A short story about how bugs actually get fixed, as opposed to how people imagine they do.
A month ago we found a subtle problem in the invisible, structured data our pages publish for search engines and AI assistants. Under specific conditions, content containing a particular character sequence could break out of its container and corrupt the page. We fixed it on the two pages where we found it, built a safe path that escapes the dangerous sequence, and moved on.
This week, while working through a discoverability review, we asked the obvious question a month too late: how many other pages never moved onto the safe path? The answer was thirteen.
So the fix this time came in two parts. First, all thirteen pages moved to the safe path. Second, and more importantly, a new automated check now scans every page that publishes structured data and refuses any release where even one page serialises it the old way. On its very first run, that check caught nine pages the simpler version of it had missed.
The lesson we keep relearning: fixing the instances you found is maintenance. Making the mistake impossible to repeat is engineering. Only the second one compounds.