AreaAudit

Which languages each generator says it supports, and where

The four shapes a language figure comes in

A published language figure takes one of four shapes, and they are not points on a single scale. Two of them can be checked against a particular market and two cannot, and the size of the number has nothing to do with which. As of 2026-09-12.

Four shapes of figure, ranked by what a reader can doA named set with its total printed can be searched, quoted and re-counted. A set without a total can be searched. A count can be quoted and checked against nothing. A lower bound can barely be quoted. The size of the number has no bearing on which shape it is.Most useful to a readerA set with its totalprintedSearchable, quotable and re-countable. The rarest shape.A set with no totalSearchable. Any figure has to be produced by counting.A count with no setQuotable and unverifiable. Answers nothing about one market.A lower bound with no setSurvives every change to the product, so it proves nothing.Least useful, and the most common
Fig. 1 Sorting a column of language figures by size puts the least checkable claims at the top, which is why this atlas does not sort it.
The four shapes, and what a reader can do with each. Recorded 2026-09-12.
ShapeSearch for one marketQuote a figureReproduce the figure
A named set with its total printedYesYesYes
A named set with no totalYesBy countingYes
A count with no setNoYesNo
A lower bound with no setNoYes, vaguelyNo

Inclusion rule. The shapes a spoken-language claim takes across the vendors this atlas reads. A vendor publishing nothing has no shape and is recorded as an absence. Order. Most useful to a reader first.

1Searchability is the property that matters

A producer has one question: is my market in there. A set of thirty names answers it in a second. A count of three hundred does not answer it at all, and no increase in the number ever will.

So a column of language figures should not be sorted by size. Two cells holding a set and a number are different kinds of object, and ranking them together suggests the larger number is the better disclosure when it is usually the worse one.

2A printed total is the one shape that is both

A set with its own count stated lets a reader search for their market and quote a figure without doing arithmetic, and lets them check the figure by counting. It is the cheapest improvement any vendor with a list could make.

It is also falsifiable, which is the point. A total over a visible set can be contradicted by its own page, so somebody had to be careful about it. A lower bound over no set cannot be contradicted by anything.

3The lower bound is engineered never to expire

Writing a figure with a plus after it converts a number into a claim that survives every change to the product. Languages can be added and removed and the sentence stays true, so its presence proves nothing about how recently anybody looked.

It also makes two vendors incomparable. One saying more than a hundred and another saying more than a hundred and thirty-five may hold inventories differing by one language or by fifty, and both sentences are true across that whole range.

4What to do when the shapes disagree

Several vendors publish a set and a figure that do not match. Quote the set: it can be recounted on any day by any reader and will give the same answer, which is the only property that makes a number worth publishing.

Record the other figure beside it rather than dropping it. The gap between a headline and a list is itself a finding, and it points in both directions: some vendors overstate their documentation and at least one understates it.

An explainer about method and structure, not a count. No vendor number appears here that is not also in the register. The sourced material is on the register. Related: Which variant, Blanks as evidence.