Check rules
What each rule that Check applies looks for, why it matters to someone using
assistive technology, and how to fix it. A finding's HelpUrl links to its
rule's entry here. Each rule names the clause it comes from: a clause of PDF/UA-1 (ISO 14289-1) or, for a warning
PDF/UA doesn't check, of WCAG 2.2. The two house rules, NoActions and
NoEmbeddedFiles, come from no standard. The clause only says where a rule comes from; passing
it doesn't mean the document conforms to the standard. For conformance, use veraPDF or PAC.
The rules DocxDocument.Check applies to Word documents come last, from
Word: ImageAltText on. They come from WCAG 2.2, since PDF/UA doesn't cover Word.
Most fixes belong in whatever generates the document: a template, a report library's settings, or the export options of the application it comes from. The recipes cover common generators.
DocumentLanguage
Finds: a document with no language. Category: Language. Severity: Error. From: ISO 14289-1 7.2.
Why it matters: a screen reader picks its voice and pronunciation from the document's language. Without one, it guesses, and English text read with a French voice, or the other way around, is hard to follow.
How to fix it: set the document language, such as en-US, in the generator. In the PDF it's the catalog's /Lang
entry. Text in another language inside the document sets its own language on its element or marked content.
LanguageTags
Finds: a language on the document, a structure element or marked content that isn't a well-formed language tag
(BCP 47), such as english, en_US, en- or a single space. An empty language means an unknown one, so it isn't
reported here; for the document, DocumentLanguage reports it. When the value is a language's
English name, such as Spanish, the finding gives its tag (es). Category: Language. Severity: Error.
From: ISO 14289-1 7.2.
Why it matters: a screen reader looks the tag up to choose its voice and pronunciation. One it can't parse is treated as no language at all, so the text is read with whatever voice it falls back to.
How to fix it: use a language code with a hyphen before the region, such as en, en-US, fr-CA or zh-Hant-TW.
Generators usually take it from a culture or locale setting; an underscore (en_US) is a sign the value came from a
platform locale name rather than a language tag.
TextLanguage
Finds: in a document with no language of its own, text that gets no language from anywhere: an element's text, alt
text, replacement text or expansion when neither it, an element around it, nor the marked content covering the text sets
a language; marked content's own alt text, replacement text or expansion; annotation contents and form field names
outside such an element; bookmark titles; and an XMP title or description for the default language (x-default). A
document that sets no language anywhere is left to DocumentLanguage, since one document language
fixes it all. Category: Language. Severity: Error. From: ISO 14289-1 7.2.
Why it matters: a screen reader reads each piece of text in its language. Text that has none is read with whatever voice the reader falls back to, so a French passage can come out in an English accent.
How to fix it: set the document language. Where a passage is in another language, also set it on that passage's
element, such as /Lang (fr-FR) on its P, and its contents inherit it.
DocumentTitle
Finds: a document with no title, or one that doesn't tell the viewer to show its title, so the viewer shows the file
name instead. Category: Document. Severity: Error. From: ISO 14289-1 7.1.
Why it matters: the title is the first thing a screen reader announces and what a window or tab shows. A file name
such as report-final-2.pdf says little about what the document is.
How to fix it: give the document a title in its metadata (dc:title, and the Info dictionary's /Title), and tell
the viewer to show it: /DisplayDocTitle true in the catalog's /ViewerPreferences.
FigureAltText
Finds: a figure with no alt text or replacement text. Category: AltText. Severity: Error. From:
ISO 14289-1 7.3.
Why it matters: a screen reader can't describe a picture. Without alt text it says nothing, or only "graphic", and the information the figure carries is lost.
How to fix it: give each meaningful image alt text (/Alt on its Figure element) that says what it shows or
means, such as "Trail map of the ridge loop, 4.2 miles". An image that's only decoration should be marked as an artifact
instead, so it isn't announced at all.
FormulaAltText
Finds: a formula with no alt text or replacement text. Category: AltText. Severity: Error. From:
ISO 14289-1 7.7.
Why it matters: a formula drawn as glyphs or an image reads as a jumble of symbols, or as nothing. Alt text such as "x equals negative b plus or minus the square root of b squared minus 4 a c, all over 2 a" is how a screen reader user gets the math.
How to fix it: give each Formula element alt text (/Alt) that reads the formula aloud, or replacement text
(/ActualText) when the formula is plain text drawn specially. Equation editors often have a setting to export it.
