It’s a familiar meeting. The dashboard on screen says 91% compliant. Someone opens the spreadsheet they exported that morning and it says 87%. Ten minutes disappear into which one is right, and the actual agenda item never gets discussed.
The instinct is to blame the data. Usually the data is fine. What’s wrong is that “compliant” has been defined twice, by two different bits of the system, and the definitions have drifted apart.
It happens innocently. The register needs a rule for what counts as compliant. Later the dashboard needs the same rule, and the quickest path is to write it again where it’s needed. Then the export needs it, and the monthly report, and the notification email. Five copies of one rule — and the moment somebody adds a new status, or decides maintenance items shouldn’t count, four of the five get updated.
Nobody notices immediately, because each screen is internally consistent. The disagreement only surfaces when two of them appear side by side, in front of someone senior.
The discipline that fixes this is unromantic: every figure that matters has exactly one definition, written down once, and every screen asks that definition rather than restating it. When the rule changes — a new status, a category excluded, a scope narrowed — it changes in one place, and everything that reports on it moves together.
It’s worth being explicit about the edges too, because they’re where the arguments are:
There’s a related trap in trend reporting. It’s tempting to reconstruct past months from current records — take the dates you hold now and work backwards. It produces a beautiful chart and it’s quietly wrong, because renewing an item overwrites its previous dates. Re-run the same report in March and February’s figure will have changed.
Stored monthly snapshots avoid this. Each month’s number is captured and then frozen. It doesn’t move, because history didn’t move. A board looking at a twelve-month trend can rely on the fact that the figure they approved in June still says the same thing in December.
Compliance reporting has one job: to be believed. A number that changes depending on where you look is worse than no number at all, because it teaches everyone in the room to discount the system — and once that habit forms, the good data gets ignored alongside the bad.