Abstract
A publisher may call a page readable after a visual review, yet that conclusion can hide several different propositions. It may mean that text did not clip when spacing changed; that a participant group completed a specified reading task under study conditions; that readers reported a preference; that content reached a nonvisual route; or merely that a designer found the hierarchy coherent. These propositions are related, but they do not prove one another.
This paper introduces reading variability: the fact that a presentation’s usability and interpretive support can change with a reader’s task, content properties, typographic variables, available adjustment, alternative route, environment, and evidence. It distinguishes documented requirements, observed results, preferences, and untested hypotheses. The paper then provides a Reading Variability Profile for recording a named reading situation, a relevant variable, adjustment or route, evidence source, uncertainty, and non-conclusion. The profile does not individualize a diagnosis or select a universal setting. It gives a publishing team a way to replace “this page is readable” with an answerable, proportionate claim.
One page, three reading situations
A small press is preparing a public-information page about a new submission policy. The page is not elaborate. It has a title, an introductory paragraph, three short sections, a compact timetable, a warning in a shaded callout, and a link to the full policy. The design team agrees that it is clean. The typeface is familiar. The hierarchy is consistent. The page has been read on the office screen and sent around for approval.
At the end of the review, someone asks a question that sounds ordinary enough to close the meeting:
Is the page readable?
The question cannot be answered yet, because no one has said what event readable is meant to describe.
Consider three plausible encounters with the same page. In the first, a reader needs to make the text larger and increase spacing before following the timetable. The reader is not asking the page to become a different policy. The reader is asking whether the page can remain intact when a legitimate presentation adjustment changes line lengths, vertical rhythm, and the amount of text visible at once. The designer’s original spacing may be coherent, but coherence is not the relevant fact if the adjustment hides a deadline, overlaps a heading, or causes the warning to detach from the rule it qualifies.
In the second encounter, a reader reaches the page through a nonvisual or assistive route. The visible shaded callout is not encountered as a shaded callout. Its importance must be carried by structure, label, order, and text. The timetable is not first understood as a visual grid. It must expose headers and relationships in a route that can be followed without seeing its columns. The link to the full policy must still announce where it goes. In this encounter, the question is not whether the page looks well typeset. It is whether the typographic and structural system has left the policy’s hierarchy and exceptions intelligible by another route.
In the third encounter, a reader sees the page quickly but is trying to separate the main submission rule from the callout, the timetable, the optional detail, and the link to a longer document. For this reader, a typeface name alone explains almost nothing. The consequential features may include heading distinction, line length, contrast between blocks, proximity, the density of the table, the amount of surrounding interface, the reader’s time, and the task of locating one answer without losing the whole policy. A visually attractive page can still make a hierarchy difficult to parse. A visually sparse page can still fail to distinguish what is mandatory from what is explanatory.
These three encounters do not prove that three kinds of readers exist in neat categories. They do not identify a diagnosis, predict a preference, or permit an editor to decide what any unnamed person can read. They hold one public page still long enough to reveal three kinds of reading situation: a task, a route, an available adjustment, a content structure, and a set of conditions in which a typographic variable becomes consequential.
The distinction matters because publishing guidance often turns accessibility into an inventory of design slogans. Use this font. Avoid that font. Make the type larger. Add more space. Never use a dense table. Provide a clean layout. Such guidance may contain useful observations, but an observation becomes unreliable when it is promoted from a condition into a universal verdict. A typeface can be a variable in reading without being the whole situation. A reader’s stated preference can be informative without being proof of measured performance. A page can survive an adjustment without proving that every route is clear. A publisher can expose accessible qualities without predicting suitability for every individual task.
The central question is therefore not “What font is accessible?” It is:
Readable to whom, for which task, through which route, and with what evidence?
The four things a typography claim might be
The word readable often works as a compressed conclusion. It may be used to praise a visual composition, to report an experiment, to summarize a technical requirement, or to repeat a reader preference. The compression is tempting because all four uses point toward a concern for readers. It is risky because they answer different questions.
| Claim category | The question it can answer | What it cannot answer alone |
|---|---|---|
| Documented requirement | What must not be lost when content is adjusted or presented through a stated standards context? | Whether every reader prefers the result or can complete every task. |
| Observed result | What occurred for a defined sample, outcome, material, and study condition? | What will occur for all readers, languages, systems, or publication settings. |
| Preference | What presentation did a person or group say they favored under a named choice? | Whether that choice improved speed, comprehension, navigation, or another outcome. |
| Untested hypothesis | What might matter and needs direct inspection or reader research? | A general claim, recommendation, certificate, or prediction. |
The categories do not make design timid. They make a claim answerable.
