Ambiguous schemes (topic, task, audience) depend on judgment

What it is

Ambiguous organization schemes group by topic, task, or audience. Membership has no single correct answer; someone has to judge. Does “expense claims” belong under Finance or Administration? Is “for new hires” an audience bucket or a task bucket? Both can be defended. Rosenfeld and Morville call these schemes ambiguous not because they are sloppy, but because category boundaries come from language and culture, not from a checkable coordinate.

Ambiguous schemes are the default for most sites, because people usually arrive with a purpose or a topic rather than a letter or a date. The cost is that judgments go stale, differ across people, and drift when the team turns over.

Why it happens

Topic, task, and audience are interpretations of content, not fields that can be read off the object. Interpretation leans on prototypes: the cataloguer’s picture of a “typical expense claim” sets the boundary, and edge cases are pushed into Other. Different stakeholders keep different prototypes, so the same item is fought over or wanted by no one. Task schemes also break when the workflow is redesigned: the process splits, the old task name stays, and the scent dies. Audience schemes assume people can recognize which class they belong to; picking the wrong class is walking into someone else’s content.

With no objective coordinate to arbitrate, consistency exists only as ongoing judgment discipline: who decides, how disputes escalate, where new items go by default. Without that discipline, an ambiguous scheme becomes a set of private drawers within a few iterations.

Studying it

Measure whether judgments converge, and whether different user groups share the same boundaries.

• Paradigms: open card sorts for spontaneous topics; closed card sorts for whether existing categories are understood the same way; then re-instruct the same deck as tasks versus audiences to see whether the cut changes.
• Independent variables: organizing dimension (topic / task / audience), category names, presence of Other.
• Dependent variables: inter-category agreement (e.g. adjusted Rand index), proportion of contested items, and whether task success depends on guessing the cataloguer’s judgment.
• Methodological note: staff sorters inflate agreement because they share organizational language. Include target users. Low-agreement items are not noise; they are evidence that the boundary is undesigned.

Where it stops holding

In law, medicine, and aviation, statutory classifications can look topical but behave as exact: membership is set by an external authority and the team has no judgment to exercise. Highly personalized streams (“recommended for you”) are no longer a shared ambiguous scheme but a private one per person; a public tree cannot be recovered by card sorting. When the use context of an item changes every time, a fixed topical bucket will systematically misfile; move to facets or search rather than splitting ambiguous categories further.

Applying it

• Declare whether this layer cuts by topic, task, or audience. Do not mix the three logics in one layer.
• Write an inclusion rule and two or three positive and negative examples for each category, especially where edge cases go and where they do not.
• If Other exists, give it a emptying cycle: past a volume threshold it must be split or merged; it must not become a permanent bin.
• Verify by giving twenty new items to two people who did not design the scheme and recording disagreements. Recurring disagreements on the same boundary mean changing the rule or the name, not training the judgment away.

Related

• Within the group: G1.02.1 Exact schemes (alphabetical, chronological, geographical) have one right place · G1.02.3 The same content can live in multiple schemes · G1.02.4 The scheme depends on what the user already knows
• Adjacent: G1.03 Taxonomy and hierarchy · G1.11 Architectural scalability and evolution · Q2.11 Card sorting
• Search terms: ambiguous organization schemes · topical organization · task-based navigation