Maintenance
What 21 years of building taught us about what breaks
The site is healthiest the day we hand it over.
After twenty-one years of building for the web, we can tell you the site we hand over on launch day is the healthiest it will ever be. Almost none of the sites we have watched die went down the way people fear. Everyone braces for the big breach, the front-page hack.
What actually kills a site is quiet, slow, and so boring that nobody schedules time for it, right up until it has already taken them down. The most embarrassing example leads the list, and it is the one nobody sees coming.
The most common cause of a major site going dark is a lapsed SSL certificate. Every visitor hits a full-screen red warning, and traffic drops to zero the same hour. It has taken down companies with infinite money and brilliant engineers: Google Voice, Spotify's podcast platform, Microsoft Teams, LinkedIn twice, all taken out by a calendar reminder nobody set.
Ericsson let a certificate lapse and took 4G offline for thirty million people in the UK. A five-minute fix, a week earlier, would have caught it.
A broken layout you can see. A certificate with three weeks left, an abandoned plugin with a known hole, a backup silently failing for four months, these all look fine from the front.
Plugins are the second killer.
A plugin is somebody else's code running on your site, and when the developer walks away the code stays behind with a door propped open that nobody is watching. That is where most of the risk now lives, not in core. You find out when a customer calls, or worse, does not.
The fix took five minutes, a week too late.
Every failure here is preventable and none of it is clever. A competent person could have renewed that certificate in five minutes, a week earlier. The gap is never skill. It is that a vulnerability gets scanned within four hours of disclosure, and the door stays open because no one whose job it is to close it was watching.
Attackers scan within four hours of disclosure. The average owner takes fourteen days to patch. For two weeks the door is on a list being actively worked through, and the owner does not know it exists. That gap is exactly what Keystone closes, and it's why our pre-launch checklist keeps getting used long after launch.