Take spacing. The W3C’s text-spacing success criterion does not tell an author to begin with one ideal line, paragraph, letter, or word spacing. Its concern is different: in markup-based content where those properties can be overridden, content and functionality should not be lost when a user applies the specified adjustments.[1] The page in the first encounter may therefore need to tolerate a change in spacing without clipping a deadline or burying a link. That is a documented requirement about loss. It is not evidence that the stated spacing values are the most comfortable, effective, or preferred arrangement for every reader.
The distinction has two parts that should remain separate. First, W3C’s explanation assigns the author a responsibility to avoid loss under the stated condition. Second, it recognizes that the user may make choices about adjustment and may choose values beyond those in the criterion. The criterion does not require the content to supply its own adjustment control, nor does it decide which arrangement will let an individual reader derive meaning most easily.[1] In other words, an author can be responsible for a resilient text system without pretending to know the reader’s best setting.
Text resizing provides a related example. W3C guidance says that visually rendered text can be resized up to 200 percent without loss of content or functionality, subject to the criterion’s stated exceptions. It locates much scaling capacity in user agents, while asking authors not to prevent effective scaling. The guidance also describes how fragile layouts can produce clipping, overlap, and narrow vertical columns as text expands.[2] A page that survives resizing has met an important condition. It has not thereby demonstrated that its hierarchy is clear, that its table can be understood in a nonvisual route, that its language is suited to its task, or that every reader will find the resulting density workable.
The same care is needed when a page is called accessible because it exposes a feature. The current EPUB Accessibility 1.2 Candidate Recommendation Draft requires EPUB accessibility metadata to disclose properties such as sufficient access modes, accessibility features, and hazards. It recognizes that a publication may have more than one sufficient access-mode set when alternatives are present. It also makes a crucial distinction: content that does not meet the specification’s requirements does not necessarily fail to meet an individual user’s needs.[3] The reciprocal insight is equally useful. A declaration that a publication has a feature or route does not predict that every reader will find the title suitable for every task. A statement can be informative without becoming a prophecy.
The W3C Publishing Community Group’s accessibility-metadata display guide gives publishers, libraries, retailers, and distributors a practical vocabulary for this distinction. It recommends presenting publisher-supplied information as claims or declarations, and organizing reader-facing information around visual adjustments, nonvisual reading, and conformance information. The guide says that reflow alone is not sufficient to establish that font, spacing, and color adjustments will be possible without other readability problems. It also treats “No information is available” as a legitimate neutral display when a relevant declaration is absent.[7] This is not a W3C standard or a suitability test. It is useful because it permits uncertainty to appear in the record instead of being covered by a confident adjective.
A publishing team should therefore pause whenever a discussion uses a broad conclusion. The question is not whether the adjective is forbidden. The question is: What category of claim is this, and what evidence could it reasonably carry?
Variables do not act alone
A typeface name is a poor container for a large promise. It does not identify the task, the content, the route, the display environment, the available adjustment, the relevant outcome, or the evidence. It does not tell a reader whether an image has a description, whether headings expose hierarchy, whether a chart is structured, or whether a callout that looks prominent can be recognized outside a visual presentation. A font may matter. It does not work alone.
A reading situation can be represented as a relation rather than a demographic label:
| Situation component | Question for the publishing team | Example from the constructed policy page |
|---|---|---|
| Task | What is the reader trying to do? | Locate the submission deadline; distinguish a rule from an exception; decide whether to open the full policy. |
| Content property | What makes this object difficult or consequential? | Dense timetable, shaded warning, abbreviated labels, link destination, multiple levels of heading. |
| Typographic variable | What presentation feature is under discussion? | Hierarchy, line length, spacing response, emphasis, table density, or the relationship between text and callout. |
| Available adjustment | What may the reader or reading system change? | Text size, spacing, display area, color treatment, navigation route, or another supported setting. |
| Alternative route | How is the needed information exposed beyond the initial visual composition? | Structured headings, table headers, link label, text description, reading order, or read-aloud/dynamic-braille route. |
| Evidence status | What is known, observed, required, preferred, or unknown? | A W3C requirement about loss, an observed test result, a reader report, a direct inspection, or an untested hypothesis. |
This account does not make every small publishing decision impossible. It identifies where a claim is doing more work than its evidence can carry.
