Abstract
A conversion can retain every word, asset, and heading while changing the relations through which a publication makes meaning. A caption can arrive before the paragraph it qualifies; a margin quotation can become a distant block; an image can retain alternative text without retaining the reason it enters an argument at a particular moment. The ordinary language of “fixed” and “reflowable” format often hides these questions by turning a publication decision into a contest between fidelity and flexibility.
This paper introduces the format promise: the stated relation among textual sequence, spatial arrangement, visual composition, reader control, and interpretive meaning that a publication claims to preserve or enable. It distinguishes three orders that should not be collapsed: textual sequence, visual/spatial arrangement, and programmatic or alternative-route order. It then offers a Format Promise Map, a decision record that makes a team name what must remain stable, what may adapt, what alternative route is required, and what evidence remains unavailable. The map does not select a winning format. It makes the burden of a publication claim proportionate to the relation it says matters.
The proof sheet with no missing words
The production note beside a converted essay contains three reassuring lines:
| Check | Result |
|---|---|
| Body text extracted | Complete |
| Image asset present | Complete |
| Caption text present | Complete |
No paragraph has vanished. Search finds the quoted sentence. The image file is in the package. The caption is visible. By the usual proofing vocabulary, the conversion has survived.
Then the editor opens the two editions side by side and discovers that the proof sheet has answered the wrong question.
In the original arrangement, an image enters immediately after the sentence that asks the reader to examine a familiar assumption. Its caption gives the visual evidence a name. A short quotation at the outer edge of the page disagrees with the paragraph without halting it. A note sits at the foot of the page because it qualifies the image, not the next section. These are not four isolated objects. They are a sequence of invitations: look here, read this sentence with that image, encounter this dissent at the margin, then carry the qualification into the turn.
The converted edition preserves all four objects. It does not preserve their agreement.
The caption now reaches the reader before the sentence that gives it a question. The margin quotation appears below a later section. The note is a link whose target sits several screens away. A nonvisual route announces the image description at a point that no longer explains why the picture interrupts the argument. Nothing has gone missing, but something is no longer held together.
This is not a claim that the converted edition has failed. It may give readers options that the original page could not offer: a changed display size, a different reading surface, a sequential alternative route, or a revised order of encounter that better serves another task. The point is more precise. A publication can retain its words and assets while altering the relations through which those words and assets do their work.
That is why the familiar argument about formats usually starts too late. A publisher is asked to choose “reflowable” or “fixed” layout, as if the decision were between two stable objects with settled virtues. Reflow is praised as flexibility. Fixed layout is praised as fidelity. One side is made to stand for reader agency; the other for design seriousness. Soon the discussion is no longer about a named publication. It is a referendum: Which format is modern? Which one is accessible? Which one wins?
Those questions are too large to teach a useful editorial decision.
The more exact question is this:
What does this publication promise to keep in relation, what does it permit to change, and what evidence makes either promise responsible?
A format is a bundle of promises
A publication format is often treated as a container selected at the end of editorial work. Text is written, images are placed, a layout is approved, and then an instruction arrives: export the book. The instruction sounds procedural, but it carries an interpretation of the publication. It assumes that the important qualities of the work have already been identified, that the selected container can represent them, and that any changes produced by a new rendering environment are either harmless or known.
The assumptions may be sound. They may also be invisible.
This paper calls the relevant bundle of commitments a format promise. A format promise is the stated relation among textual sequence, spatial arrangement, visual composition, reader control, and interpretive meaning that a publication says it will preserve or enable. It is neither a marketing claim nor a metaphysical claim that one file has become “the book” in a final sense. It is a compact editorial statement of responsibility.
A narrow format promise might say that ordinary body text may respond to available screen space and reader adjustments, while a labelled diagram must retain a stated relationship among labels, image, and explanatory passage. Another promise might say that a double-page sequence matters visually, while a logical reading order and a description route remain separately available. A scholarly edition might say that a page location supports citation in one version without implying that page position is the only valid way to locate a passage. A children’s illustrated work might say that a particular image sequence performs a narrative function, while also identifying what information an alternative route must convey.