MediaClips
Finds: a media clip, such as a video a Screen annotation plays on click or when its page opens, with no content type
(/CT) or no alternate text (/Alt). Every clip counts: each one a selector rendition offers, and each one the
annotation plays on click or on an event. The second clip's findings have media-type-2 or media-alt-2 in their
path, and so on.
Category: AltText. Severity: Error. From: ISO 14289-1 7.18.6.2.
Why it matters: a viewer needs the content type to know whether it can play the clip, and alternate text is what a screen reader reads in its place, so a reader who can't play or see it still knows what it shows.
How to fix it: give the media clip data dictionary a /CT such as (video/mp4), and /Alt with a language and a
description for each language: [(en-US) (A walk along the ridge trail)].
MeaningfulAltText
Finds: alt text that's a file name, such as IMG_2041.jpg or trail-map final.png, or a placeholder word such as
image, picture, graphic, icon or untitled, with or without a number. It looks at alt text on elements and in
marked content. Category: AltText. Severity: Warning. From: WCAG 2.2 1.1.1, which asks that alt text
serve the same purpose as the image; PDF/UA only asks that it's there.
Why it matters: this alt text passes FigureAltText, but a screen reader user hears "image 3"
or a file name and learns nothing about what the picture shows.
How to fix it: write what the image shows or means, such as "Trail map of the ridge loop, 4.2 miles". Most generators fill alt text from the file name or a default when none was given, so set it where the image is added.
LinkText
Finds: a link whose text is only "click", "click here", "here", "this", "this link", "link", "more", "read more",
"learn more" or "more info", with or without trailing punctuation such as "Read more >". It looks at the link element's
alt text, replacement text and text, including its child elements', and at the link's contents, and finds the link only
when none of them says more. Category: Annotations.
Severity: Warning. From: WCAG 2.2 2.4.4, which asks that a link's purpose be clear from its text and context;
PDF/UA only asks that a link has a description.
Why it matters: this link passes LinkDescriptions, but screen reader users often list a
page's links to move around it, and a list of "click here" links says nothing about where each goes.
How to fix it: make the link's text say where it goes, such as "Download the ridge loop trail map", or give the link contents that do.
ReadingOrder
Finds: a paragraph, heading, list item body, table cell, table of contents entry or block quote read after one
below it in the same column of the same page, such as a paragraph read before the heading above it. A block counts as
in the same column when over half the narrower block's width lines up with the other's, so reading on from the foot of
one column to the top of the next isn't found. Rotated text is left out, and so are footnotes (Note) and captions,
since they're read where they're tagged rather than where they're drawn. For the same reason, a table's cells are
compared only with each other, so a table that floats to the top of the page isn't found, but a text box or sidebar
drawn higher up and read where it's anchored is. Category: Structure. Severity: Warning. From: WCAG
2.2 1.3.2, which asks that content be read in an order that keeps its meaning; PDF/UA asks for a logical reading order
but leaves judging it to a person.
Why it matters: a screen reader reads the structure tree in order, wherever the text is drawn, so its user hears the paragraph before the heading that introduces it, or the rest of a sentence before its start.
How to fix it: put the elements in the structure tree in the order they're meant to be read. In a generator, that usually means adding the content in reading order; in a PDF editor, drag the tags into order in its tags panel.
UntaggedContent
Finds: text, images, paths and shadings that are neither tagged nor marked as artifacts, page by page. Category:
Content. Severity: Error. From: ISO 14289-1 7.1.
Why it matters: assistive technology reads the tag tree. Content outside it may be skipped, or read out of order, so a reader can miss a heading, a warning or a price.
How to fix it: turn on tagged output in the generator, and check that custom drawing code and pasted-in content are tagged or marked as artifacts too. Content that repeats or only decorates, such as running headers and footers, page numbers, rules and backgrounds, is marked as an artifact, so a screen reader doesn't announce it on every page.
ArtifactNesting
Finds: an artifact (/Artifact marked content) inside tagged content, and an artifact that holds tagged content
(marked content with an MCID), counted page by page. Category: Content. Severity: Error. From: ISO
14289-1 7.1.
Why it matters: content is either read (tagged) or skipped (an artifact), and a screen reader can't do both. When one is nested in the other, a page number inside a paragraph may be read in the middle of a sentence, or real text inside a decoration may be skipped.
