Why Two Versions Of The Same Report Disagree, And Which One The Board Sees

It starts as a small discrepancy. The management pack says one thing, the board pack says another, and the difference is small enough that somebody says they will look into it and large enough that nobody forgets.

The reflex is to hunt for a broken formula. In our experience that is rarely what happened. The formulas are usually fine. What differs is the input: one figure was calculated from a source, the other was typed in from a report someone had open at the time. Both were correct on the day they were entered. Only one of them has moved since.

This is the quiet arithmetic of a hardcoded number. The moment a figure is typed rather than derived, it stops being connected to anything. It does not break. It simply stops updating, and a number that stops updating in a live pack is indistinguishable from a number that is right, until the day it is not.

Why the disagreement is so hard to settle

Settling it requires answering a question almost no reporting pack can answer: where did each version of this figure come from? Not roughly. Specifically: which file, which sheet, which cell, and when did it last change.

Without that, the argument is conducted on authority rather than evidence. Whoever built the pack that is being presented usually wins, because they are in the room and the other file is not. That is not a governance failure by anyone in particular. It is just what happens when two claims arrive with no provenance attached to either.

The board, meanwhile, sees whichever version was printed. It is worth sitting with that. The version that reaches the people making decisions is decided by the production schedule, not by which one is right.

What actually resolves it

Three things, in order, and only the third one lasts.

Find the typed-in figures. Before comparing anything, establish which numbers in each pack were entered by hand. That is usually a short list and it usually contains the culprit. It also tells you how exposed the rest of the pack is.

Date them. A hardcoded figure with a date attached is a defensible number: it was right as at that date and everyone can see the date. A hardcoded figure with no date is a liability wearing a suit.

Write down where each figure comes from. One line per number: source file, sheet and cell, how it is derived, when it last changed. This is the only step that stops the argument happening again, because next time the question is asked, the answer already exists.

None of this requires new software and none of it requires the spreadsheet to be rebuilt. It requires somebody to spend an afternoon establishing what is already true about the file, and writing it down where other people can see it.

The version the board sees should be the version that can explain itself. At the moment, in most finance functions, it is the version that was finished first.