# The Fields This Archive Cannot Count, and What That Costs > 2,592 technical facts wear 1,598 different labels and 1,754 ticket prices wear 1,355. A fact written in prose is not data, and here is what that costs. > The Study - 2026-09-24 > Canonical HTML: https://iconotheca.com/study/the-fields-this-archive-cannot-count/ ## Key points - The 293 published records carry 2,592 rows of technical evidence under 1,598 different labels. The commonest label, Dimensions, is used 168 times; most labels are used once, so almost none of that evidence can be counted or compared across objects. - The split is often a spelling. Inscription appears 41 times and Inscriptions 21, Material 31 and Materials 20. Nothing in the data says those are the same field, so any query for one silently loses the other. - The venues corpus is worse. Its 353 museums carry 1,754 admission rows under 1,355 distinct tier labels, and the ordinary adult ticket is written at least as Adult, Adults, General admission, General, Adult (18+) and Full price. - Founded is not a date. Of 353 museums, 224 give a bare year and 129 give a passage of prose holding several different events, so a median founding year computed from this field is an artefact of whichever year happens to appear first in each sentence. - Two records carry an attribution status of documented, which is not one of the attribution values at all. It is borrowed from the provenance vocabulary, where documented, disputed and unknown are the three grades. ## A fact written in prose is not data ## 2,592 technical facts under 1,598 labels ## The adult ticket has at least six names ## Founded in is not a date, it is several events ## One institution, written two ways, in the same column ## A value borrowed from the wrong vocabulary ## The count nobody could run ## What the object does better than the database ## What to fix, and what to leave alone ## Why this archive is free to read An archive that sources every sentence and labels nothing consistently can defend any claim it makes and cannot total a single column. Two thousand five hundred and ninety-two technical facts wear 1,598 labels, 1,754 ticket prices wear 1,355, and one museum appears twice in the same list under two spellings of its own name. None of that makes a fact wrong. It makes the facts incomparable, which is the specific thing a structured record exists to prevent. The makers in this archive solved it without a database: sign in the same place, mark the underside, use the same formula every time. A field is only a field when the same thing always goes in it under the same name, and that discipline is cheaper to impose on the day a row is written than on the day somebody asks how many. ## Questions **Why does an archive with sources on every claim still struggle to count its own facts?** Because sourcing and structure are different disciplines. A sourced claim proves that a statement is true; a structured field proves that two statements are about the same kind of thing. This archive enforces the first rigorously and the second hardly at all, so it can defend any single sentence and cannot reliably total a column. **What is a controlled vocabulary?** A fixed list of permitted values for a field, checked when the value is written rather than when it is read. Museums use them for object type, medium and role, most often against a published authority such as the Getty vocabularies, so that a search for one term returns every record that means it. **Does the labelling problem mean the facts in this archive are wrong?** No. Every row measured here carries its source and its value is correct. What the labelling costs is comparison: the archive cannot say how many objects carry an inscription, or what the typical adult admission is, without a person reading the rows, because the same fact is filed under different names. **Why is a median founding year of 1926 unreliable?** Because 129 of the 353 founding entries are prose holding several events, such as a collector's bequest, a building's completion and a reopening under a new name. A count that takes the first four-digit year in each entry is averaging whichever event each writer chose to mention first, not the foundings. **Should every field be a fixed list?** No. Technical evidence about art is genuinely various, and a label used once is often the most interesting row on the page. The workable rule is a closed list for the labels that recur plus an explicitly marked free-text label for the one-offs, so that counts can exclude them honestly instead of silently. ## Sources - [1] The Metropolitan Museum of Art, Open Access Collection API, object 451753, accession 64.178.2, the Abu Zayd bowl. Queried directly on 24 September 2026. The response carries medium, date and dimensions as fields and has no objectHistory and no provenance key at all. - [2] The Walters Art Museum, object record 54.456, Ewer, 644 AH/AD 1246-1247, acquired by Henry Walters 1917. Read in the in-app browser on 24 September 2026 after HTTP 403 on the plain fetch. Source for the signature formula quoted in the hero caption and for the museum's own transcription of the inscription. - [3] The Walters Art Museum, object record 48.559, Potpourri Vase (Vase potpourri a vaisseau), 1764, acquired by Henry Walters 1928. Read in the in-app browser on 24 September 2026. Source for the designer, the decorator and the museum's own marks section. - [4] The Archimedes Palimpsest Project, The History of the Archimedes Manuscript. Read 24 September 2026. Source for the scribe, the 1229 dating of the prayer book and the imaging programme behind the plate reproduced here. - [5] The Metropolitan Museum of Art, press release, The Met Makes Its Images of Public-Domain Artworks Freely Available through New Open Access Policy, 7 February 2017. Read in the in-app browser on 24 September 2026 after HTTP 429 on the plain fetch. Source for the CC0 designation and for the structured collection data released alongside the images.