An uptime monitor has one job, and it isn't alerting you. It's alerting you only when it's true.
The first false alarm trains you to check. The second trains you to glance. The third trains you to ignore, and from then on the monitor is a notification you swipe away, including the one that's real. I've watched this happen with tools I paid for. When I built MyUptimeBot, the whole design started from that failure.
The fifteen seconds
Here's what happens when a check fails.
A request from one region gets an error or times out. That's a signal, not an outage. Networks hiccup. A region has a bad moment. A server takes one slow breath. So the bot doesn't alert. It asks a second region to check the same URL. If the second check fails too, that's an outage, and the alert goes out: push, email, SMS, with what failed and where.
Total time from the site going down to a phone buzzing: under a minute. The double-check costs roughly fifteen seconds of that. Fifteen seconds is the price of an alert you can trust, and it's the best trade in the product.
Why not alert on the first failure
Because speed you can't trust isn't speed. A monitor that alerts fifteen seconds sooner and is wrong one time in ten is slower in practice, because you stop reacting to it. The user's response time is part of the system, and false alarms destroy it.
There's a version of this product that advertises "instant alerts" and ships the first-failure design because it demos well. It would also get uninstalled in a month.
The recovery alert matters as much
When the site comes back, the bot sends a second message: recovered, with the downtime duration, and the incident gets logged. This is the part people forget to build. Without it, every outage alert is an open question. Is it still down? Did someone fix it? You end up checking manually, which is the thing the monitor exists to prevent.
An alert is a promise to follow up. The recovery message is keeping it.
What else the bot won't wake you for
Certificates. An expiring SSL cert isn't an outage until it is. The bot warns at thirty days, fourteen, and seven. Three reminders, none of them urgent, so the urgent one never happens.
Content changes. You can tell the bot a word that should appear on the page. If it disappears, or unexpected text shows up, you find out. That catches defacements and broken deploys that return a perfectly healthy 200 status with the wrong page on it.
Slowdowns. Response time is tracked over hours and weeks, so you can see a site getting slower before it goes down. That's a chart, not an alert. Not everything that's worth knowing is worth a buzz.
If you're hiring me
Every product with notifications has this problem, and most solve it in the wrong direction: more alerts, more badges, more urgency. The right question is what has to be true before you interrupt someone, and the answer is usually stricter than the first draft. I build that check in before the notification, because the notification is the easy part.
An alert you ignore is worse than no alert. Build the one you'll believe.