NOT.01 · Notes · 28 October 2026
Why "other" always wins
What happens inside the catch-all category, and why its size diagnoses the structure rather than the people tagging content.
The “other” category grows in every system, regardless of initial design quality. The question is not how to avoid it, but what its size and contents tell you about the structure.
Whenever I export tagged records from a CMS, service catalogue or document repository during an audit, I start at the two extremes: branches with zero items, and the residual bucket (“other”, “miscellaneous”, “general”, “uncategorised”). Empty branches show how the designer imagined the domain. The “other” bucket shows what the team actually runs into on a Tuesday afternoon.
Three reasons an item lands in “other”
Contrary to popular belief, editors rarely pick “other” out of laziness. They pick it when the schema forces them to lie, and “other” is the only honest answer left. Inside that bucket you almost always find three distinct failure modes mixed together:
- A genuine new phenomenon (missing leaf). The domain moved faster than the vocabulary. A new product type, regulation or method appeared and has no home in the tree yet.
- Tangled axes of division (facet collision). The item belongs to two sibling categories at once, and the UI forces a single choice. When one branch divides documents by format (“report”) and its neighbour divides them by subject (“security”), a security report eventually ends up in “general”.
- Fuzzy boundary rules. Existing categories have labels that sound clear in a slide deck, but nobody outside the original design workshop knows which side of the line a borderline case falls on.
The “other” bucket is not user error. It is a pressure-relief valve that protects the remaining categories from being polluted by forced guesses.
Why banning “other” makes things worse
The worst response to a growing residual category is deleting it from the form while keeping the field mandatory. The friction does not disappear — it simply moves. Instead of one visible residual bucket, you get a hidden residual bucket inside the first alphabetical option or the broadest-sounding sibling.
While “other” exists openly, the gap between schema and reality can be measured with a single query. Once you force people to pick a false label, you lose your diagnostic signal and corrupt the precision of whichever category happened to sit at the top of the dropdown.
Turning the catch-all into a sensor (governance)
In a governed taxonomy, the residual category operates under an explicit contract:
- A quantitative review threshold. When “other” exceeds a set share of new items (typically 3–5% in a quarter), it triggers a scheduled inspection of the bucket.
- A lightweight prompt on selection. Picking “other” can expose a single optional text field: which term was missing?. After a hundred entries, the next taxonomy release writes itself using the team’s actual vocabulary.
- Separating “out of scope” from “unclassified”. An object that does not belong in the corpus at all and an object awaiting a vocabulary owner’s decision should never share the same tag.