EPUB Accessibility 1.2 Explainer

There are two [wcag2] accessible authentication success criteria — a minimum (3.3.8) and an enhanced (3.3.9) version — as well as a related success criterion for re-authentication (2.2.5). These have a combined goal of ensuring that users with certain cognitive disabilities are provided an easy-to-use and accessible login method for protected web sites, and that information is not lost if users have to reauthenticate themselves.

Not all users are able to remember usernames and passwords, transcribe verification access codes, or solve puzzles to gain entry which is why it is important that logins do not rely solely on methods that require these kinds of cognitive capabilities. It is also vital to ensure that users do not lose any information they have input if they are required to reauthenticate as this could prevent some users from being able to complete the task (i.e., they may continue to reach the time out just trying to re-enter the lost information).

It is not common that these success criteria will apply to EPUB publications, however. When a login is required, the more typical scenario is that a vendor will require users to log in to their reading system to access their publications. In this case, if the login does not meet the requirements of the minimum success criterion, then it is the reading system that is not accessible, not the individual publications it provides access to. Similarly, if the reading system loses the current state of a quiz or task at the time it forces reauthentication, there is nothing the publication can do about this.

Even in the case of EPUB publications served through a learning management system (LMS), it is more typical for the user to log in to the LMS. The LMS may use that information to track their activity, but again it is rarely the publications within the LMS that are secured.

One major reason that EPUB publications typically do not build in login requirements is because scripting support is not guaranteed. The reading system is more reliably going to have a connection to, or be able to connect to, the user's network. It is also more capable of being able to store information for later transmission to a publisher's server when a connection is not immediately available, as persistent storage is also not commonly available within EPUB publications. The lack of persistent state also means that a login in one EPUB content document likely will not persist into any of the others. That means a user might, for example, have to log in at the end of every chapter to complete embedded quizzes.

Despite the potential drawbacks of adding authentication to EPUB publications, nothing prevents a publisher from using logins. They are much rarer than on the web because a publication belongs to one user and so it does not have to be blocked internally from public access and there is no question if the user is human or not. Authentication typically only matters in the case where the publisher has to collect information; for example, from quizzes and other interactive content.

For these reasons, it is unlikely that logins will be found outside of educational material, which can help narrow down when to look for them. Since a login will typically require the user to have an account with the publisher, they are less likely to be found in general reading material.

But, while it may be highly uncommon to find logins, their possibility cannot be ignored when performing an accessibility review.

Unfortunately, there is no quick way to determine if a publication has a login attached to any of its content short of manually checking. The only useful aid that an EPUB publication provides is the scripting designation that has to be expressed in the package document manifest.

Knowing which files are scripted may help reduce the amount of work involved in finding logins, but that is only if scripting is used sparingly in the publication. If most or all of the EPUB content documents are scripted, it might be easier to read the JavaScript files. Even without a strong knowledge of scripting, the names of the various functions might provide clues to their purpose. For example, a function with "login" in its name would be a major flag to look more deeply into what the scripting is doing.

Ultimately, direct testing of all scripted content for accessibility is always advised (i.e., not trying to gauge the accessibility by looking only at the markup), so as long as an assessment is thorough in this regard any logins should be found in due course.

If a login is found, checking for a reauthentication requirement may still take some work. It should never be assumed that lack of mention of a timed session means that users will not be prompted to reauthenticate after a certain amount of time, or a certain amount of time without activity being registered. Verifying if there is a timeout, and checking what happens after reauthenticating, may require inspecting code to understand when a timeout is triggered. If the evaluator is not the publisher of the content, it may be faster to check with publisher whether session timeouts can occur and how best to test them.

The [wcag2] consistent navigation success criterion (3.2.3) has a similar goal to the consistent help success criterion but in this case the objective is to make it easier for users on the web to find and use repeated navigational aids such as tables of contents, search boxes, and skip to content links. Having these aids in the same location, and the links within them ordered in the same way (when the links repeat), simplifies navigating a site.

But many of these kinds of navigation aids are not included in EPUB publications, at least in a way that they fall on the publisher to make accessible (i.e., being available across multiple content documents). It is the user's reading system that is going to make access to the table of contents, or landmarks, or page list, or a search box consistently available in its user interface. Likewise, information about what (virtual) page the user is on is handled by the reading system. Even the EPUB equivalent of a skip to content link — a landmark to the first page of body matter — is information for the reading system to use to advance the user; it is not content the user directly encounters in the publication.

