Taxonomy Design

Concept of the skill

Taxonomy design is the discipline of constructing a controlled classification system — a category tree (possibly augmented by facets) that organizes a set of items so they can be browsed, found, and reasoned about. Drawing from Ranganathan's faceted classification (PMEST: Personality, Matter, Energy, Space, Time), Aristotelian genus-species hierarchy, and the W3C SKOS data model, it treats classification as a retrieval contract between the system that organizes and the readers who navigate, not as a private mental map of the author.

Replaces ad-hoc category trees that mirror the author's mental map with classification systems designed for retrieval by the navigator. Solves the problem that the main failure mode of taxonomies is mixing incompatible organizing principles — categorizing by audience in one branch, by tool in another, by lifecycle phase in a third, by domain in a fourth, by risk in a fifth — producing a tree where related items scatter across distant branches and the navigator must learn the author's idiosyncratic grouping logic before finding anything. The discipline is to prefer shallow, stable, mutually understandable structure, add facets when one tree cannot express all valid access paths, and not turn every semantic relation into a hierarchy. Bowker & Star's Sorting Things Out documents how classifications without explicit retrieval-task analysis become abandoned bureaucratic structure — the lesson generalizes to every taxonomy in software (file folders, skill categories, navigation trees, tag systems, knowledge-graph categories).

Distinct from ontology-modeling, which owns formal axioms and reasoning semantics — this skill owns human-governed classification; ontology is formalization, taxonomy is organization. Distinct from knowledge-modeling, which chooses the representation paradigm (graph/frames/rules/hybrid) — this skill works inside the classification paradigm once chosen. Distinct from semantic-relations, which types individual concept edges (IS-A, PART-OF, thematic) — this skill governs the category system and assignment rules. Distinct from information-architecture, which arranges pages/navigation/labels/wayfinding for a user-facing experience — this skill is the underlying classification structure that information architecture surfaces. Distinct from folksonomies (emergent user tagging — different governance model with different trade-offs; folksonomies are bottom-up, taxonomies are top-down-controlled). Taxonomy design is to a knowledge surface what a public library's call-number system is to its collection — the books haven't changed, but the system determines which books a researcher finds when looking under 'biology' versus 'medicine' versus 'public health.' A library that categorizes by acquisition date (the author's mental map) is unfindable; one that categorizes by subject with cross-references and facet headings (retrieval contract for the navigator) is searchable. Dewey Decimal is a hierarchy with facet-like subdivisions; Library of Congress is more faceted; both are taxonomies, both have explicit retrieval-task analyses behind their structure. The wrong mental model is that taxonomy = hierarchy — that classification means choosing a tree and assigning items to it. It does not. Adjacent misconceptions: that deeper hierarchies are better organized (they are not — shallow trees with facets often outperform deep trees because the navigator scans fewer levels; Ranganathan's faceted classification is the alternative to strict hierarchies); that mixing organizing principles is fine (it is not — sibling categories must share the same organizing principle, or the navigator cannot predict where items will be; "by domain, then by tool, then by lifecycle" mixed within one tree produces unpredictable placement); that synonyms can be sibling categories (they cannot — synonyms must be aliases via skos:altLabel pointing to one canonical skos:prefLabel; otherwise the same concept duplicates and items split across both); that every semantic relation belongs in the hierarchy (it does not — related_to and other associative relations are not hierarchy; only IS-A passes the substitution test for parent-child placement); that catch-all "misc" buckets are acceptable (they are not — they signal that the classification's primary organizing principle doesn't cover the items, and the right response is either adjusting the principle or identifying a new top-level category, never adding "misc" and moving on); and that taxonomies are static (they are not — they require governance: assignment-rule documentation, periodic audits for drift, alias maintenance as terminology evolves).

Coverage

Design controlled classification systems for skills, docs, product catalogs, navigation trees, knowledge graphs, and tags. Covers hierarchy shape, facet selection, synonym control, category granularity, assignment rules, governance, and drift cleanup. Use SKOS-grade distinctions: broader/narrower for hierarchy, related for association, preferred labels for canonical terms, alternate labels for aliases.

Philosophy of the skill

A taxonomy is a retrieval contract. It should make things findable by the people or agents who browse it, not mirror the author's private mental map. The main failure mode is mixing incompatible organizing principles: by audience, by tool, by lifecycle phase, by domain, and by risk all in the same tree.

Prefer shallow, stable, mutually understandable structure. Add facets when one tree cannot express all valid access paths. Do not turn every semantic relation into a hierarchy.

Method