Knowing is not the same as fixing

I killed something today that I'd known was broken for five weeks.

Not broken in a quiet way. A monitoring script had been catching it on every single run - more than three thousand times - logging it faithfully, doing exactly what it was built to do. The system worked perfectly. The problem did not get fixed.

The gap between detection and repair is something founders learn slowly, and usually expensively. We build the alert. We see the alert. We write the note. We move on to the thing that's on fire right now. The alert runs again tonight. We'll see it again. We'll move on again.

I'm not describing negligence. I'm describing the arithmetic of running several things at once. Every day there's a finite number of fires. The ones with live flames get water. The smoldering ones get reprioritized for tomorrow, which becomes the day after, which becomes five weeks and three thousand logged detections and a note somewhere in a task list that nobody has touched.

The quiet tax

What I actually fixed today was cheap. Twenty minutes. The thing had been running in a broken state for five weeks because the note said "address this when the other fire is out" - and the other fires kept coming.

But here's the thing about tolerated brokenness: it doesn't sit still. It generates waste - real, measurable waste - every single day. And it sits in the back of your mind too. Every time you pass something adjacent to it, you remember. The reminder is brief. The cost is small. Multiply it by thirty-five days and it adds up to something you never quite accounted for.

The hardest part is that you almost never measure what it costs you to carry a problem - only what it would cost to fix it. So the problem stays on the list. The carrying cost is invisible. The fix feels optional.

"Real artists ship" gets quoted at founders constantly. What it usually means is stop waiting for it to be perfect, just put it out. But there's a harder version of it that nobody quotes: the real artists also clean up. They kill the things that stopped working. They close the loops that opened months ago and nobody ever came back to.

Not because someone forced them to. Because they know that a monitoring script logging the same failure three thousand times isn't a system working - it's a system telling you, very patiently, that something is waiting.

I closed one today. The fix took twenty minutes.