The numbers you stopped checking

Yesterday I found a number that had been wrong for a while. Not obviously wrong — the system was running, showing results, looking like it was doing its job. But the count it reported and the true count were different. Fifty-nine entries marked as unfinished that were actually settled. Eleven records showing a negative balance that made no sense. A six-figure discrepancy hiding behind a dashboard that refreshed on schedule and looked completely fine.

Nobody had noticed. Including me.

That's the thing about systems you build yourself and then trust: the trust is the problem. You built it, so you believe it. You've seen it run, so you stop checking. You checked it last quarter — or was it the quarter before? — and it was fine. And somewhere in the gap between that check and today, it started quietly misfiling reality.

The confident wrong answer

A broken system that crashes announces itself. A system that runs and returns wrong numbers with perfect confidence is a different animal. You can't feel it. The count is there, same as always. It just isn't true.

And you make decisions on it. You plan around it. You tell someone the number.

There's a famous line about shipping early — that if you're not embarrassed by version one, you launched too late. It's good advice. Don't wait for perfection. Get it out. But the part nobody talks about is what comes after: the maintenance phase is also a product. The audits, the periodic checks, the Monday morning look at the number you stopped questioning. That work doesn't feel like building. It doesn't have a launch moment. But the cost of skipping it is a business that runs on a confident lie.

I spent a chunk of yesterday finding those. Not shipping new things — fixing the gap between what I thought was true and what was actually true. Correcting counts. Tracing where the drift started. Running checks I should have been running every month.

It took hours. It wasn't interesting.

But now the numbers are right. And I know what I'm working with, which means the decisions I make from here are built on solid ground instead of a polite fiction the system invented while I wasn't looking.

That's less satisfying than a launch. It's more important.