---
name: research-report-design
description: >
  Applies professional information design to a research document: visual
  hierarchy, typography for long-form reading, whitespace, grid and page
  templates, one idea per page, chart hierarchy, accessibility, and the
  headline convention where the headline states the finding rather than the
  chart topic. Use for "make this report look professional", "the deck is a
  wall of charts", "design the report template", "fix the layout", "why does
  this report look amateur", "how big should the type be", "is this report
  accessible".
category: 12 Report Design and Compilation
ref: 12.02
tier: 1
inherits: [K2, K3, K4, K5]
---

# Research Report Design

## 1. One-line description
Turns a compiled and checked research report into a designed document, using visual hierarchy, typography, whitespace, grid, page hierarchy and accessible treatment so that the argument is legible on the page rather than buried in it.

## 2. What this skill is used for

**The research problem it solves.** A research report can be analytically sound and functionally unreadable. The evidence is right, the argument is right, and the reader cannot find either, because the page gives them no way to know what matters. The failures are consistent across the profession. Pages carry three charts of equal size, so the reader must work out which one is the point. Headlines name the chart topic ("Satisfaction by Segment") rather than state the finding, so a reader who skims the headlines learns nothing at all, and readers do skim the headlines. Body type is set too small and too wide for reading, or too large and too short, and neither was decided. Whitespace is treated as leftover rather than as the mechanism that groups related things and separates unrelated ones. Charts vary in style from page to page because they were made by three people over five weeks, so the reader spends attention decoding the format instead of the content. The base and the caveat are set in six-point grey at the bottom of the page, which is a design decision that quietly removes a disclosure the researcher was required to make. Design is often treated as the cosmetic layer applied at the end. It is not: it is the layer that determines what a reader actually receives, and a document whose hierarchy is wrong communicates the wrong thing regardless of what the words say.

**Where it sits.** Reporting, after compilation and before or alongside final QA. It takes a report whose claims are settled and makes them legible. It never alters a claim.

**Typical use cases.**
- A compiled report that reads well and looks like a working document.
- A deck that has become a wall of charts, where nobody can say what the point of any page is.
- Building or revising a report template that a team will use across many projects.
- A report that must meet an accessibility standard for a public sector or regulated client.
- A document assembled by several people whose charts and pages do not match.
- A dense reference deliverable that also has to be readable, where the density trade-off needs to be made deliberately rather than by accident.

**Who uses it.** Research directors and senior researchers who own the finished deliverable; insight managers building a team template; consultants without a designer; designers working on research documents who need to understand what the research disclosures require.

## 3. When to use it

- The report is compiled, the claims are settled, and the document now has to be made readable.
- Nobody can state what the primary idea of a given page is, which is a hierarchy problem rather than a content problem.
- Headlines name topics rather than state findings, so the document cannot be skimmed usefully.
- The document will be read on screen, printed, projected, or all three, and no one has decided which reading mode it is built for.
- Several people produced charts and pages, and the document does not look like one document.
- The deliverable must meet an accessibility requirement, or will be read by an audience with a known range of visual acuity, which in practice is every audience.
- A team template is being created or revised, so the decisions will be reused across many projects and are worth making properly once.
- The document is dense and the client wants it both as a reference and as a narrative, and the trade-off must be made explicitly.

## 4. When NOT to use it

- **The claims are not settled.** Designing an unfinished report is expensive, because every claim change moves the page. Worse, design pressure starts to shape content: a page with room for four bullets acquires a fourth. Compile and check first, per **12.03 Research Report Compilation**, then design. The tell is a designer asking whether a finding can be shortened to fit.
- **The problem is structural rather than visual.** If the chapter order is wrong, if a chapter has no conclusion, if the answer arrives on page 40, no amount of typographic care will fix it. That is **12.01 Research Report Architecture**. Design makes a good structure legible. It cannot rescue a bad one, and applying it to one produces an attractive document that still does not work.
- **The question is which chart to use.** Whether a finding is best shown as a bar, a line, a distribution or a table is chart selection, and it belongs to **11.04 Data Visualisation and Chart Selection**. This skill takes the chart type as given and owns how charts are treated as a family: consistent scales, consistent colour meaning, consistent labelling, and their hierarchy on a page. In the other direction, where design reveals that a chosen chart cannot carry its claim legibly at the size available, that goes back to 11.04 rather than being solved by shrinking the type.
- **The deliverable is a live presentation.** Slides built to be spoken over have different rules: less text, larger type, one idea held for a longer time, and a build sequence. That is **11.05 Research Presentation Development**. Designing a presentation as a document, or a document as a presentation, produces the two most recognisable failures in the field.
- **The deliverable is a short executive artefact.** A one-page summary or board note is designed around what survives on a single page, which is **12.05 Executive Research Reporting**.
- **Brand guidelines are being asked to determine content.** Design implements the commissioning organisation's visual system. It never determines what the report concludes, what a base says, or whether a caveat appears. Where a template genuinely cannot accommodate a required disclosure, per K4 §4.3 and §7, the template changes.
- **The document is being checked rather than made.** Reviewing a finished report against its own standards, including its presentation and accessibility, is **12.06 Research Report QA**. Do not redesign a report as a way of reviewing it.
- **There is no time to do it properly and the alternative is a partial redesign.** A half-applied design system is worse than none: inconsistency reads as carelessness and undermines the analysis it presents. Where time is short, apply the small set of decisions with the largest effect (headline convention, one idea per page, consistent base placement, body type size and measure) and leave the rest alone deliberately.