These examples do not decide which choice should be made. They demonstrate the level at which a decision becomes inspectable. An editor cannot responsibly say that fixed presentation is necessary until the relation needing stability is named. A publisher cannot responsibly say that reflow solves an access problem until the relevant adjustment, task, and evaluation are named. Words such as faithful, accessible, and flexible are not useless. They are incomplete until attached to an object, a relation, a reader task, and a limit.
EPUB 3 supplies a technical foundation for this distinction without settling it. The W3C Recommendation defines EPUB as a distribution and interchange format for structured and semantically enhanced web content. Its package document holds information used in presentation, including rendering metadata and a default reading order. EPUB publications are intended by default to reflow to available screen space; the standard also permits pixel-precise fixed layouts, and identifies fixed-layout content documents as pre-paginated.[1] The standard therefore establishes that both adaptable presentation and deliberately constrained presentation belong inside an actual technical landscape. It does not tell an editor whether a given relationship in a given publication is significant.
Johanna Drucker’s account of the virtual codex is helpful precisely because it moves the discussion away from a contest of icons. She argues that a book should be understood through what it does as well as what it looks like, using marginalia, tables of contents, and other paratextual features to think about page space and electronic space.[5] Her argument is theoretical, not a current-format test or an instruction to reproduce print in digital form. Its teaching value is more modest: reading arrangements perform actions. They create adjacency, delay a qualification, invite comparison, separate commentary from primary text, make a return route visible, or demand that a reader hold a relationship in memory.
A format promise begins when a team is willing to name those actions instead of treating them as the natural background of a familiar page.
Two formats, three orders
The language of fixed and reflowable presentation is useful only when it does not erase the different kinds of order involved in reading. At minimum, a publication team should distinguish three:
| Order | Core question | What may carry it | What can go wrong | What the format promise must state |
|---|---|---|---|---|
| Textual sequence | In what order does the language or content proceed? | Paragraphs, sections, spine, heading hierarchy, numbered items. | A reader encounters supporting material before its premise, or cannot recover an intended progression. | What the default sequence is and whether a sequence change affects meaning. |
| Visual/spatial arrangement | What appears together, apart, above, beside, or across a turn? | Page composition, spread, table, image/caption adjacency, margin, placement, white space. | A relation that depended on spatial adjacency becomes distant, ambiguous, crowded, or falsely emphasized. | Which spatial relation is semantically or pedagogically significant, if any. |
| Programmatic / alternative-route order | What sequence does a nonvisual, keyboard, assistive-technology, or structural route expose? | Semantic markup, tags, source order, navigation document, labels, descriptions, links. | A visually clear arrangement becomes an incoherent or incomplete sequence outside the visual page. | How the relevant relation is made recoverable through another route, and what remains different. |
The table does not imply that all three orders must be identical. A comparison table may be clearer when sighted as two columns, while a sequential route needs headers and an ordered account of each relationship. A decorative running header may be visible but unnecessary in a spoken route. An image description may need to explain information that a visual reader receives through a placement and a caption. The issue is not sameness. It is whether a publication has examined the difference deliberately.
W3C fixed-layout accessibility guidance makes this distinction explicit. The guidance recognizes that publishers may choose fixed layouts because print arrangement matters to a text or its artistic purpose. It also identifies recurring challenges: navigation, reading order, text alternatives, smaller displays, visual adjustability, and the need to account for visual, audible, and tactile ways of encountering content.[2] The guidance advises authors to make programmatic order match visual order as far as possible. That does not mean every visual relation can be reduced to one linear path. It means that visual coherence cannot excuse an unintelligible source sequence.
The W3C’s tagged-PDF technique offers a useful comparison. For tagged PDF, logical reading order is primarily determined through tag order and the document’s content-tree structure. The technique warns that a multi-column document can look clear to a visual reader while being read as one interleaved column if it is not correctly structured. It also notes that complex layouts with graphics, tables, footnotes, sidebars, and form elements may not convert into correct order without repair.[4] This is not a judgment about PDF as a format, nor a guarantee about a particular authoring process. It is a disciplined reminder that a visible page and an exposed sequence are related but separate editorial facts.
