A number with a gap in it is still a number — if it says so
The morning meeting starts at nine. At seven, one property has not finished its night audit, one till has not exported, and one feed has arrived twice. The picture is incomplete, and it will be incomplete every morning for as long as the group exists. The only question is whether the reporting admits it.
Complete is a property of the source, not of the day
In a group, “is yesterday closed?” has no single answer. Night audit at one property closes at a different hour than at another. The restaurant platform exports on its own schedule. The accounting system posts when the accountant posts. Money from the payment gateway settles a day or two later.
So completeness cannot be a flag on the date. It has to be a status on each source, for each property, for each business day: arrived, arrived in part, not yet arrived, arrived and rejected. A reporting layer that keeps that status can say what a figure is made of. One that does not can only guess, and it will guess “complete” because that is what the screen looks like.
Three ways to lie with a partial number
There are three ways a partial number lies. The first is to show yesterday’s figure as though it were today’s: nothing crashed, so nothing alerted, and the number on the screen looks exactly as fresh as it did when it was true. The second is to show zero. A property whose feed has not arrived posts no revenue, the group total drops, and the owner reads a collapse that is really a delay. The third is to fill the gap with an average or a forecast without saying so, which is the most sophisticated of the three and the hardest to catch.
All three have the same cause. The status of the source was known somewhere in the pipeline and thrown away before the screen. Keeping it is most of what honesty costs.
Quarantine, do not zero
When a feed arrives and something in it is wrong, the tempting fix is to drop the bad rows and load the rest. The total is then wrong by an amount nobody recorded.
The better rule is quarantine. The suspicious rows are held to one side, the figure they would have touched is marked as partial with a named reason, and a named person is told. The rows are not adjusted to make the total balance, because a total that balances by dropping something is a total that the auditor will unpick at year end.
That gives a figure four honest states. It is complete: every source for this property and this business day has arrived and passed its checks. It is partial, and the screen says which source is missing and since when. It is stale, and the last good figure is shown with the time it was true rather than dressed up as current. Or it is absent by design, because the screen is switched off for this group and says so, rather than showing a plausible zero.
An incomplete figure is fine as long as it says so on its face.
The first run always sends
Honesty about partial data has a consequence for delivery. If the morning summary waits for the day to be complete, it will never go out, because the day is never complete at seven. So the seven o’clock run goes regardless of what is missing, with each source’s status alongside it. The morning meeting needs a signal more than it needs perfection.
Later runs go out only when a source has actually landed and moved the figure. The delivery rules themselves are the subject of the next piece in this series.
Dark screens are part of the honesty
Partial data has a larger cousin: data the group does not keep at all. A budget-versus-actual screen needs a signed budget, and in more groups than will admit it there is none. On our largest deployment a dozen screens are dark for exactly that reason. The data behind them is not kept anywhere.
Those screens say why they are dark. It is the same principle at a different scale: a reporting layer that shows a plausible number where there is no data has stopped being reporting.