The document says what its marks stand for

In plain words

Written for anybody, with nothing taken for granted. It says what the rule is for and what it changes on the document you end up with.

Somebody who cannot see the page listens to it. The software reading it out has to know what is a heading, what is a paragraph, what is a table, which cell sits under which column heading, and above all in what order to say it all. A page that records only where the ink went says none of that, so the reading aid works through it in the order the marks were drawn, which is rarely the order a person would read them.

The answer is a second list running alongside the drawing: this group of words is a title, that one a paragraph, this block is a table and here is its heading row. The drawing does not change. What changes is that the file now knows what it drew — which also lets software pull the figures back out, and lets a search engine index the text properly.

Most of it is built as the page is laid out. What stays with you is the outline: which part of the document is a section, which picture means what, and in what order it should be read.

For a developer

The same thing without the detour: the objects, the constraints, and where the line falls between what is produced and what is declared.

Level A of part 3 asks for a structure hierarchy and refuses the document without one. The accessibility standard asks for the same tree and for three things around it: a title, the language of the content, and every shape enclosed in an element that says what it is for.

Built for you: the tree root and its parent tree, the role map, the mark on the catalogue that says the document is tagged, the numbers tying each drawn run to its element, the tab order of a page carrying links or fields, the document language taken from the tree's own, and — under an accessibility claim — the request that a reader show the title rather than the file name. A table lays out its own structure, heading cells included, as it is drawn.

Declared by you: the tags themselves, the nesting, the reading order, the text alternative of a picture, the language, the title, and the marking of running heads and rules as decoration rather than content. Nothing is inferred from the layout — a heading is a heading because the calling program said so.

Two rules refuse the document under an accessibility claim: a shape no element of the tree names, and — under the second part — a field that acts when something happens to it while saying nothing of itself. Since the tree names links and form widgets and nothing else, a document claiming that standard carries no free-standing shape at all.

Clause by clause

The published text behind each requirement, what the engine measures the document against, and what it leaves to a validator.

The accessibility rules on the tree, the title and the language are reported before writing rather than refused at writing: a document that claims the standard and carries no tree is written, and a validator declines it. Ask for the report and the answer is there before the file exists.

The second part of the accessibility standard asks that a way into the document name a piece of its structure rather than a page. The engine writes page destinations, so every internal link, bookmark and opening view of such a document is reported, and the only way to clear the report today is to carry none. Structure destinations are planned for a future version.

Marking decoration as decoration is asked of the calling program and measured by nobody here. The table rules are read and reported: the shape of a table, one caption at one end of it, rows that all cover the same width, no cell over another, and every heading cell either saying what it heads or being named by the cells that use it.

What the engine does on its own

Nothing on this list has to be asked for. It comes out of a document that claims the standard, whatever else the calling program says.

• Builds the tree root, the parent tree and the role map, and marks the catalogue as tagged.
• Ties every drawn run to the element that owns it, as the page is laid out.
• Lays out a table's own structure, heading cells included.
• Writes the document language from the tree, and sets the tab order of a page carrying links or fields.
• Asks a reader to show the title rather than the file name, under an accessibility claim.
• Reports the tree, the title, the language and the tables before anything is written, and refuses a shape no element names.

What you supply

Nothing on this list can be guessed. The engine states what it was given and declines to invent the rest, which is why the document is refused rather than written with a value nobody chose.

• The outline: which element is a heading, a paragraph, a figure.
• The reading order, where it differs from the drawing order.
• A text alternative for every picture that carries meaning.
• The language of the content, and the title of the document.
• Running heads, folios and rules marked as decoration.

The published texts behind it

Glossary

standard

A rule argued out in committee, published under a number anybody may buy and read, and identical for every firm claiming it. A claim to follow one can therefore be checked against the text. A way of working that merely spread because it worked is a habit of the trade: useful, widespread, and answerable to no text at all.

veraPDF

A free program that checks whether a file really is what it claims to be, against the published rules. It is the tool archives use, so it is the one used here: a document is not called compliant because we say so, but because veraPDF passed it.

CMYK

The four inks a printing press lays down: cyan, magenta, yellow and black. A screen mixes light instead, so a document going to a press states its colours in inks, not in screen colours.

tagged PDF

A PDF that carries the structure of its own text alongside the drawing: this is a heading, this is a table, this cell belongs to that column. It is what lets a reading aid, or a program pulling data back out, make sense of the page.

structure tree

The list, held inside a tagged PDF, that says what each part of the page is and what it belongs to: this block is a section, this line its heading, this run of cells one row of that table. It is the shape of the document, kept apart from the ink.

reading order

The order a document is meant to be read in, which is not always the order it was drawn in. On a page in two columns the ink may go down one column and then the other, or jump between them; software reading the page aloud follows the stated order rather than guessing.

text alternative

A sentence written into the document to stand in for a picture, read out to somebody who cannot see it. It says what the picture shows, not what it is: a photograph of a shop front, not a photograph.

bookmark

The list of headings a PDF reader shows down the side of a document, each line leading straight to its place in it. It is the table of contents the software owns rather than the one printed on a page, and a long document without it is one nobody can find their way around.

request

One call to the service: you send what the document should say, and get the document back. Your bill counts these calls, one for each document. The number of pages inside a document is never counted.

Every word the site explains