Scaltritti and colleagues provide one useful empirical reminder that typography is multi-variable work. Their study used 50 existing webpages rather than a narrowly constructed set of artificial stimuli and recorded eye-movement measures while 79 Italian native speakers read the pages. The final participant sample included adults and middle-school children, with typical readers and readers with dyslexia represented in the study design. The authors examined multiple visual and typographic variables together and report that several variables affected reading-process measures; effects were mostly homogeneous across the groups, with some exceptions.[4] Their ecological design is valuable because webpages combine variables: font size, spatial distribution, spacing, contrast, content, and surrounding choices often arrive together.
The same design is also a limit. Because the study examines co-occurring properties in existing webpages, it does not establish a universal causal rule for one isolated typeface or one spacing value. It studies Italian readers, its own webpage sample, its own eye-movement and performance outcomes, and an explicitly bounded set of conditions. The study can teach a publication team to distrust a one-variable slogan. It cannot tell that team what every reader in every language, format, device, or task will need.
This is the core of reading variability. The phrase does not mean that every visual choice is equally useful or that design should dissolve into arbitrary options. It means that the unit of judgment is a situation: a reader task using a particular route and content under stated conditions, supported by an appropriate kind of evidence. A typeface is one possible component of that situation. It is not the situation itself.
What evidence says—and declines to say
The evidence most likely to be misused is often evidence that deserves to be taken seriously. A well-designed study may show a meaningful difference under its stated conditions. The error occurs when the study’s conditions are silently removed and the result becomes a sales claim, a typeface label, or a prediction about unnamed people.
Two contrasting font studies make the problem concrete.
Kuster, van Weerdenburg, Gompel, and Bosman tested a named typeface, Dyslexie, in two experiments. In the first, 170 children with dyslexia read text in Dyslexie and Arial. In the second, 102 children with dyslexia and 45 children without dyslexia read word lists in Dyslexie, Arial, and Times New Roman. The authors report that the Dyslexie font did not produce faster or more accurate reading in their experiments and that preference was not related to reading performance.[5] This is a valuable counterweight to a claim that a specially branded typeface proves itself through its name, or that a favored presentation automatically produces a measured outcome.
It is not a proof that typography does not matter. It is not a reason to dismiss readers’ stated preferences. It is not a claim about every font, language, age group, device, reading task, or text. A null result for a named comparison in specified experiments does not become a universal null result for typographic design. Its strongest lesson is methodological: special-purpose branding, reader preference, and measured performance are separate questions.
Wallace and colleagues provide a different bounded result. An Adobe Research record for their 2022 ACM Transactions on Computer-Human Interaction article describes an adult “interlude reading” context with perceptually normalized font size and hundreds of crowdsourced participants. The record reports a 35 percent difference in reading speed between the fastest and slowest fonts in the study without a reported change in comprehension, and it interprets high speed variation across fonts as evidence that one font does not fit all for that measured outcome and setting.[6] This finding belongs in a serious discussion of typography because it identifies variation at the participant level rather than simply comparing group means.
Its scope must remain visible. The result concerns adult interlude reading, the study’s font set, its speed and comprehension measures, and its online research arrangement. It does not establish that a publisher should personalize every document, that any particular reader needs a bespoke font, that a typeface improves all forms of reading, or that measured speed is the only meaningful outcome. A researcher may value speed in one task and a reader may value detailed comprehension, navigational confidence, annotation, fatigue, reference stability, or another outcome in another task. The responsible response is not to discard the study. It is to record the outcome, setting, and unclaimed remainder.