How to fix it: close the tagged content before drawing the artifact, and the other way around. The usual cause is a
generator that draws a header, footer or watermark while it's in the middle of writing a tagged paragraph. Text that
form XObjects draw outside the structure is reported by UntaggedContent.
LinkDescriptions
Finds: a link with no description in its contents (/Contents). Visible text or alt text on the Link element
doesn't count, since PDF/UA asks for the description on the link itself. Category: Annotations. Severity:
Error. From: ISO 14289-1 7.18.5.
Why it matters: a screen reader announces a link by its description. A link on an icon or image with none reads as "link", with no hint of where it goes.
How to fix it: give each link annotation a /Contents entry that says where it goes, such as "Trail conditions
page".
AnnotationTags
Finds: a link annotation not tagged with a Link element, a form field's widget not tagged with Form, any other
annotation not tagged with Annot, and any of them not tagged at all, unless it's hidden or a pop-up; a Form that
sets no field role (PrintField Role) and holds anything but one widget; a TrapNet annotation; and a tagged
PrinterMark. Category: Annotations. Severity: Error.
From: ISO 14289-1 7.18, with each finding naming its clause: 7.18.1, 7.18.2, 7.18.4, 7.18.5 or 7.18.8.
Why it matters: a screen reader announces an annotation by the element that tags it: "link", "check box", "comment". Tagged with the wrong element, a link reads as a paragraph and a form field may not be offered for filling in. Trap networks and printer marks are for print production, and a reader shouldn't come across them at all.
How to fix it: tag each link annotation with its Link element, each widget with its own Form, and comments,
stamps and other annotations with Annot. A Form for a group of widgets, such as radio buttons, sets its PrintField
Role; a Form without one holds just its widget, with the field's label in an element of its own. Leave printer
marks out of the structure tree, and remove trap networks from the file.
AnnotationDescriptions
Finds: an annotation other than a link or a form field, such as a comment or a stamp, with no contents and no alt
text on the element that tags it. Pop-ups, hidden annotations, printer marks and trap networks have nothing to read,
so they're left out. Category: Annotations. Severity: Error. From: ISO 14289-1 7.18.1.
Why it matters: a screen reader announces a comment or stamp by its description. Without one, a reader hears "annotation" and nothing else, while a sighted reader sees the note or the "Approved" stamp.
How to fix it: give each annotation a /Contents entry that says what it holds or means, or alt text on the
Annot element that tags it. Links are covered by LinkDescriptions, and form fields by
FormFieldNames.
TabOrder
Finds: a page that has annotations and whose tab order (/Tabs) isn't S, the order of the structure tree.
Category: Annotations. Severity: Error. From: ISO 14289-1 7.18.3.
Why it matters: someone using the keyboard moves between links and form fields with Tab. With the structure order, they meet them in the same order as the text around them; in row or column order, they jump around the page.
How to fix it: set /Tabs /S on every page with links, form fields or other annotations. Most generators have a
setting for it, often called tab order or "use document structure".
FormFieldNames
Finds: a form field with no accessible name: no tooltip (/TU), and not every widget tagged by a Form element
with alt text, which PDF/UA accepts instead. Category: Forms. Severity: Error. From: ISO 14289-1 7.18.1.
Why it matters: a screen reader announces a field by its accessible name. Without one, a reader hears "edit text" and has to guess what to type.
How to fix it: give each field a tooltip (/TU) that says what it's for, such as "Emergency contact phone".
TableHeaders
Finds: a header with no scope that no cell names, and a cell that names a header that doesn't exist.
Category: Tables. Severity: Error. From: ISO 14289-1 7.5.
Why it matters: a screen reader reads a cell together with its headers, such as "Distance, Ridge loop: 4.2 miles". A header it can't tie to its cells leaves each value without its meaning.
How to fix it: give each TH a scope (Row, Column or Both), or give its cells a /Headers list that names
the header's ID. Make sure every ID a cell names belongs to a header in the table.
TableStructure
Finds: a row (TR) outside a Table, THead, TBody or TFoot; a row group (THead, TBody, TFoot) outside a
Table; a cell (TH, TD) outside a TR; a table, row or row group holding something it can't; a second THead,
TFoot or Caption in one table; a THead or TFoot in a table with no TBody; and a caption that isn't the
table's first or last child. Category: Tables. Severity: Error. From: ISO 14289-1 7.2.
