Every number a human maintains is already wrong

Company sites are full of stale numbers. Someone typed "10,000+ users" when it was true, and it has been quietly wrong ever since — usually in the direction that undersells the product.

Every figure on this site updates itself. The product pages, the home page and the about page all read the same freshly gathered values, so what you see is at most a day old.

The rule that makes it trustworthy

Automating figures introduces a failure mode that hand-typed ones do not have: when the thing being measured cannot be reached, the number can become zero.

So the rule is that a figure which cannot be refreshed keeps its previous value. Nothing is ever replaced by a zero or an empty space because a measurement failed. A day-old number is fine. A zero is a lie, and it is a lie the site tells confidently, in large type, to everyone who visits.

This sounds obvious written down. It is not what a naive implementation does — the naive one writes whatever it got, and what it got was nothing.

Say which number you are showing

The second lesson was subtler. A figure can be gathered correctly and still be mislabelled.

One of our products was reporting a count that was a third of the real figure. Nothing was broken: it was counting something narrower than the label claimed, and the label had never been checked against it. Both numbers were true. Only one of them was the one people were reading.

Any number a person has to remember to update is a number that is already wrong.