Not every mistake announces itself

There's a moment before you approve something to go live when you feel the weight differently depending on what the thing is.

Some things: if they break, you'll hear about it in an hour and fix it. Cost of being wrong — a few embarrassing hours, a quick repair. Others: if they break, you don't find out for months, and by then the wrong version has been teaching the wrong numbers to everyone who looked at them.

Yesterday I shipped both kinds in the same afternoon.

The first was a redesign — a new feature in a product I run. I'd built it, tested it, and it looked right. I approved it. If I'd missed something, someone would tell me, I'd fix it. Standard loop.

The second was a correction on a set of financial figures that had been quietly wrong for months. Not dramatically wrong — wrong in the way that makes every downstream report slightly off, where you get used to the gap between two numbers that should agree. I'd found the root cause, built the fix, and ran it in a test environment until the output matched the expected result exactly. Then I applied it to the live data.

If I'd gotten that one wrong, there would be no quick iteration. The data would be wrong in the corrected direction, and nobody would know which version to trust.

The rule and its exception

"Done is better than perfect" is mostly right. Most things: ship it. Iterate. The fear of imperfection is usually the enemy, not imperfection itself. A product that ships rough and improves does more work in the world than a product that lives in your head while you refine it toward ideal.

But there is a category where this rule inverts — where the cost of a wrong version isn't friction or embarrassment but silent, compounding damage. Financial records. The logic that determines what a customer gets charged. A migration that touches years of old data. These deserve the slow path: test, verify, get a second look, then go.

The skill isn't perfection. The skill is sorting. Before you push, ask: if this is wrong, do I find out immediately and fix it? Or does the wrong version just become the new reality?

Build fast. Ship often. Know, before you approve, which category you're in.