How a true number can describe something that no longer exists
A number measured on a version that no longer exists is a true fact about something that no longer exists.
This is the cheapest question in the volume. Two words — when, and what version — and it is skipped more often than any other, because the answer feels like bookkeeping rather than evidence.
Notice what the drawing does not say. It does not say the measurement was sloppy or that anybody acted in bad faith. The number was true when it was taken and it is still true about the thing it was taken on. The thing it was taken on no longer exists.
This matters more here than it used to, and for a specific reason: the things being measured now change in ways that leave no trace a manager would notice. A prompt is edited. A tool is added. A threshold moves. None of those look like a release, none of them appear in a change log you read, and any of them can move a result more than a month of engineering would.
How to tell without reading a change log
| ask | the answer that is fine | the answer that is not |
|---|---|---|
| when was this run? | a date, immediately, without checking | “recently”, or a date that arrives after a search |
| what exactly was running? | something you could point at again and re-run | a description of the system rather than an identifier of one |
| has anything changed since? | a short specific list, or a confident no | “nothing significant” — which is a judgment, made by somebody with an interest, about the thing in question |
A result nobody can date is not a stale result. It is a result that cannot be placed at all, which is worse, because a stale number can at least be re-run on purpose.