## 5. Required inputs

**Required.** Without these the skill cannot run. If absent, ask. If no answer is available and work must proceed, state the assumption at the point where it bites, per K5 §5.

- **The compiled report with claims settled**, including every headline, chart, base description, question reference and caveat. Design over moving content wastes the work twice.
- **The reading mode.** Screen, print, projection, or a stated combination, with the primary one named. Every typographic and density decision depends on it and none of them can be made without it.
- **The document's purpose on the density spectrum:** a narrative document to be read, a reference document to be searched, or a stated hybrid. This is the single decision that most determines how the pages look.
- **Any brand or template system supplied by the commissioning organisation**, including its palette, type and logo rules, and a statement of what in it is mandatory and what is guidance.
- **The accessibility requirement**, if one is stated. Where none is stated, apply the baseline in step 9 anyway, because the absence of a stated requirement is not evidence of an audience that does not need it.

**Optional, and what each one adds.**

- **The architecture document from 12.01**: supplies chapter altitude and page budgets, which are the input to page templates and to the density decision per chapter.
- **A previous report in the same series**: shows the visual conventions the audience is used to and where deviation will need explaining. Its risk is inheriting decisions nobody made.
- **The chart source files rather than images**: allow scales, colour assignment, label sizes and segment order to be fixed consistently rather than patched. Working from exported images guarantees an inconsistent chart family.
- **A stated output pipeline** (how the document will be produced, exported and distributed): determines what is achievable, particularly for accessible export, and prevents a design that cannot survive its own production.
- **Known audience characteristics** such as typical age range or viewing conditions: raise the floor on type size and contrast, which is a design decision better made from information than from a default.

## 6. Questions to ask before starting

1. **How will this be read, and in what conditions?** Screen at arm's length, printed and annotated, projected in a bright room, or read on a phone in a taxi. Determines type size, measure, contrast, density and how much a page can carry. *Default if unanswered:* design for screen reading at 100%, check that it survives print at A4, and state the assumption.
2. **Is this a document to be read or a reference to be searched?** Determines the density contract (step 1). A narrative document with reference density is unreadable; a reference document with narrative density is incomplete and doubles in length. *Default:* narrative document with a dense appendix, which serves both without compromising either.
3. **What is genuinely mandatory in the brand system, and what is convention?** Determines the design space, and inherited constraints are almost always assumed to be firmer than they are. *Default:* treat colour, logo and typeface as mandatory and everything else as available, and confirm.
4. **What accessibility standard applies, and is an accessible export required?** Determines contrast floors, minimum type size, colour independence and structural tagging, and whether the production pipeline can deliver them. *Default:* apply the baseline in step 9 and state that a formal conformance claim has not been tested.
5. **Will pages be extracted and reused in other documents?** Determines how self-contained each page must be: its own headline, base, source and context. *Default:* assume yes, because it is nearly always true and the cost of designing for it is low.
6. **Who produces the final file, and in what tool?** Determines what design decisions can actually be implemented and maintained. A system that only one person can apply is not a system. *Default:* design to what the team can reproduce.

## 7. Step-by-step methodology

**What report design is.** The construction of a small set of rules, applied without exception, that let a reader see the structure of an argument before reading a word of it. Almost all of the value comes from a handful of decisions made once and held. Most bad research documents are not badly designed in a hundred ways; they are undesigned, and the hundred inconsistencies follow from that.

**Step 1. Set the density contract.** Decide, and write down, where this document sits between two honest extremes. A **narrative document** carries one primary idea per page, generous space, few charts, and is read front to back; it persuades well and answers a specific question badly. A **reference document** carries high information density, complete tables, consistent locations for every element, and is searched rather than read; it answers specific questions well and persuades badly. Neither is superior, and the failure is not choosing. A document trying to be both without saying so becomes dense enough to be unreadable and incomplete enough to be unusable. The usual right answer for a research report is a narrative main document with a genuinely dense appendix or data book, and the two are designed to different rules and look different, which is correct and should be visible. Where a client requires a single dense deliverable, accept it, and compensate with navigational structure (step 9) rather than by thinning the content. *Correct result:* a written density decision, per section, that a designer could apply without asking.