If the reading system were to change the order of access to these features in the middle of reading a publication, that would be a failure of the reading system not the publisher. Similarly, if it were to change the order of the links within a navigation element — for example, change the order of the landmark links — that would also be a failure of the reading system.

The most common case where the consistent navigation success criterion applies to EPUB publication is when chapters include mini tables of contents at their start. These are most typically found in educational works to provide quick access to all the subsections of a major topic.

In this case, the position of each of these mini tables of contents needs to be consistent relative to any other content at the beginning of the chapter or section. If the first mini table of contents occurs after the first chapter heading, for example, then the same pattern has to be followed in each subsequent chapter.

It is also worth noting that the pattern of placement has to be consistent, not that the same elements always have to appear. For example, if some chapters have a title followed by an epigraph, followed by an image to illustrate the concept being covered before the mini table of contents is provided, then it is acceptable if some chapters omit the epigraph and/or image so long as the order of the elements before the navigation is always consistent. It would only be a failure if the mini table of contents is sometimes placed before the heading, or between the epigraph and image.

The order of links within the mini tables of contents is not likely going to be relevant to check for this success criterion. Because each chapter's content is unique, the links are never going to be repeated from one chapter to the next so there is not a consistent order that has to be maintained.

Although page numbers provide a form of static navigation aid — indicating to the user where they are in a publication — their consistent placement in reflowable EPUB publications is out of scope for this success criterion. Page break markers in these documents will occur arbitrarily at locations that typically correspond to a statically paginated equivalent, like a print copy of the book.

The consistent identification of page break markers in reflowable publications is more typically a concern of the reading system, as it will provide the user the page number somewhere in its user interface.

The [wcag2] reflow success criterion (1.4.10) seeks to ensure that content remains readable for users with low vision as they enlarge it. It requires that there be no loss of information or readability for vertically scrolling content at a width equivalent to 320 CSS pixels [wcag2].

The first thing to note about this success criterion is that the "scrolling" aspect of the definition applies to all EPUB publications whether they are read in a reading system that lays out the content in a scrolling manner or whether the reading system dynamically paginates the content. Dynamic pagination is just a feature of EPUB that takes a vertically scrolled document and splits it up for reading in a book-like manner.

The second thing to note is that the width and height sizes do not refer to the actual screen resolution of the device. The emphasis of the success criterion is on equivalence up to these sizes, which is why it notes that a viewport with a width of 1280 CSS pixels is equivalent to a width of 320 CSS pixels when it is zoomed to 400%. The example is just a common screen resolution and exhibits easy math to the 320 CSS pixel maximum.

What happens when zooming a 1280 CSS pixel viewport beyond 320 CSS pixels is out of scope, as is zooming any other starting resolution after it exceeds 320 CSS pixels in width. The success criterion is not saying that all viewports greater in width than 320 CSS pixels are out of scope.

The good thing about reflowable EPUB publications when it comes to meeting this success criterion is that they were designed to work exactly in the manner it anticipates. A reflowable EPUB publication is expected to work on many different device screens and with many different user preferences applied. The content has to reflow to fit within the paginated environment that reading systems provide as users rarely can scroll the content — if the content does not fit in the viewport, it is typically inaccessible to all users. Publishers are typically already aware of these reflow issues and how to ensure their content does not break.

It does not mean that all reflowable EPUB publications get an automatic pass on this success criterion, but only that the circumstances where the content will not reflow are not common.

The most likely source of failure for the reflow success criterion is going to arise with container elements, such as with the [html] aside, div, nav and similar elements used to group content. When these elements are assigned fixed widths, their ability to adapt to changes in zoom diminish.

The content of an [html] aside element used as a sidebar, for example, will reflow by default to so that its content fits the available screen space. But if a width, or a minimum width, greater than 320 CSS pixels is assigned, then as the content is zoomed the dimensions of the viewport may become smaller than the width of the aside.

When a user zooms their publication, what typically will happen with EPUB reading systems is that the sidebar will extend outside the boundary of the page/viewport, causing its text to be cut off and become inaccessible.