The distinction is especially important for a publication that contains images, captions, annotations, tables, sidebars, or critical apparatus. The editorial question is not merely, “Will the caption appear?” It is, “What does the caption qualify, how is that relation recognized, and what does another route need in order to recover the relevant connection?” A production proof that checks only object presence cannot answer this question.
One controlled study helps show why an arrangement question should not be dismissed as visual fussiness. Kłyszejko and colleagues compared pagination and scrolling, with and without margin callouts, in a four-condition experiment involving 42 participants reading a 400-word moderately challenging text on a 22-inch monitor. They reported condition-specific differences in recall and eye-movement measures; the scrolling condition involved more glances and longer glance duration, and the effect of callouts depended on presentation mode.[6] The study is not an EPUB/PDF comparison, an e-book field trial, a mobile study, or a general rule that pagination produces better reading. Its text was short, its conditions were controlled, and its callouts may themselves have affected results. Its relevance is narrower: the way material is segmented and placed can become part of the task a reader performs.
A format promise therefore includes a question that a standard file check cannot resolve: what relation is the reader being asked to navigate?
Adaptability is a capability, not an automatic verdict
Reflowable presentation often receives a necessary defense because it can let a publication respond to available screen space and reader adjustment. That capability matters. A reader may enlarge text, change a display condition, use a different surface, encounter content in a sequential route, or rely on functions provided by a reading system. The capacity to avoid a rigid visual snapshot can make a publication more usable in circumstances that a single page cannot anticipate.
But reflow is not an automatic accessibility verdict.
The W3C’s EPUB Accessibility 1.2 Explainer is expressly limited to reflowable EPUB publications and directs readers to separate fixed-layout guidance for the distinctive difficulties of that form. It further notes that the accessibility standard does not prescribe a single evaluation or reporting method.[3] The implication is not that an editor should give up on evaluation. It is that a feature label cannot perform the evaluation in advance. Reflowability describes an adaptable presentation possibility. It does not, by itself, establish a publication’s structure, navigation, reading-system behavior, alternative route, or reader outcome.
A situated-access study by Muir, Veale, and Nichol offers a complementary warning against format labels. Their initial in-depth work with seven students, including students with dyslexia and visual impairment, examined the experience of using e-books for academic coursework and reported information-accessibility issues.[7] This small study is not statistically generalizable, not a comparison of fixed and reflowable formats, and not an account of contemporary reading systems. It does, however, justify an editorial discipline: access claims must name a reader task and a delivery condition. A publication cannot be called broadly usable merely because its text can change shape. Nor can it be dismissed because it retains a stable visual element. The relevant question is always what a particular arrangement lets a particular route and task accomplish, and what the available evidence can actually support.
The Format Promise Map does not ask an editor to predict every reader’s experience. It asks the editor to stop using a property of a format as if it were a conclusion about every reader.
Stable composition is a claim with burdens
The strongest counterpoint to a reflow-first instinct is not nostalgia for print. It is the fact that some visual arrangements do genuine work. A complex diagram can use spatial placement to state a relation. An illustrated sequence can establish action, tone, recurrence, or contrast through position across a page or spread. A caption may be a precise guide to evidence. A scholarly facsimile may need a stable relation between transcription, image, and note. A page turn may be part of how a work withholds, reveals, or compares material.
The W3C fixed-layout guidance recognizes this possibility when it describes print layout as potentially important to text or artistic purpose.[2] The guidance does not let that importance become a waiver. It notes that fixed presentation can make text resizing, layout adjustment, orientation changes, and reflow difficult or impossible in some circumstances, and that reading order and text alternatives require specific care. A stable composition is therefore not a self-justifying achievement. It is a claim with obligations.
A responsible claim has three parts.
| Claim component | Unhelpful version | Responsible version |
|---|---|---|
| Named relation | “The layout must be preserved.” | “The reader needs to see the label, the marked area of the image, and the explanatory paragraph in an identifiable relationship.” |
| Reason for significance | “It is part of the design.” | “The comparison between two image states performs an evidentiary step that the prose alone does not make.” |
| Alternative-route account | “An image description is included.” | “The description identifies the comparison, labels the relevant image states, and tells a nonvisual reader what relation the visual arrangement establishes.” |
The difference is not bureaucratic. It prevents two errors. The first is to call every inherited print arrangement necessary because it is familiar or attractive. The second is to treat every spatial relation as expendable because it can be difficult to re-present. Neither conclusion can be reached from the words fixed or reflowable alone.