**Step 2. Set the grid.** Choose a column grid and a set of anchor positions, and hold them for every page. The grid does one job: it puts the same kind of thing in the same place on every page, so the reader stops looking for it. Fix the position of the headline, the main visual area, the supporting text area, and the base, source and footnote line. A twelve-column grid subdivides flexibly for most document work, and the number matters far less than the consistency. Then build a small family of page templates from it, typically four to six: full-width chart, chart with commentary, two-chart comparison, narrative text, quote, and section opener. **Resist adding a seventh template.** A page that fits none of them is usually a page carrying two ideas, and the fix is at step 5. *Correct result:* a grid with named anchor positions, and a template family small enough to be memorised.

**Step 3. Make the typographic decisions that matter, and stop.** Long-form legibility rests on a small number of decisions, and the rest is preference. Make these deliberately. **Type size**: body text large enough for the reading distance, which for screen reading is meaningfully larger than most research documents use, and never smaller than the accessibility floor at step 9. **Measure** (line length): roughly 55 to 75 characters per line for continuous prose. Wider and the eye loses the return; narrower and reading becomes choppy. This single decision fixes more legibility problems than any other and is the one most often left to whatever the page width happened to be. **Leading** (line spacing): roughly 1.4 to 1.5 times the type size for body text, tighter for large headlines, looser for small type or long measures. **Hierarchy steps**: a small number of clearly distinct sizes, typically four (page headline, subhead, body, caption or base), each visibly different from the next. Steps that are close in size read as inconsistency rather than as hierarchy. **Typeface count**: one or two families, no more, each with a constant job. Choose faces with a large x-height and clearly distinguishable numerals: research documents are full of figures, and a face that confuses a one with a seven is disqualified whatever its aesthetics. Weight, not colour, carries most emphasis. *Correct result:* a type specification of five or six lines that fits on an index card, and no decision in the document that is not on it.

**Step 4. Use whitespace as structure, not as leftover.** Space is the primary tool for showing what belongs with what, and it is more powerful than lines, boxes or colour. Two rules do most of the work. **Proximity**: elements that belong together sit closer to each other than to anything else, so a chart, its title and its base form one visual group and the next group is clearly separated. **Consistent spacing units**: pick a base spacing unit and use multiples of it, so gaps are one of a few sizes rather than a hundred. Then apply the discipline that is genuinely hard: **when a page is too full, remove something rather than reducing the space.** Compressing margins to fit one more chart is the mechanism by which a document becomes unreadable, and it happens incrementally, page by page, with each individual decision looking reasonable. Where content will not fit at the correct spacing, the page has too much content, and the correct fix is at step 5. *Correct result:* a spacing scale, and no page where whitespace was reduced to accommodate content.

**Step 5. Establish page hierarchy: one primary idea per page.** Every page communicates one primary idea, and the design makes that idea unmissable within about three seconds. Design the eye path deliberately: the reader's eye should land on the headline, move to the element that proves it, then to supporting detail, then to the base and source. Build that path with size, weight, position and space, in that order of power. Position matters more than most people credit: the top-left region of a page in left-to-right reading carries the most attention, so the headline goes there and nothing decorative competes for it. **The test for a page carrying two ideas** is simple and reliable: cover the headline and ask a colleague what the page says. If they name two things, one of them is being ignored by every reader, and the page splits. **Chart hierarchy within a page** follows the same rule. Where a page carries more than one chart, one is dominant and the others are visibly subordinate in size and prominence. Three charts of equal size on a page is not a hierarchy, it is an abdication: the reader has to work out the point, and most will not. *Correct result:* every page passes the covered-headline test, and no page has co-equal visual elements.

**Step 6. Fix the chart family.** Charts across a report must look like they came from one system, because inconsistency costs the reader real decoding effort on every page. Fix once and apply everywhere: **colour meaning** (a colour means the same thing in every chart in the document, and a segment keeps its colour throughout), **scale conventions** (whether axes start at zero, when they do not and how that is signalled, and whether comparable charts share a scale, which they should), **segment and category order** (the same order everywhere, chosen for meaning rather than for the order the data file happened to be in), **label treatment** (direct labelling in preference to legends wherever the chart allows it, because a legend makes the reader look away and back), **gridline and axis weight** (light enough to recede, present enough to be read), and **the data-to-decoration ratio** (no element that does not carry information). Chart type selection is **11.04**. This step owns consistency, and the diagnostic that a chart family has failed is that the reader can tell which analyst made which chart. *Correct result:* a chart specification covering colour, scale, order, labelling and weight, applied to every chart in the document including those inherited as images, which are rebuilt rather than patched.