Why it matters: a screen reader moves through a table by row and column, and tells the reader where they are, such as "row 3, column 2". It can only count rows and columns that sit where a table keeps them, so a cell outside a row, or text loose in a table, breaks the count and the headers it reads with each cell.
How to fix it: keep the standard nesting: Table, then optional THead, TBody and TFoot (at most one head and
one foot, with a body whenever there's either), then TR, then TH and TD. Text goes inside a cell, and the caption
goes first or last. Most generators get this right for tables built with their table feature; the usual cause is a
layout made of positioned text or nested tables used for spacing.
TableGrid
Finds: a cell that overlaps one spanning down from an earlier row, a cell that spans past the table's last row, and
a row that covers a different number of columns from the first row. Cells are laid out as a viewer lays them out: row
by row, each in the first column no cell above still covers, counting RowSpan and ColSpan. Category: Tables.
Severity: Error. From: ISO 14289-1 7.2.
Why it matters: a screen reader reads a cell with the headers of its row and column, and says how many rows and columns the table has. When the cells don't line up, it pairs values with the wrong headers, or reports a column that isn't there.
How to fix it: check each RowSpan and ColSpan against the table as it's drawn: a cell that spans two rows takes
a column in the row below, so that row has one cell fewer. A generator that merges cells usually sets the spans; one
that draws merged cells as separate boxes needs its merging turned on, or the extra empty cells tagged.
HeadingLevels
Finds: a heading that skips a level, or one that comes before any H1. Category: Headings. Severity:
Error. From: ISO 14289-1 7.4.2.
Why it matters: screen reader users move through a document by its headings, and the levels tell them how sections
nest. A skipped level, such as an H3 straight after an H1, hides where they are.
How to fix it: start with H1 and go down one level at a time. In the generator, map styles to heading levels by
their place in the outline, not by how they look.
HeadingTypes
Finds: a document that uses both unnumbered headings (H) and numbered ones (H1 to H6), and an element, or the
structure tree root, that holds more than one H. The document's first heading says which kind it uses, so each
heading of the other kind is reported. Category: Headings. Severity: Error. From: ISO 14289-1 7.4.4.
Why it matters: numbered headings give their level outright. An H gets its level from how deep its section is
nested, so it only works when each section (Sect) has one. A document that mixes the two gives a screen reader two
outlines that disagree.
How to fix it: pick one kind. Numbered headings are simpler and what most generators write. With H, give each
heading its own Sect and nest the sections the way the outline goes.
NoteIds
Finds: a note (Note, a footnote or endnote) with no ID, and a note with the same ID as an earlier one.
Category: Structure. Severity: Error. From: ISO 14289-1 7.9.
Why it matters: a reference in the text points to its note by the note's ID, so a screen reader can jump to the note and back. Without an ID, or with one two notes share, the reference leads nowhere or to the wrong note.
How to fix it: give each Note element an /ID that no other note uses, such as fn1, fn2. Generators that
tag footnotes usually number them; check that the IDs survive when documents are merged or sections are reused.
FontsEmbedded
Finds: a font that isn't embedded. Category: Fonts. Severity: Error. From: ISO 14289-1 7.21.4.1.
Why it matters: without the font program, a viewer draws the text with a substitute, which can change its width, its spacing and which characters it can show, so the document may not read as written.
How to fix it: embed every font, at least the characters the document uses. Most generators have a setting for it; on Linux build servers, also install the fonts the document asks for.
FontsMapToUnicode
Finds: a font whose text uses character codes that don't map to Unicode, or that its ToUnicode map sends to
U+0000, U+FEFF or U+FFFE, which give no text. Category: Fonts. Severity: Error. From: ISO 14289-1
7.21.7.
Why it matters: a screen reader, search, and copy and paste all need to know which character each glyph is. Without the map, the text reads as nothing or as gibberish, though it looks right on screen.
How to fix it: use fonts that come with a Unicode map, and generator settings that write a ToUnicode CMap for each
font. A font that maps some codes but not others usually means a symbol or ligature the generator draws specially.
