The most dangerous number is the one that used to be right

I keep a sheet with every rupee I own in it. Gold, stocks, funds, the loan I am paying off. I have looked at it most weeks for a year.

This week I checked it properly. Line by line, against my bank statement and my broker records.

Four things were wrong.

The worst one had a hole of 24,213 sitting behind it. And nobody had made a mistake. The number was correct on the day it was typed.

A wrong number shouts. A stale one says nothing.

There are two ways a number goes bad.

The first is loud. The total does not add up. A cell turns red. Something breaks and tells you it broke. I found one of those too. It had been sitting there for months, and it was still the easiest of the four to fix, because it was asking to be fixed.

The second kind keeps working.

It looks exactly like a good number. Same column. Same font. Same confident two decimals. It agrees with everything around it. It has no way of telling you that the world moved and it did not.

Mine was the cost of everything I had sold. Typed in once. True that day. Then I kept trading, and the real figure moved. The cell stayed where it was, being accurate about a day that had already gone.

Being right once is how it hides

This is the part I did not see coming.

A number that has been right before is harder to catch than one that was always wrong.

You have checked it. It passed. So the next time your eye goes down the column, it skips. Its history of being correct is what buys it permission to stop being checked.

That is why these are always old. Nobody finds a stale number in the first week. You find it in the second year, while looking for something else.

I was asking the wrong question

I kept asking whether the number was right.

You cannot answer that by looking. A stale number and a good number look the same. That is the whole trick.

The better question is where the number comes from.

If something works it out, you are fine. If someone typed it, it is a guess in good clothes, and it is only as fresh as the day it was typed.

I replaced the typed figure with one that reads its own source. The hole went from 24,213 to 28.

Not everything should move

I should be fair here, because it is easy to take the wrong lesson.

Some numbers should be frozen. A balance that has been signed off. The price on a bill that already went to a customer. Software that quietly recalculates those is worse than software that leaves them alone.

So the test is not whether someone typed it. The test is whether the thing it describes has stopped moving.

Freeze a settled fact and you are right. Freeze a live one and you have planted something that goes off later.

Why I care about this more than I used to

I want to own something. That is the whole plan.

The things I can own are made of software that people trust with money. Bills. Stock. What a shop is owed and what it owes.

Nobody buys that because the code is good. They buy it because the number at the bottom is right, and stays right, without anyone standing over it.

So the boring work is the product. Checking a total against a record I did not produce. Making sure a figure comes from somewhere instead of from a keyboard. Nobody is ever going to compliment me for it.

But it is the difference between a thing people rely on and a thing they quietly stop trusting. And you only get to own something people rely on.