The following evidence check keeps the distinctions visible.
| If the source says… | A proportionate publishing statement is… | Do not turn it into… |
|---|---|---|
| Content does not lose information or function under a stated spacing override. | “We inspected whether the content survives this specified adjustment without loss.” | “This is the optimal spacing for all readers.” |
| A study found an effect among a named sample using named material and measures. | “This result occurred under those study conditions.” | “This variable is universally better.” |
| Participants preferred one presentation. | “Preference was reported under this choice and setting.” | “Preference proves performance, comprehension, or conformance.” |
| A publication declares a visual-adjustment or nonvisual route feature. | “The publisher declares this feature or route.” | “The title will meet every individual need.” |
| No relevant evidence or metadata is available. | “This condition remains unknown and needs direct inspection or a more limited claim.” | A silence disguised as “readable.” |
The chart does not convert uncertainty into failure. It makes uncertainty legible enough to guide the next decision. A team may need a direct review of one dense table. It may need to test a task with readers relevant to a new publication. It may simply need to stop claiming that an untested feature is broadly readable. Each outcome is preferable to presenting a design judgment as evidence it does not possess.
Coherent defaults still matter
The rejection of a universal readability verdict can be misunderstood as a demand for endless variability. It is not. A publication needs an initial hierarchy, an ordinary reading route, a system of emphasis, a default sequence, and a coherent visual organization. A reader who first meets a page should not encounter arbitrary variation because a publisher has mistaken uncertainty for an excuse not to design.
A coherent default is a declared starting condition whose purpose and limits are inspectable. It is not an implied prediction that every reader will find the initial setting best. A coherent default may make an editorial hierarchy visible, reduce unnecessary visual noise, provide a predictable order, and give a team a stable artifact to inspect. It becomes less defensible when it prevents an adjustment that a reading system can otherwise provide, hides content when text changes, fails to expose structure through an alternative route, or is described as a universal solution without evidence.
W3C text-spacing and resize-text guidance supply a practical counterpressure. A team can create a deliberate initial form while also ensuring that legitimate adjustment does not create loss.[1] [2] The W3C metadata display guide adds that visual adjustment and nonvisual reading should be exposed as publisher claims so readers can see relevant information before and during use.[7] Together, these sources do not tell a team which default to select. They show why the default must coexist with adjustment, route, disclosure, and honest limits.
The goal is not “every setting for every person.” The goal is a page whose initial design is coherent, whose meaningful content does not disappear under stated adjustments, whose alternative route can expose the information required for comprehension, and whose claims remain no broader than the evidence.
The Reading Variability Profile
The Reading Variability Profile is a decision record for a bounded reader-facing object: an article opening, library notice, policy page, chapter lead, table, illustrated explanation, reference note, or other identifiable publication component. It is not a diagnostic of a reader. It is not a font-selector, a conformance certificate, or a substitute for testing. Its job is to prevent four different claims from being silently treated as one.
| Profile field | Record | Why the field matters |
|---|---|---|
| Reading situation | A task-and-route statement, not a demographic label. | It identifies the practical encounter under review. |
| Reader task | Locate, compare, study, skim, annotate, follow an instruction, or another named activity. | A presentation can support one task while remaining uncertain for another. |
| Content property | Dense table, long paragraph, heading hierarchy, embedded image, callout, citation block, multilingual passage, or another relevant object. | It keeps a typographic variable attached to what the reader is actually meeting. |
| Typographic variable | Hierarchy, spacing response, visual grouping, line measure, text scale, emphasis, table density, or another named feature. | It prevents a font name from standing in for the whole presentation. |
| Available adjustment | Resizing, spacing override, display change, navigation route, magnification, or another verified possibility. | It separates the initial visual form from reader or system control. |
| Assistive or alternative route | Structured sequence, heading navigation, text alternative, read-aloud/dynamic-braille availability, table header relation, or another relevant route. | It identifies what must be exposed beyond an initial visual arrangement. |
| Claim category | Documented requirement, observed result, preference, or untested hypothesis. | It stops a preference or design intuition from borrowing the authority of a standard or study. |
| Evidence source | Inspected standard, study, published declaration, direct review, reader research, or explicit lack of evidence. | It makes the warrant inspectable. |
| Observed uncertainty | What remains untested, unknown, context-dependent, or outside the source’s scope. | It gives uncertainty a place in the record rather than hiding it. |
| Non-conclusion | What the profile does not decide. | It prevents the record from becoming a certificate, font ranking, reader diagnosis, or universal rule. |
Worked profile: the submission-policy page
The following example is constructed. It does not identify a product, a reader, a diagnosed condition, a typeface, a size, a test result, or an actual accessibility outcome.