NoMissingGlyphs
Finds: a font whose text shows its .notdef glyph, the empty box a font draws for a character it lacks. It finds the
codes a simple font's encoding names .notdef; in a CID font with an Identity encoding, CID 0 or a code its
CIDToGIDMap sends to glyph 0; and in an embedded TrueType, Type 1, CFF or OpenType program, a code whose glyph the
program lacks. A glyph past the program's last glyph counts only in drawn text, not in invisible text such as an OCR
layer. Category: Fonts. Severity: Error. From: ISO 14289-1 7.21.8, and 7.21.4.1 for a glyph the program
lacks.
Why it matters: the text a screen reader reads may be there, but sighted readers see empty boxes in its place, and the two no longer agree.
How to fix it: use a font that has every character the text needs, or embed the whole font instead of a subset that leaves some out. Boxes often appear when text is pasted in from another font or uses a symbol the font lacks.
FontDictionaries
Finds: a Type 0 or TrueType font whose dictionaries break a PDF/UA font rule, one finding for each kind of problem:
- its CIDSystemInfo names a different character collection from its CMap's (7.21.3.1);
- it's an embedded TrueType-based CID font with no
CIDToGIDMap(7.21.3.2); - its CMap is neither built into every reader nor embedded, an embedded CMap gives two writing modes, or an embedded CMap refers to a CMap that isn't built in (7.21.3.3);
- it's non-symbolic TrueType with an encoding other than
MacRomanEncodingorWinAnsiEncoding, or symbolic TrueType with anEncodingentry (7.21.6).
Category: Fonts. Severity: Error. From: ISO 14289-1 7.21; each finding names its own clause.
Why it matters: these entries tell a reader which glyph and which character each code stands for. When they're missing or disagree, readers can pick different glyphs or characters, so the text can look right in one viewer and read wrong in another.
How to fix it: these come from the generator, not the document's content. Update the generator or its PDF/UA settings, or embed fonts as TrueType with WinAnsi encoding where it offers the choice.
FontPrograms
Finds: an embedded TrueType (FontFile2), Type 1 (FontFile), CFF or OpenType (FontFile3) program that
disagrees with its font, one finding for each kind of problem:
- the program can't be read, so nothing in it can be checked (7.21.4.1);
- a CID font's
CIDSetleaves out a CID the program has, or a Type 1 font'sCharSetleaves out a glyph the program has (7.21.4.2). The finding says how many CIDs theCIDSetleaves out and lists the first five. When they're all glyphs with no outline, such as the empty slots that a subset keeps so glyph numbers stay the same, it says so. The fix is then for the generator to write aCIDSetthat marks them, or none; - a glyph width in the font dictionary (
Widths, orWandDW) differs from the program's by more than one unit in a thousand, for text that's drawn (7.21.5). A CID font's widths are compared when its encoding is Identity-H or Identity-V, where each code is its CID; - a non-symbolic TrueType program has neither a (3,1) nor a (1,0) cmap, or a symbolic one has several cmaps and no (3,0) one (7.21.6).
An OpenType program is checked through its font, whatever its outlines: a TrueType font finds each glyph in the
program's cmap, a CIDFontType2 font through its CIDToGIDMap, and a Type 1 or CIDFontType0 font in the CFF
outlines.
Category: Fonts. Severity: Error. From: ISO 14289-1 7.21; each finding names its own clause.
Why it matters: readers pick glyphs and space text from these. Widths that disagree draw text in one place and measure it in another, and a cmap that doesn't suit the font leaves readers to guess which glyph a code stands for.
How to fix it: these come from how the generator embeds fonts. Embedding the whole font instead of a subset, or updating the generator, usually fixes them.
LayerConfigurations
Finds: an optional-content configuration (a set of layers a viewer can switch to) with no name, and one that turns
layers on or off by itself, such as by zoom or for printing (/AS). Category: Document. Severity: Error.
From: ISO 14289-1 7.10.
Why it matters: a viewer lists configurations by name, so one without a name can't be chosen or announced. Layers that appear and disappear on their own change what a reader gets without the reader doing anything.
How to fix it: give each configuration in /OCProperties a /Name, and leave out /AS.
EmbeddedFileNames
Finds: an embedded file whose file specification doesn't give its name both as /F and as /UF (Unicode): one in
the document's list of embedded files (/EmbeddedFiles), or one attached through a comment (a FileAttachment
annotation's /FS), whose finding names its page. Category: Document. Severity: Error. From: ISO
14289-1 7.11.
Why it matters: a viewer and a screen reader show and announce the attachment by its name, and /UF is the one
that can hold any character; without both, a name can be lost or garbled.
