Today I almost shipped broken structured data because I’d stopped reading output that’s usually fine. I was adding JSON-LD to this site, generated the schema, it looked right, and I nearly committed it. Then something nagged at me and I actually validated it.
The bug: the JSON-LD was built with Hugo’s jsonify but rendered into the <script> tag without piping through safeJS. Hugo dutifully escaped it, double-encoding the JSON into a string that looks like valid markup at a glance but parses as garbage. A validator flagged it instantly. My eyeballs hadn’t, because the output was 95% the shape I expected and I’d quietly trained myself to skim it.
That’s automation complacency. A tool that’s right most of the time teaches you to stop checking, and the failure hides in the fraction you stopped looking at. The more reliable the generator, the more dangerous the rare miss, because by then you’ve outsourced the verification to a vague sense that it’s probably fine.
What tipped me off wasn’t expertise, it was a habit: validate the thing rather than eyeball it, especially when a machine produced it and it looks right. Now the structured data goes through a validator every time, because “looks right” is exactly the state automation is best at manufacturing, and you can’t catch the 5% by looking harder.
