Failure catalogue · The number is plausible, and wrong · 7 of 51
OBSERVED FAILURE MODE
Phantom margin from one bad cost line.
Margin is a difference between two numbers, so a single corrupted cost value moves it far more than it moves either input. One bad cost line manufactures margin swings across a whole category, and every percentage, ranking, and comparison downstream inherits the swing while the revenue beside it stays perfectly correct.
What we saw
At Meridian Journal Group one corrupted cost value in a print production feed produced margin swings across an entire category. Revenue was untouched and tied to its anchor, so every revenue check stayed green while the margin column moved enough to reorder which titles looked profitable. The row was eventually found by eye. Nothing in the build had asked whether a cost sat far from what its group normally costs, and nothing had computed what the category's margin would be if that one line were priced like its neighbours.
Why it passes a glance
Revenue ties, units tie, and the cost column has no anchor of its own because the file states no total for it. Margin is derived, so it inherits a green badge from inputs of which one was checked and one was not. A glance at a margin percentage carries no clue about which single line produced it, and the percentage itself looks entirely ordinary.
What addresses it
Doctrine principle 7, clean by relabelling, never by revaluing, forbids quietly repairing the value and requires the alternative: a labelled what-if computed by code under each open reading, so the corrected category margin appears on the page instead of being left as the reader's homework. The contested-metrics skill supplies the branch discipline, giving each open reading its own number rather than one figure and a caveat.
Check your own file in two minutes
- Screen the cost column the way you screen the price column, against each group's typical cost, in both directions.
- For any flagged line, recompute the category's margin with that line costed at its group's typical figure.
- Publish both figures, labelled, and name the row that separates them.
- Leave the original value in the data; the what-if belongs beside it, never in place of it.
What this does not catch
A what-if shows what a figure becomes under one reading. It does not decide which reading is right, and it cannot detect a cost that is wrong by a believable amount or a whole feed wrong in the same direction, since both leave the group's typical value looking normal.
Quick answers
- Should the bad cost line be corrected in the data?
- No, the value stays as the source stated it and the corrected figure is published beside it as a labelled what-if.
- Why does margin suffer more than revenue?
- Because margin is a difference, so an error that is small against revenue can be large against the gap between revenue and cost.
- What does a flag owe the reader?
- Its stakes: the share of every headline it feeds, and the figure that headline becomes under the alternative reading.
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.