Governance
What a provenance label means, what the external schemes permit, how a term is retired, and the one thing this vocabulary will not claim.
Provenance ranks the source, not the accuracy
Every term carries where its content came from and when. The label describes the source and nothing else: a term lifted verbatim from an upstream dictionary is vendored whether or not the upstream got it right. Reading it as a confidence score is how confident wrong answers get made.
| Basis | What it records |
|---|---|
| vendored | Copied unedited from a named upstream dictionary at a stated snapshot. |
| authored | Written here, from scratch, by a named editor. |
| evidenced | Written here, but derived from measured usage in a named corpus. |
| standard | Adopted from a published external standard, with its identifier carried. |
Usage is a band, never a count
Where a term carries a usage figure, the page shows a band and a five-step meter. The number underneath it is not published, in the markup or in the JSON-LD. Those figures derive from Material Bank’s warehouse and DesignShop’s index, and they are somebody else’s business. The band carries the signal a reader needs - is this term common or rare - and nothing more.
External schemes, and what may be shown
Getty Art & Architecture Thesaurus
ODC-By 1.0. Redistributing a subset of AAT identifiers and prefLabels as skos:exactMatch targets and as displayed labels is exactly what §3.1 grants, commercially included, subject only to notice.
Contains information from the J. Paul Getty Trust, Getty Research Institute, the Art & Architecture Thesaurus, which is made available under the ODC Attribution License.
Uniclass 2015
CC BY-ND 4.0, with three binding guardrails. Tables are sourced from the free downloads and never from the NBS API, whose terms are contractual and revocable. Codes and titles are reproduced verbatim or not at all, and no code is ever coined in the Uniclass namespace. Only a true subset is carried: §4(b) makes a database holding a substantial portion of the licensed contents into Adapted Material, which ND forbids us to share, so there is no browse-all-EF page here and no downloadable EF dataset.
Uniclass classification codes and titles © NBS Enterprises Ltd, reproduced unmodified under CC BY-ND 4.0 (https://creativecommons.org/licenses/by-nd/4.0/). Uniclass is published by NBS at https://uniclass.thenbs.com.
A negative result is evidence
Every Style and Period term either carries a verified AAT mapping or a dated negative result naming the scheme searched, the query, the endpoint and the date. That is a stronger commitment than requiring a mapping, because it turns a gap in the world’s best art-and-architecture thesaurus into published evidence a reader can repeat rather than take on trust. All 25 Style terms satisfy it today: 21 mappings and 4 dated negatives.
AAT’s consumer-interiors coverage is partial rather than absent, and the gaps are specific, which is what makes them evidence rather than a complaint. maximalism and parametricism return nothing in any domain. traditional returns cultural properties, dances and a bookbinding format. transitional returns skyscrapers, a typeface group and Stone Age periods, and it is the one that matters: it is among the most-used values in that corpus and the thesaurus has no concept for it.
An absent mapping and a dated negative are opposites. One means nobody has looked yet; the other means somebody looked, on a date, with a query anyone can run again. The crosswalk table on a term page shows them as separate rows for that reason.
Concepts classify, anchors retrieve
The package also ships style ANCHORS: retrieval prose describing what a photograph of each style would contain. They are a separate export and they are versioned separately, and no term page shows one. A concept and an anchor have different lifetimes - an identifier must never move, while an anchor is tuned to whichever embedding model is current and is expected to be rewritten. Printing an anchor beside a definition would make retrieval prose read as part of the term’s meaning, which is the conflation the vocabulary exists to undo. Anything relying on matching against anchor text is relying on something deliberately outside the stable-identifier promise.
Where the colours come from
Palette hue is the one axis on this site where colour carries meaning, and the swatches are not chosen by eye. Each is the representative hex that @instruments/colorscope 4.0.0 publishes for that colour family, which is the hue facet’s own declared provenance. The library is a build-time dependency, so its version belongs on this page in the same way a term’s snapshot date belongs on the term.
Supersession, never deletion
Stable identifiers are the whole promise of this package. A term whose parent cannot be stated as an is-a sentence is on the wrong axis, and the fix is supersession: the term keeps its identifier, is marked deprecated, and gains a pointer to what replaced it. Consumers reading the old identifier keep working and migrate when they choose.
application.bath and application.kitchen are the worked example. Both are rooms sitting inside a surface axis, which is the diagnostic for leakage. Neither is deleted. The claim underneath them was never application to a surface but suitability for a space, and that claim is now carried by a relation instead.
A deprecated term still renders in full, at normal contrast, with its supersession pointer directly under the identifier. It is still a term somebody cited.
Declining is a first-class answer
On the open-world axes - Style, Period, Mood, Palette, Place - there is no other term, and a crosswalk that cannot map a source value declines rather than guessing. On a fuzzy axis other costs a classifier nothing, absorbs every hard case, and quietly becomes the largest class, at which point the axis reports high coverage while carrying no information. A Style axis with 40% nulls tells you something true; a Style axis with 40% other tells you nothing and looks fine.
The largest single decline in the crosswalk is one word. DesignShop carries Modern and Contemporary as separate values at almost identical volume, together 38% of its tagged schemes. Its merchandisers plainly distinguish the two; nobody wrote down how. This canon mints both style.modernism and style.contemporary and declines the row, on the record. It will look like a coverage failure in any dashboard that counts mapped rows, and it is the opposite.
A limit this vocabulary asserts
A vocabulary can be correct, shared, and worth publishing, and still not be predictive of an individual. Measured on the DesignShop corpus on 20 August 2026: enriching a taste signal with style and category made prediction worse, by twelve points at broad taste, because a richer description of a person’s taste is also a richer description of the room type they were shopping for.
This canon promises correctness and shared meaning. It does not promise prediction. Nothing built on it, and nothing published about it, may imply otherwise.
The reasoning behind every decision above is recorded in ADR 0001 in the package repository. The vocabulary itself is on the axes.