| Profile field | Constructed entry |
|---|---|
| Reading situation | A reader needs to identify the submission deadline and understand a policy exception from a public-information page. |
| Reader task | Distinguish a mandatory rule from a shaded exception and confirm the timetable’s relevant date. |
| Content property | Multi-level heading hierarchy, compact timetable, shaded callout, one link to fuller policy text. |
| Typographic variable | The visible distinction among headings, ordinary paragraphs, the callout, and table labels; the page’s response to increased text spacing. |
| Available adjustment | A markup-based presentation is inspected for loss under the W3C text-spacing condition; no claim is made about every reading system’s controls. |
| Assistive or alternative route | Headings expose their structure; the callout’s status is stated in text; table headers and link destination are identified in a sequential route. |
| Claim category | Documented requirement for no loss under stated spacing adjustments; untested hypothesis regarding the clarity of the callout for a named task. |
| Evidence source | W3C text-spacing guidance for the non-loss condition; direct structural inspection for the constructed route. |
| Observed uncertainty | No direct reader study has been conducted on whether the hierarchy supports rapid policy comparison under varied conditions. |
| Non-conclusion | The profile does not declare the page universally readable, certify conformance, rank its typeface, or predict any individual’s outcome. |
The worked profile demonstrates a useful change in editorial language. Instead of saying “The page is readable,” a team can say: “The page’s structured text and layout were inspected for loss under a stated spacing override; its headings, table headers, callout label, and link destination are intended to expose the policy’s hierarchy through a sequential route; the effectiveness of the resulting hierarchy for rapid comparison remains untested.” This statement is less dramatic. It is more useful because another editor can inspect it, revise it, or decide what evidence to seek next.
A claim that remains answerable
A publication does not become more humane by abandoning every default or pretending that no evidence can guide design. It becomes more accountable when it stops making a feature, a preference, a standard, and a study result do the work of one universal promise.
The Reading Variability Profile offers a small discipline: name the reading situation; state the task; identify the content property and relevant variable; record what can adjust and what alternative route is available; classify the evidence; and retain the uncertainty that the record cannot resolve. The result is not a perfect prediction. It is a claim that remains answerable.
The final test is therefore not whether a typeface has been called accessible. It is this:
Readable to whom, for which task, through which route, and with what evidence?
When a publishing team can answer those four questions—and can say where the answer stops—it has moved from a visual verdict to a responsible reader-facing practice.
References
[1] World Wide Web Consortium Web Accessibility Initiative. Understanding Success Criterion 1.4.12: Text Spacing. https://www.w3.org/WAI/WCAG22/Understanding/text-spacing.html
[2] World Wide Web Consortium Web Accessibility Initiative. Understanding Success Criterion 1.4.4: Resize Text. https://www.w3.org/WAI/WCAG22/Understanding/resize-text.html
[3] World Wide Web Consortium Publishing Maintenance Working Group. EPUB Accessibility 1.2. Candidate Recommendation Draft, 18 August 2026. https://www.w3.org/TR/epub-a11y-12/
[4] Scaltritti, M., A. Miniukovich, P. Venuti, R. Job, M. De Caro, and M. L. Marangolo. “Investigating Effects of Typographic Variables on Webpage Reading Through Eye Movements.” Scientific Reports 9 (2019). https://www.nature.com/articles/s41598-019-49051-x
[5] Kuster, Sanne M., Marjolijn van Weerdenburg, Marjolein Gompel, and Anna M. T. Bosman. “Dyslexie Font Does Not Benefit Reading in Children With or Without Dyslexia.” Annals of Dyslexia 68, no. 1 (2018): 25–42. https://pmc.ncbi.nlm.nih.gov/articles/PMC5934461/
[6] Wallace, Shaun, Zoya Bylinskii, Jonathan Dobres, Bernard Kerr, Sam Berlow, Rick Treitman, Nirmal Kumawat, Kathleen Arpin, Dave B. Miller, Jeff Huang, and Ben D. Sawyer. “Towards Individuated Reading Experiences: Different Fonts Increase Reading Speed for Different Individuals.” ACM Transactions on Computer-Human Interaction (2022). https://research.adobe.com/publication/towards-individuated-reading-experiences-different-fonts-increase-reading-speed-for-different-individuals/
[7] W3C Publishing Community Group. Accessibility Metadata Display Guide for Digital Publications 2.1. Community Final Specification, 22 December 2025. https://www.w3.org/publishing/a11y/metadata-display-guide/guidelines/