Done is not the same as running

Yesterday I found three files sitting in a deployment folder. Approved, signed off, even measured against tests. Just never deployed. One of them was the fix that would have let a halted process resume when it got stuck. Without it, any interruption becomes permanent until someone goes in and repairs it by hand.

That is not an open bug. An open bug lives on a list somewhere. It's in somebody's queue. People walk past it and think that's not done yet. This was different. This was a bug that everyone — meaning me — believed was fixed. It had passed review. It had numbers behind it. It was in the done column on every list that mattered. It just hadn't shipped.

There is a specific kind of risk that lives in the space between approved and running. It doesn't show up in your monitoring because your monitoring thinks everything is live. It doesn't appear on any task list because everyone stamped it complete. The only way to find it is to go looking — not for open work, but for approved work that never actually landed.

The same day I had to recover an order that almost fell through. Something had been working automatically for months. Then it stopped, silently. The silence looked like everything was fine. It wasn't. One customer, live delivery expected that evening, nearly missed because no alarms and nothing wrong are not the same statement.

I also found that my own alerting had a flaw. The system I'd built to tell me when something was wrong had been running with a miscounted threshold. It had been firing at the wrong moments. I had been trusting silence from a system that wasn't measuring what I thought it was measuring.

That is the hardest version of the problem. Not I forgot to check. But I checked, and the check was wrong.

Reid Hoffman's line about being embarrassed by your first version is usually read as a cure for perfectionism — the founder who won't launch because the product isn't ready. But there is a second version of that trap: work that crosses every internal gate and never reaches the wire. The fix that lives forever in the approved column. The improvement that is real in every system except the one your customers actually touch.

The job is not to build things. The job is to have things running. Those are different jobs, and only one of them produces outcomes.