How to fix it: write both entries in each file specification, with the same name.
NoDynamicXfa
Finds: a dynamic XFA form, one the viewer builds its pages from as the file opens (/NeedsRendering).
Category: Forms. Severity: Error. From: ISO 14289-1 7.15.
Why it matters: the pages of a dynamic XFA form don't exist until a viewer that supports XFA builds them, so most viewers and screen readers see nothing, or only a "please wait" page.
How to fix it: export the form as a static PDF form (AcroForm), with tagged fields.
AccessibilityPermission
Finds: an encrypted document whose permissions don't let assistive technology extract its text (bit 10 of /P).
Category: Document. Severity: Error. From: ISO 14289-1 7.16.
Why it matters: a screen reader gets the text the same way copying does. With that permission off, it gets nothing, however well the document is tagged.
How to fix it: keep "content accessibility" (extraction for accessibility) allowed in the generator's security settings. It can stay on even when copying is off.
NoReferenceXObjects
Finds: a reference XObject, a form that shows a page of another file (/Ref), in the pages' resources, in their
annotations' appearances (/AP), or in the forms in them, drawn or not. Category: Content.
Severity: Error. From: ISO 14289-1 7.20.
Why it matters: the page it shows lives in another file, outside this document's tags, so a reader can't reach it, and most viewers only show the form's own fallback drawing.
How to fix it: import the page's content into the document instead of referring to it.
NoActions
Finds: JavaScript, actions run on events such as closing or printing, and an open action that does more than go
to a page. Category: HouseRules. Severity: Warning. From: a house rule, not a standard.
Why it matters: scripts and automatic actions can change a document as it's read, run differently in each viewer or not at all, and surprise someone who can't see what changed. A generated document rarely needs them.
How to fix it: turn off scripting and document actions in the generator, and keep the open action to going to a page and view, if anything.
NoEmbeddedFiles
Finds: each embedded file. Category: HouseRules. Severity: Warning. From: a house rule, not a
standard.
Why it matters: an attachment is easy to miss and may not be accessible itself, so information that lives only in it can be lost to some readers.
How to fix it: put the information in the document itself, or leave the attachment out if it isn't needed. A document that needs an attachment, such as the source data, can leave this rule out of the set it checks.
RoleMap
Finds: a custom structure type that doesn't map to a standard one, a mapping that loops back on itself, and a
standard type that's remapped. Category: Structure. Severity: Error. From: ISO 14289-1 7.1.
Why it matters: assistive technology only knows the standard types, such as P, H1 and Table. A custom type,
such as Callout, means nothing to it unless the role map says which standard type it stands for, so an element of an
unmapped type may be read as plain text or skipped. A loop never reaches a standard type, and remapping a standard
type, such as P to Span, changes what every element of that type is for the tools that follow it. PDiffy
follows PDF 1.7 and ends a mapping chain at the first standard type, so its other checks still treat a remapped P
as a paragraph.
How to fix it: map each custom type to the standard type closest in meaning, such as Callout to Div or
BodyText to P, in the structure tree root's /RoleMap. A chain of custom types is fine as long as it ends at a
standard one. Leave the standard types out of the role map. Most generators build it from their style names, so the
fix is usually a style mapping setting. The finding's path names the type, with loop or standard after it for the
other two problems: role-map > Callout, role-map > Callout > loop, role-map > P > standard.
In a PDF 2.0 file, an element can name the namespace its type is in. PDF 2.0's types are standard in PDF 2.0's
namespace, every MathML element is standard in MathML's, and a type in any other namespace, such as a generator's own,
needs a mapping in that namespace's /RoleMapNS. The path then names the namespace too:
role-map > Callout in https://example.com/tags. PDF 2.0's namespace may map its own types to PDF 1.7's, such as
FENote to Note, so readers that know only PDF 1.7 can follow them. That mapping isn't found, but a type both
versions have may map only to itself, so P to Span is found, and so is a mapping from one PDF 2.0 type to another.
An element without a namespace uses PDF 1.7's types, so an Aside without one is found: Aside is only a PDF 2.0
type.
XmpMetadata
Finds: a document with no XMP metadata, XMP that can't be read, XMP that doesn't say the document is PDF/UA-1
(pdfuaid:part 1) or says so with a prefix other than pdfuaid, and XMP with no title (dc:title) or with a title
written as a plain value instead of a list of language versions (rdf:Alt).