The proof-sheet opening can now be re-read as a practical test. The converted essay does not need to preserve the exact margin width, line break, or physical page size in every condition. It does need a team to decide whether the dissenting quotation’s relation to the claim, the caption’s relation to the image, and the note’s relation to the evidence are meaningful. If they are, the team must state what is needed to represent those relationships through visual and alternative routes. If they are not, the team should be able to say why adaptation does not alter the argument.
That is the difference between a format decision and a format promise.
The Format Promise Map
The following map is the paper’s teaching instrument. It is a record for one publication object or one bounded component of a publication: an article opening, illustrated essay, diagram set, table, critical note, children’s spread, source edition, or similar unit. It is deliberately not a product scorecard. It cannot certify accessibility, select a file format, predict reader response, or replace direct testing. Its purpose is to expose which part of a decision is already known and which part has been left as an assumption.
| Map field | Decision record |
|---|---|
| Publication object | Question: What precisely is being represented? Record: A bounded object, such as “opening illustrated essay with caption and marginal quotation.” Do not infer: that the whole book has one uniform requirement. |
| Required stable relation | Question: What must remain recognizable together, apart, or in sequence? Record: The stated relation—image/caption, text/note, paired panels, parallel columns, or cited page/quotation. Do not infer: that a visual feature is important merely because it was present in print. |
| Adaptable relation | Question: What may change without altering the stated work? Record: Line breaking, page count, display width, position of a nonsemantic decorative element, or another named variable. Do not infer: that adaptation has no cost or that every reader uses the same setting. |
| Textual sequence | Question: What order must content follow? Record: Default section, paragraph, item, or panel order. Do not infer: that visible placement alone supplies a meaningful sequence. |
| Visual/spatial arrangement | Question: What visual relationship contributes to the object’s work? Record: Adjacency, comparison, orientation, grouping, separation, or page-turn relation. Do not infer: that spatial arrangement can be described by a generic “preserve layout” instruction. |
| Programmatic/alternative order | Question: What must another route expose? Record: Headings, labels, order, description, table headers, link targets, or a stated difference that remains. Do not infer: that source order is automatically correct or that one description reproduces every visual effect. |
| Reader-controlled variable | Question: What legitimate adjustment should remain available? Record: Text size, spacing, display mode, navigation route, magnification, or another verified reader control. Do not infer: that a particular reading system will always expose it. |
| Semantic dependency | Question: Why is the relation meaningful? Record: An evidentiary, narrative, navigational, citation, or instructional reason. Do not infer: a universal user preference or an aesthetic value without a stated role. |
| Evidence source | Question: What backs the promise? Record: Standard, inspected study, documented artifact, direct test, or an explicit absence of evidence. Do not infer: that one authority proves every component of the claim. |
| Conversion risk | Question: What could change, and what is the observed signal? Record: Caption displacement, lost label, ambiguous reading order, clipped table, broken target, or an unresolved relation. Do not infer: that a file is sound because it opens or validates. |
| Unresolved question | Question: What remains untested or outside scope? Record: Reader evaluation, reading-system behavior, alternative description quality, legal status, or a future test. Do not infer: that the map has decided more than it has. |
| Non-conclusion | Question: What is the map explicitly not deciding? Record: “No format choice made,” “no conformance claim,” “no universal readability claim,” or another boundary. Do not infer: that a completed map is a recommendation or certification. |
Worked map: the conversion proof revisited
The following worked example is constructed. It does not describe a named title, provider, file, device, test result, or actual user experience.
