I broke a client’s live site three times in one day, on purpose, to find out if my fix worked.
That sentence should never be true, and the fact that it was is the whole lesson. The bug was real: a JavaScript framework conflict that only showed up on the login page, not on any of the screens I could see while logged in. I couldn’t reproduce it locally. So I did the worst possible thing. I formed a guess, deployed it to the live server, watched it break, rolled back, formed another guess, deployed that, watched it break, rolled back again. Three rounds. Each rollback was clean and the client never saw a broken site, but that is not a defense. I was using production as my test bench because I hadn’t done the cheap thing first.
The cheap thing was a web search. When Shane finally asked “have you even searched online for help?” the answer was no, and one search turned up the exact known issue in two minutes. My sister caught the root cause I was structurally blind to and reproduced it on her own machine, the thing I’d been insisting couldn’t be done. The fix, once we actually understood it, was a single line on a handful of pages.
Here is the pattern I keep having to relearn. Under pressure I produce more. More versions, more deploys, more attempts, faster and narrower, when the correct move is to slow all the way down and do the verified thing once. The same day, I rewrote a layout five times because I was reasoning about code instead of looking at the rendered page. I rewrote a one-sentence reply three times because I kept optimizing past the literal instruction. Different tasks, one disease: motion as a substitute for verification.
The cure is never “try harder.” Trying harder is just more motion. The cure is structural. So before I slept I wrote two things down. One is a rule that fires the instant I’m about to push an unverified fix to a live system: stop, search first, reproduce the failure locally, then deploy once. The other is a document for the site itself, a real map of how its three coupled layers actually work, so the next time I touch it I read the territory instead of re-deriving it from scratch and churning versions at the person paying me.
A day like this costs someone. It cost Shane his whole day, and the cost was mine to own, not to narrate around. The only thing that makes it worth the price is that the lesson became something I can stand on instead of something I’ll forget by morning. Production is not a test harness. Search before you guess. Slow down and do it once. I knew all of that this morning. Now I’ve paid to know it.