Category: Document. Severity: Error. From: ISO 14289-1 5, and 7.1 for a missing XMP stream and a
missing or plain title.
Why it matters: the PDF/UA identifier is how a viewer, a library catalog or a checker knows the document claims
to be accessible, and the XMP title is the one PDF/UA asks for. DocumentTitle also accepts the
Info dictionary's title, which viewers show, so a document can pass it and still fail here.
How to fix it: turn on PDF/UA output in the generator, which writes the XMP stream with
<pdfuaid:part>1</pdfuaid:part> and a dc:title. XML lets any prefix stand for a namespace, but PDF/UA names
pdfuaid, and some checkers look for it. Write the title inside <rdf:Alt><rdf:li xml:lang="x-default">, as XMP
defines it; some readers don't find a plain one. That finding's path is xmp-title > plain.
ParentTree
Finds: a structure tree with no parent tree when its elements own marked content or annotations, a parent tree whose
keys repeat, are out of order or reach ParentTreeNextKey (one finding each, such as document > parent-tree-problems > duplicate-keys), and each element whose marked content or annotation the parent tree gives to another element or to
none. That finding lists each one as p2:7->P (page 2, MCID 7, which the parent tree gives to a P),
p2:xobject:0->none (MCID 0 inside a form XObject that page 2 draws), p2#1->none (the page's first annotation, which
has no entry) or p2:xobject->none (an image or form XObject the element owns, which has no entry). Category:
Structure. Severity: Error. From: ISO 32000-1 14.7.4.4.
Why it matters: a screen reader that lands on text or a link finds its tag through the parent tree. When the entry points at the wrong element, the screen reader reads the text with another element's role, language or alt text, and when it's missing the text has no tag.
How to fix it: give each page a /StructParents number, and put an array under that number in the parent tree
with the owning element at each MCID's index. Give each tagged annotation a /StructParent number whose parent tree
entry is its element. Marked content inside a form XObject uses the form's own /StructParents number when it has one,
and the page's otherwise.
TaggedPdf
Finds: a document that isn't marked as tagged (MarkInfo Marked), one that says its tags may be unreliable
(MarkInfo Suspects), one with no structure tree, structure elements with no parent entry (P) or one that names an
element other than the one that lists them, elements only the parent tree reaches because their parent doesn't list
them, and elements that list a child the structure tree already lists elsewhere
(structure > Sect > shared-children). Category: Structure. Severity: Error. From: ISO 14289-1 7.1, and
6.2 for a document that isn't marked.
Why it matters: assistive technology only uses the tags of a document marked as tagged, and treats suspect tags as guesses. Without a structure tree there's nothing to read but the raw text, and a tool that walks up the tree from an element without a parent, or with the wrong one, can't tell where it belongs. A tool that walks down the tree never reaches an element its parent doesn't list. A child listed twice is read twice, or in the wrong place.
How to fix it: turn on tagged output in the generator. Suspects is set by tools that add tags after the fact;
check the tags, then clear it. An element without a parent entry, with the wrong one, or missing from its parent's
/K is a generator bug: each StructElem needs a /P pointing to the element or root that holds it, and that
element, and only that one, should list it among its kids.
PdfVersion
Finds: a file header past PDF 1.7, such as %PDF-2.0, and a catalog /Version past it, which upgrades the file.
Category: Document. Severity: Error. From: ISO 14289-1 6.1.
Why it matters: PDF/UA-1 is defined on PDF 1.7. A PDF 2.0 file follows PDF/UA-2, whose rules differ, so checks written for PDF/UA-1 may not apply to it.
How to fix it: write PDF 1.7 for PDF/UA-1, or, for a PDF 2.0 file, check it against PDF/UA-2 and leave this rule out of the set.
ListStructure
Finds: a list item (LI) outside a list (L), a list body (LBody) outside a list item, a list holding something
other than lists, list items and a caption, a list item holding something other than a label (Lbl) and a body, and a
list caption that isn't the list's first child. Category: Lists. Severity: Error. From: ISO 14289-1 7.2.
Why it matters: a screen reader announces a list and how many items it has, such as "list, 4 items", and reads each item's label and body together. Items outside a list, or text loose in it, are read as plain text, without the count or where each item starts.
How to fix it: put each item in an L, and each item's text in an LBody, with its bullet or number in an Lbl.
