Failure catalogue · The metric means something else · 25 of 51
OBSERVED FAILURE MODE
The average of incomparables.
A portfolio view that averages one source's metric with another's assumes both sources mean the same thing by the word. When two retailers define it differently, the blended figure describes neither and belongs to nothing. Averaging across definitions produces a number with no referent at all. Every input tied; the output cannot.
What we saw
Meridian Journal Group tracks the same channel metric across two retail partners. One partner's system counts one way, the other counts another, and each figure ties perfectly to the portal that produced it. The portfolio page averaged the two into a single headline that leadership reviewed every month. It moved with partner mix rather than with performance. Both underlying numbers were correct. The blended figure was the one thing on the page that described nothing, and it was the figure everyone read.
Why it passes a glance
Both inputs are certified, so the average inherits the look of certification. An average of two tied figures has no anchor of its own to fail against, and no check exists for a total nobody asserts. The blended headline is also the number leadership asks for, which makes its absence feel like a gap rather than a refusal.
What addresses it
The contested-metrics skill settles the fork per source, stamps each figure with its own basis, and refuses to place two differently based figures on one axis or into one sum. Rates roll up by recomputing from summed components, never by averaging the ratio. Principle 10, a metric without its definition is not a number, states plainly that a definition never travels across sources.
Check your own file in two minutes
- For each source in the portfolio view, write down the definition its system uses, in one sentence.
- Check whether any two of them differ in base, population, or period.
- Replace the blended average with the components, each labelled with its own basis.
- Where a rollup is genuinely needed, recompute it from the summed numerator and summed denominator instead of averaging the rate.
What this does not catch
Per source definitions stop the silent blend. They do not make sources comparable, and a leadership team that wants one number still has to choose which definition the portfolio adopts. Restating a partner's figure onto your definition needs data the partner may not export.
Quick answers
- Why can I not average two sell-through rates?
- Two reasons. The definitions may differ, so the inputs are not the same measure, and even where they match, a rate rolls up by recomputing from the summed numerator and denominator. Averaging a ratio is arithmetic on the wrong object.
- What should a portfolio page show instead?
- The components, each with its own basis printed beside it, and a rollup only where every source shares one definition and the numerator and denominator can be summed.
- Is a blended number ever acceptable?
- When every source has been restated onto one recorded definition and the page says so. The problem is not the blend, it is a blend that hides a definition fork inside a single figure.
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.