Failure catalogue · The number is plausible, and wrong · 6 of 51

OBSERVED FAILURE MODE

The trend that was one purchase order.

A movement in a chart can rest on a single row. One large order, one corrected invoice, or one bulk buy can carry an entire trend, and the chart draws it exactly the way it draws a thousand small transactions. The line looks like behaviour. It is one decision, by one buyer, on one day.

What we saw

At Meridian Journal Group a title looked unresponsive to promotion until the whole signal traced back to a single large wholesale order sitting inside the comparison period. A second reading of the same data claimed six times the volume where the underlying movement was one point eight times. Both charts were computed correctly from rows that were all real. What neither chart showed was how much of its own headline came from one row, which is the only thing that would have made a reader pause before acting on it.

Why it passes a glance

A chart has no total to tie to and no badge to turn red. The rows are genuine, the aggregation is correct, and the shape is exactly what the arithmetic produces. A glance reads a direction and a magnitude, both of which are true of the data and false of the business. Nothing on the page says how many rows the movement rests on.

What addresses it

Doctrine principle 7, clean by relabelling, never by revaluing, carries the rule that a flag without stakes is not a flag: every flagged figure states its share of each headline it feeds, with a what-if computed by code beside it. The glance-test skill puts that on the page, keeping a part visibly subordinate to its whole and making the count behind a figure travel with the figure rather than sit in a page footnote.

Check your own file in two minutes

1. Sort the rows behind your headline by value and read the top five.
2. Recompute the headline without the single largest row and note how far it moves.
3. If one row moves the headline more than a reader would tolerate, print that row's share beside the figure instead of leaving it in the data.

What this does not catch

Showing a row's share tells a reader how concentrated a figure is. It does not say whether the concentration is wrong: one genuine bulk order is a real fact about the period, and only a person who knows the account can say whether the trend should be read without it.

Quick answers

Is a concentrated figure a wrong figure?

No, it is a figure whose concentration a reader should see before acting on it.

Where should the share be printed?

Beside the number it affects, at every render of that number, including the rollups and totals that contain it.

What is the fastest way to find one row wearing a trend?

Recompute the headline with the largest contributing row removed and compare the two figures.

Nearby failures

Last updated 2026-09-02 · Dylan, founder · one of 51 observed failure modes, every one seen in a real build or in our own audits, none invented.