But even if the sidebar is made scrollable using CSS overflow properties [css-overflow], the result still fails this success criterion because the content now requires scrolling on the x axis to reach the hidden overflow text.

Any hoped-for value in assigning fixed widths to container elements in reflowable content is generally going to be offset by the problems it will cause with the reflow success criterion. If a size must be specified (e.g., to ensure that boxes of content always fill the width of the viewport), the more accessible way is to use a unit that adapts to changes in the viewport. Setting a percentage (%) or viewport width (vw) no greater than 100 percent of the viewport size will avoid scroll bars appearing as the content is zoomed.

When testing this success criterion, be aware that app-based reading systems typically do not allow zooming of reflowable content. They are more likely to support font size increases up to 400% of the default, but increasing the font size is not the same as zooming the content.

The reflow success criterion does not, in fact, specify requirements on the font size as the content is enlarged. The requirement for text to enlarge up to 200% is instead addressed by the [wcag2] resize text success criterion.

Large words, hyperlinks, code examples, and other strings of contiguous text with no break points can also pose problems for reflow as the viewport reaches 320 CSS pixels, but the text is not required to stay the same font size as the content is enlarged. The font size could, for example, be reduced using a media query so that it stays at 200% the original size as the viewport nears 320 CSS pixels.

Other strategies to prevent large text strings causing scrolling include using the [html] wbr element to specify word break locations, using the CSS overflow-wrap and word-break properties, or using the Unicode soft hyphen character.

Not all content is required to reflow as it is enlarged. The success criterion makes an exception for content that would lose its meaning if it were reflowed. For reflowable publications, the most common content that requires two-dimensional layout are diagrams, charts, maps and similar images.

Data tables are also exempt due to their grid nature, but the individual cells within them are not. This means that it would not be unreasonable for the user to have to scroll to reach all the columns, but the content with a column must not require scrolling in two dimensions to read.

Content that can be scrolled in two dimensions needs to always be checked that it has been authored so that it can indeed be scrolled this way. Reading systems will often try to compress tables to fit the available space, for example, but when that cannot be done the overflow outside the page boundary will not be accessible. A common practice to avoid this is to wrap the two-dimensional content in a container that allows scrolling (e.g., a table can be placed inside a div that provides it more space for rendering and includes scrollbars to move around the columns and rows). But when using this practice, content that does not depend on two dimensions for comprehension cannot be captured in the scroll. If a table has a caption, for example, only the table can scroll on the x axis to meet the success criterion; the text of the caption cannot require scrolling.

Although the reflow success criterion does not pose a lot of problems for reflowable EPUB publications, it is one of the most problematic to meet for fixed-layout EPUB publications as reading systems do not reflow the content of a fixed-layout page as it is zoomed.

This guide does not address the problems with making fixed-layout publication accessible. For more information on this topic, refer to EPUB Accessibility - Fixed Layout Challenges and Best Practices [epub-fxl-a11y].

A page list [epub-a11y] is a list of links to static page break locations in an EPUB publication. These destinations remain the same across reading systems, device screen sizes, and user text preferences.

A page list is essential for accessibility when digital publications are used in coordination with a static alternative like a print edition. The most common scenario is when digital publications are produced for use in educational settings. Students who cannot read a print edition need to be able to coordinate their digital edition with their sighted peers. But any social reading environment, such as a book club, will also be more inclusive when users are not disadvantaged by their need to use a digital edition.

It is true that many reading systems provide a paginated reading experience and allow users to move between these pages. The key difference is that this pagination is typically not static and does not match up with a statically paginated source. The number of pages and what content appears on any given page changes depending on how much content can fit in the device's screen. That in turn is influenced by whether the user has altered the fonts and spacing to their own preferences. A publication rendered using larger font sizes and greater spacing will have more dynamic pages than one using the reading system's standard settings. Consequently, coordinating access across devices is impossible without a static baseline.

When a page list is provided, a reading system can provide an interface that allows users to enter the page number from the static source that they wish to move to. This provides users the ability to find the same location in their digital edition as their peers using a print edition, for example. Without a page list, page coordination with a print source becomes extremely difficult as it would rely on navigating the individual page break markers (if they are even present).

