: switch off "Powered by Moonitor" entirely

- Your users can subscribe to updates by email

- Maintenance windows, so planned work doesn't page anyone

- SMS alerts

That's the lot for now. Keep the feedback coming — a good chunk of the above started life as a question in this very tab.

— Chris & the Moonitor team