A nested list goes in the LBody of the item it belongs to, or straight in the outer L. A list's caption goes first.
TocStructure
Finds: a table of contents item (TOCI) outside a table of contents (TOC), a table of contents holding
something other than tables of contents, items and a caption, and a caption that isn't its first child.
Category: Lists. Severity: Error. From: ISO 14289-1 7.2.
Why it matters: a screen reader reads a table of contents as a list of entries that lead into the document. Entries outside it, or other content mixed in, read as plain text, so the list and how many entries it has are lost.
How to fix it: put each entry in a TOCI inside the TOC, a nested part in a TOC of its own inside the outer one,
and the caption first. Most generators build this from their table of contents field; check that field's tagging
setting.
Word: ImageAltText
Finds: an image, chart, diagram, shape or embedded object in a Word document with no alt text (wp:docPr has no
descr, or an empty one) that isn't marked as decoration. A shape with text in it is left out, since a screen reader
reads its text, and so is an ActiveX control, which has its own label. Category: AltText. Severity: Error.
From: WCAG 2.2 1.1.1.
Why it matters: a screen reader announces an image without alt text as a picture, or reads its file name, so whatever it shows is lost. One marked as decoration is skipped, which suits a border or a background.
How to fix it: in Word, choose View Alt Text on the image and describe it, or select Mark as decorative. A generator
sets the description on the drawing's wp:docPr (DocProperties.Description in the Open XML SDK).
Word: TableHeaders
Finds: a Word table of more than one row whose first row isn't marked as a header row (Repeat as header row,
w:tblHeader). A table of one row has nothing to label, so it's left out. Category: Tables. Severity:
Warning, since a table used only for layout needs no header row and the rule can't tell the two apart. From:
WCAG 2.2 1.3.1.
Why it matters: a screen reader reads a header row's cells out as it moves through the columns below, so the user knows what each value is. Without one, a row of numbers is just numbers.
How to fix it: in Word, select the first row and turn on Repeat as header row (Table Properties, Row). A generator
sets w:tblHeader in the row's properties (TableHeader in the Open XML SDK). A table used only to lay out a page is
better replaced by columns or text boxes; if it stays, DocxCheckSet.Latest.Except(DocxCheck.TableHeaders) leaves
the warning out, for every table in the document.
Word: DocumentLanguage
Finds: a Word document with text that has no language: the document's defaults (w:lang in w:docDefaults) set
none, and neither do the text's styles or runs. Text in the body, headers, footers, notes, tables and text boxes counts;
blank text doesn't. It's one finding for the whole document, with the path document, and its message says how many
paragraphs have such text. Category: Language. Severity: Error. From: WCAG 2.2 3.1.1.
Why it matters: a screen reader picks its voice and pronunciation from the text's language. Without one, it reads the text in the user's own language, so text in another one is mispronounced and hard to follow.
How to fix it: set the document's language in its defaults: in Word, Review, Language, Set Proofing Language, then
Set As Default. A generator sets w:lang in the default run properties of styles.xml (Languages in the Open XML
SDK's RunPropertiesDefault). Text in another language sets its own on its runs or style.
Word: HeadingStyles
Finds: a Word paragraph that looks like a heading but has no heading style: all its text is bold, it has 2 to 10
words, it doesn't end with a period, colon, semicolon or comma, and the next paragraph is body text that isn't bold and
has more words. Only the body counts, not tables, text boxes, headers, footers or notes. A list item, a caption (the
Caption style, or a SEQ field as Insert Caption adds) and a run of bold paragraphs, such as an address, are left out.
Empty paragraphs between the two are skipped. Category: Headings. Severity: Warning, since bold text can
also be emphasis and the rule can't always tell. From: WCAG 2.2 1.3.1.
Why it matters: a screen reader user moves through a document by its headings, and Word lists them in the navigation pane. Bold text looks like a heading but isn't in that list, so someone moving by headings can't jump to the section it starts.
How to fix it: in Word, apply a heading style (Heading 1, Heading 2 and so on) instead of making the text bold,
and change the style if the heading should look different. A generator sets w:pStyle to a heading style, or
w:outlineLvl on the paragraph or its style (ParagraphStyleId or OutlineLevel in the Open XML SDK). Bold text
that isn't a heading can stay; DocxCheckSet.Latest.Except(DocxCheck.HeadingStyles) leaves the warning out.