The page list is not required to reproduce every page break from the static source. Digital editions often omit content from their static print counterparts, for example, typically in the front or back matter. Blank pages are another feature of print editions that have no equivalent in digital editions. Page break markers can be included for them but the effort to add them may not be worth the effort knowing that users never actively seek this locations.

What is required is that the page list include links to every page of content that is reproduced from the static equivalent. Moreover, if there are page break markers in the EPUB publication, the page list has to include links to all these markers. This is to ensure that the EPUB publication does not contain content that cannot be reached.

The page list requirement is currently only mandated in a limited set of scenarios because the benefit to users only accrues when there is a statically paginated equivalent to sync up with. As already noted, when a publisher knows their content will be used in a mixed reading environment, such as when they sell publications to schools, ensuring the digital edition has a page list is essential. Likewise, if they are providing students with a print and digital bundle, the need to coordinate reading can be anticipated. And finally, if a digital publication is produced from the same workflow as a print publication, making it easy to retain the page locations in the digital version, then publishers should opt for maximum accessibility by producing the page list.

A page list is not required for EPUB publications that do not have a statically paginated source. Digital-only publications can benefit from a page list for coordinated access within reading groups, but the lack of a page list affects all users equally. It does not specially disadvantage any one type of user over another.

The ability to identify the current page a user is on, and have that page be consistent across devices and user settings, is another essential feature for users who are trying to coordinate their reading in a mixed print-digital environment. The page list provides quick access to the page break locations, but without static page break markers [epub-a11y] in the content the context of where a user is relative to the statically paginated equivalent is lost.

Assistive technologies [epub-a11y] can use the page break markers to announce the current page number the user is on. This makes it easier for users to verify they are in the right location, for example, or to cite content on the page. Page break markers also allow assistive technologies to provide shortcut keys that allow the user to move forward or backward by page — a feat that is not easy to accomplish with only a page list.

The inclusion of page break markers also simplifies the creation of a page list, as they provide easily referenced destinations for the links.

It is not required that there be page break markers for every page from the static equivalent. Some content from print sources is not reproduced in digital editions, for example. There are also print conventions, like inserting blank pages so new chapters always begin on a recto page, that do not translate to digital editions. Page break markers can be included for these blank or missing pages, but additional context needs to be provided so users are aware why there is no content.

Since some content can also be rearranged from print to digital editions, it is also not a requirement that the page break markers be arranged in numerical order.

As with the page list, it is not required that every publication include page break markers to be accessible. The same conditions that apply to including a page list apply to include page break markers.

An EPUB publication containing a page list and page break markers cannot be considered fully accessible if users are unable to identify the corresponding source edition. Pagination and content layout frequently vary across different editions of the same publication; for example, the pagination of a hardcover edition often differs from that of the softcover edition.

Consequently, the source of the pagination [epub-a11y] has to be specified in the metadata to allow users to determine if they have the right digital edition.

The source publication is typically identified using a standardized identifier such as an ISBN or ISSN. These identifiers are more accurate than text descriptions of the source. In the case of a publication that does not have a standard identifier, the source can be identified by providing as much information about it as possible (e.g., title, publisher, edition, publication date, etc.).

Providing the source identifier allows vendors, for example, to present the information to users in their bookstore metadata. This allows users to be sure of their purchase before making it.

A page source identifier is only set when an EPUB publication contains one or both of a page list or page break markers. Setting it when neither is not only incorrect but can mislead users into thinking the features are present even though, for example, they will not be listed in bookstore metadata.

The EPUB 3 standard defines how to use media overlays [epub-3] to synchronize text and audio. What it does not do is require any amount of coverage when using the technology. Consequently, an EPUB publication might contain text and audio synchronization for only part of a work and rely on the user having access to text-to-speech playback for the rest.

Text-to-speech playback unfortunately offers poorer playback quality than pre-recorded narration. Text-to-speech engines have limited built-in vocabularies which causes them to mispronounce any uncommon words they encounter. Consequently, some users require that words be repeated or spelled out to understand what is being voiced. That in turn slows down reading and reduces comprehension.

This is not to diminish the value of being able to use text-to-speech playback. While many users prefer hearing professional narration, for some works it forces users to read at a slower speed than they prefer due to a lack of playback speed control. Being able to stop playback and inspect words manually is also a valuable tool. A user may generally be interested in learning how to spell a word they have never heard before.