| Field | Constructed entry |
|---|---|
| Publication object | Opening of a short illustrated essay containing one image, a caption, a marginal quotation, and a footnote. |
| Required stable relation | The caption must be identifiable as qualifying the image; the note must be identifiable as qualifying the image-based claim; the marginal quotation must be recognizable as a dissenting voice rather than ordinary body prose. |
| Adaptable relation | Exact page location, line break, margin width, image scale within a usable range, and whether the note appears through a linked route or adjacent presentation. |
| Textual sequence | Claim about a familiar assumption; image/caption evidence; dissenting quotation; qualification note; next section. |
| Visual/spatial arrangement | Visual readers may encounter claim, image, and caption as a compact evidentiary group while seeing the dissenting quotation as adjacent but distinct. |
| Programmatic/alternative order | A route identifies the image, caption, dissenting quotation, and note in a sequence that explains their respective roles; the publication does not pretend that the alternative route reproduces the page’s visual timing exactly. |
| Semantic dependency | The image is evidence, not decoration. The quotation complicates the claim; the note qualifies the evidence. |
| Evidence source | Constructed teaching artifact; source-backed distinctions about reading order and alternative routes from W3C guidance.[2] [4] |
| Conversion risk | Objects remain present but their relation becomes ambiguous or temporally distant. |
| Unresolved question | Direct reader evaluation of whether the alternative route makes the evidentiary relation sufficiently clear. |
| Non-conclusion | The map does not select a fixed or reflowable format and does not certify accessibility. |
Notice what the worked map does not say. It does not claim that margins are inherently meaningful, that image descriptions are always enough, that reflow must avoid all sequence changes, or that fixed layout is required whenever an image appears. It makes a smaller demand: if a team invokes a relation as a reason for a format decision, the team should be able to state the relation, explain its significance, describe the route that must expose it, and name the evidence it lacks.
A proportionate decision threshold
A paper can become less honest when its final paragraph forces a winner. Reflowable presentation and fixed presentation each describe capacities and constraints. Neither word specifies the publication object, reader task, semantic relation, alternative route, or evidence burden that an actual decision requires.
The responsible question is not, “Which format is the book?” It is:
Which relation must stay stable, which relation may adapt, and what evidence would make that claim proportionate?
The answer may be a narrowly documented stable composition. It may be an adaptable reading text with structured alternatives. It may be a publication that carries different promises for different components. It may be an unresolved question requiring direct testing rather than a confident declaration. The Format Promise Map has succeeded when it makes this uncertainty usable—when it replaces a format slogan with a clear account of what a publication says it will hold together and what it still needs to learn.
References
[1] World Wide Web Consortium. EPUB 3.3. W3C Recommendation, 13 January 2026. https://www.w3.org/TR/epub-33/
[2] World Wide Web Consortium Publishing Maintenance Working Group. EPUB Accessibility — Fixed Layout Challenges and Best Practices. W3C Group Note, 2 April 2026. https://www.w3.org/TR/epub-fxl-a11y/
[3] World Wide Web Consortium Publishing Maintenance Working Group. EPUB Accessibility 1.2 Explainer. W3C Group Note Draft, 4 August 2026. https://www.w3.org/TR/epub-a11y-explain-12/
[4] World Wide Web Consortium Web Accessibility Initiative. Technique PDF3: Ensuring Correct Tab and Reading Order in PDF Documents. https://www.w3.org/WAI/WCAG22/Techniques/pdf/PDF3
[5] Drucker, Johanna. “The Virtual Codex from Page Space to E-space.” Lecture presented at the Syracuse University History of the Book Seminar, 2003. https://www.philobiblon.com/drucker/
[6] Kłyszejko, Zuzanna, Anna Maria Wieczorek, Justyna Sarzyńska, Agnieszka Szóstek, Karolina Chmiel, Tomasz Soluch, and Anna Brzezicka. “Mode of Text Presentation and Its Influence on Reading Efficiency: Scrolling versus Pagination.” Studia Psychologica 56, no. 4 (2014): 309–321. https://www.studiapsychologica.com/uploads/WIECZOREK_SP_4_vol.56_2014_pp.309-321.pdf
[7] Muir, L. J., T. Veale, and A. Nichol. “Like an Open Book? Accessibility of E-book Content for Academic Study in a Diverse Student Population.” Library and Information Research 33, no. 105 (2009): 90–109. https://eprints.whiterose.ac.uk/id/eprint/74739/