This is why it is necessary to provide full audio synchronization and the full text content of a publication to meet this objective. Users can then decide which reading modality they prefer: text, audio, or a mix of the two.

Media overlays [epub-3] allow par elements to reference text content without corresponding pre-recorded audio. This serves as a request for the reading system to synthesize the referenced text, but the feature is not widely supported across reading systems. Consequently, it does not satisfy the criteria for a complete synchronized media presentation.

Media overlays also do not require the full text content of a publication by default. Some producers provide only the major headings of a work as the text content and synchronize the full audio of the work to them. This technique is called structured audio. While this is perfectly acceptable for providing as close to a pure audiobook experience as possible in EPUB, it does not meet the general accessibility requirement of this objective.

Every EPUB publication has a default reading order that allows users to progress through the content. The default reading order consists of two parts: the order of references in the spine provides a high-level progression through the EPUB content documents that make up the publication, while the markup within each EPUB content document provides the default progression through the content elements (i.e., as represented in the document object model [dom]).

For many languages, the default reading order also matches the logical reading order — the way that users will naturally follow the narrative. It ensures that readers can follow the primary narrative and that they encounter secondary content where the author intended it to be read. The default reading order also establishes some less obvious relations, like the progress within a table from cell to cell and row to row.

If the sequence of the synchronized text-audio playback does not match this progression, it can cause confusion for readers, whether they are only listening to the audio or trying to also follow visually.

Ordering the playback to match the default reading order is the safest way to ensure that users can follow the text. In some cases, however, strict adherence to this practice can result in a suboptimal reading experience. For exmaple, playback of a table by column instead of row might make more natural sense in some cases. These publications have a logical reading order that cannot match the default order of the elements of the host format.

The goal of this objective is not to forbid alternate presentations, but to ensure that deviations are only made to better represent the logical reading order of the content.

Being able to read the primary narrative of a work without interruption is central to reading comprehension. EPUB publications are typically structured to visually represent secondary information such as page break markers, sidebars, and footnotes outside the main narrative flow. Different background colors or content placement are commonly used, for example, as cues to visual readers that this information is outside the primary narrative.

Readers who prefer auditory playback, however, cannot skip this information with the same ease in a linear audio-based reading experience. While media overlays do not represent a true audiobook, they suffer from a default lack of structure and semantics. Media overlays only require a linear sequence of par elements [epub-3], where each par element defines the text and audio to synchronize.

This default lack of structure and semantics means that users can only move forward and backward one par element at a time. In more practical terms, that means they might only be able to skip forward a single word or sentence at a time until they believe they have reached the end of the content they wish to skip.

When the synchronized text-audio playback instructions are structured and include structural semantics, however, reading systems can create reading experiences that allow users to decide which secondary content to skip. That allows users to filter out the content they do not want to hear by default.

When reading visually, users can easily resume reading the primary narrative once highly structured content such as sidebars, lists, and figures stop being of interest. Visual readers can skim lists and quickly return to the primary narrative once they locate the desired information, for example. The same is true for reading figures and sidebars, as they are visually offset from the primary narrative so their end is easily found.

A media overlays file without any structure or semantics prevents users from navigating out of this kind of content with the same ease. Like the skippability problem, the user has no option but to keep moving playback forward one item at a time. They will also typically have to guess when the primary narrative resumes, which can be especially problematic when trying to get out of content like sidebars that also consist of paragraphs of text.

Users can exit highly structured content during media overlays playback if the synchronized text-audio playback instructions include document structure and semantics. This structural data enables a reading systems to provide navigation mechanisms, such as keyboard shortcuts, to escape the content. For example, a user can exit individual table cells and rows, or bypass the entire table.

In addition to body text synchronization, media overlays [epub-3] can be applied to the EPUB navigation document. Including synchronized text-audio playback within the EPUB navigation document enables a reading system to generate auditory labels for hyperlinks, enhancing navigation clarity for audio users.

Reading systems typically provide their own interfaces to the navigation aids in the EPUB navigation document. For example, they open the table of contents as a specialized interface on top of the content the user is reading. To access these interfaces, users typically rely on text-to-speech playback, when available, to hear the entries.

Providing synchronized text-audio playback for the EPUB navigation document provides reading systems the ability to use the synchronized audio as auditory labels for the links, improving the experience for auditory readers.

To meet this objective, it is also required that audio synchronization of the navigation document be complete. All the navigation components, not just the table of contents, need to have audio synchronization.

There are fifty-six success criteria that fall at level A or AA in WCAG 2.2 [wcag2]. They are organized by whether they improve the perceivability, operability, understanding, or robustness of the content.

This organization matters because success criteria across multiple categories can apply to different content types. The non-text content success criterion, for example, can apply to images, audio, video, and interactive content. It requires that static images and dynamic images drawn on an [html] canvas element have alternative text and/or descriptions. And for audio and video, it requires labels to describe the nature of the content.

At the same time, there are various success criteria that only apply to a single type of content. The time-based media success criteria, for example, only apply to audio and video.

When carrying out an accessibility evaluation, it is important to know the possible applications of each success criterion. This avoids both mistakenly missing critical checks and wasting time on success criteria that cannot apply the content.

It is not possible for a document like this to determine exactly which success criteria can apply to an EPUB publication because so much depends on the nature of the content. But it is possible to help steer evaluators away from success criteria that do not apply.

Many publications do not contain images, audio, video, or scripted content. The filtering table identifies success criteria that can be skipped when they are not present in EPUB publications.

It is possible to determine what types of content are present, or not present, in an EPUB publication by using validation tools and inspecting the package document. With this information, the filtering table can be used to skip any success criteria that do not apply.

Before attempting to filter the success criteria it is critical to be sure what content an EPUB publication contains. The traditional way to verify this is through the package document manifest [epub-3]. Each manifest item has to list its media type in a media-type attribute, so all that is required is a knowledge of which media types define which type of content.

The media types themselves typically aid in understanding the nature of the listed resources, as the start of each usually classifies the type. Image formats, for example, begin with "image/", while audio formats begin with "audio/". EPUB also maintains a list of core media types which also limits the number of formats that are actually used in publications.

Unfortunately, the package document manifest is an imperfect way of assessing the content of an EPUB 3 publication. It only works well for EPUB 2 publications because EPUB 2 lacks the more advanced content options of EPUB 3.

Determining whether there are images in an EPUB publication is a prime example of how the manifest can fail. Scripts can be used to draw images on the [html] canvas element or to create interactive charts or games. These kinds of images will not be listed in the manifest because they are created using native html functionality.

Similarly, data URLs allow raw image data to be embedded directly in html, once again bypassing a manifest entry. A search for URLs that begin with "data:" might not even be sufficient as a script could dynamically insert an image this way.

Scripting is also problematic because it limits the ability of validators to check whether resources are missing from the manifest. If an [html] audio or video element is dynamically written into an html file, for example, a validator might only report the corresponding audio or video file in the EPUB package as an unused resource in a usage message. Usage messages are not usually emitted by default making the issue easy to miss. Moreover, if the script writes a reference to a remotely hosted resource, the validator will not report anything at all.

Evaluators will always have to be aware of these limitations and be on the lookout for content that is not listed in the manifest.

In an odd twist, although scripting makes the package document manifest less reliable for detecting non-text content, the package document is the best way to detect if scripting is used. When scripting is used in an EPUB content document, no matter what form it takes, it has to be declared in a properties attribute [epub-3].

Scripts are not always stored in standalone JavaScript files, so it can be quite difficult to determine if scripting is present without the help of a validator. Scripts embedded using [html] script tags and scripting inlined in the markup using event handler attributes like onclick and onkeypress are a couple of ways that scripting can be present at the content level. Trying to locate these possibilities manually is prone to mistakes. But so long as the EPUB publication is validated prior to carrying out an accessibility evaluation, detecting scripting will be straightfoward.

The following table provides a list of success criteria that can be skipped if an EPUB publication does not contain the specified type of content.

If an EPUB publication does not contain any of the above content types, and does not use CSS animations, the following Level A and AA success criteria will also not apply:

• non-text content (1.1.1)
• non-text contrast (1.4.11)
• pause, stop, hide (2.2.2)
• three flashes or below threshold (2.3.1)