**Step 7. Apply the headline convention.** Every page headline states the finding, not the topic. This is the highest-value single change available in research document design, because the population of readers who read only the headlines is large and the headlines are usually empty. Three worked pairs:

> **Before:** Satisfaction by Segment
> **After:** Satisfaction is 18 points lower among customers who contacted support in the last year

> **Before:** Reasons for Lapsing
> **After:** Lapsed members left over scheduling, not price

> **Before:** Awareness Trends 2023 to 2025
> **After:** Awareness has been flat for three years despite two campaigns

Test every headline three ways. **Is it a claim?** A claim contains a verb and can be disagreed with; a topic cannot. **Is it true of this page and nothing more?** A headline that summarises the chapter on a page showing one chart is overreach. **Would it survive alone?** Headlines get extracted into emails and summary slides with no chart attached, so a headline is a standalone assertion whether or not it was designed as one. Then apply the constraint that stops this convention doing damage: **the headline may not claim more than the page's evidence supports.** "Support contact destroys satisfaction" adds causation the design does not license, per K4 §3.2. "Lapsed members left over scheduling" needs its base on the page, and if the base is 38 the headline is directional and says so, per K4 §7. Design is where overclaiming most easily enters a report, because the pressure to make a headline land is aesthetic and the cost is evidential. *Correct result:* every page headline is a claim, defensible from the page beneath it, safe when quoted alone.

**Step 8. Fix the treatment of bases, sources, footnotes and caveats.** Give every disclosure a consistent, legible, permanent position and treatment. The base description and base size, the question reference, and the source appear on every page carrying a figure, in the same place, at a size that can actually be read, per K4 §7 and K2 §4. **Set the floor deliberately:** a base line in six-point grey satisfies the letter of a disclosure while defeating its purpose, and doing it by design rather than by accident makes it worse, not better. A material caveat is not a footnote at all: where a limitation changes how a finding should be read, it sits with the finding in body text, per K4 §4.3. The practical rule that keeps this honest: **if a reader would act differently knowing it, it is not a footnote.** Where a page carries a small base, an unverified or client-supplied input, or an untested difference, the label sits adjacent to the number rather than at the foot of the page. *Correct result:* a disclosure specification with position, size and contrast stated, and no material caveat set below body size.

**Step 9. Run the accessibility pass.** This is a baseline, not an enhancement, and it improves the document for every reader. **Contrast:** body text and small elements need a contrast ratio of at least 4.5 to 1 against their background, large text and meaningful non-text elements at least 3 to 1. Measure it rather than judging by eye; light grey on white fails routinely and looks fine to the person who made it. **Type size:** set a floor and hold it for every element including bases, captions and axis labels. **Never rely on colour alone** to carry meaning: a difference shown only by colour is invisible to a substantial minority of readers and to anyone printing in greyscale. Add a second channel, which for charts means direct labelling, position, pattern or shape, and for status indicators means a word. **Structural navigability:** real heading levels in the correct order rather than large bold text; a logical reading order; alternative text on meaningful images that states what the image shows; tables with header rows marked as headers rather than as bold text; a table of contents that navigates. **Do not use colour as the only distinction between adjacent series** in a chart, and check every chart in greyscale, which is a two-minute test that catches most failures. *Correct result:* measured contrast, a stated type floor, no colour-only encoding, and structural tagging that survives export.

**Step 10. Run the five-failure check and the consistency sweep.** Check the finished document for the five design failures peculiar to research documents. **The wall of charts:** page after page of equally weighted charts with topic headlines and no argument; the fix is steps 5 and 7, and usually the deletion of a third of the charts. **The slide that is a document:** eleven-point body text in a projected deck, or a slide format used for a document that will be read alone, which produces pages too sparse to inform and too dense to present; decide the medium, per step 1, and hand presentation work to **11.05**. **The inconsistent chart family:** the reader can tell which analyst built which chart; the fix is step 6, applied to every chart including inherited ones. **The decorative image:** stock photography and background textures that carry no information, occupy the position the evidence should hold, and signal to a sceptical reader that the page had nothing to say. Images earn their place by carrying evidence, and a research image (a shelf, a form, a queue) carries a caption stating what it shows rather than what it proves. **The caveat in six-point type:** step 8. Then sweep for consistency: same element in the same place on every page, same colour meaning throughout, same type sizes, same spacing units, same headline grammar. *Correct result:* a document where a reader can predict where to find anything after two pages.

## 8. Analytical framework

The design chain:

    Reading mode → Density contract → Grid and templates → Type system
        → Space and hierarchy → Chart family → Headline convention
            → Disclosure treatment → Accessibility → Consistency sweep

**Applying it.** Each decision constrains the ones after it, and the constraint runs one way. When a page will not work, walk back up the chain rather than adjusting the page: a page that cannot be laid out is nearly always a page carrying two ideas (hierarchy) or a document without a density decision (link two). Fixing it locally produces the exception that starts the drift toward an undesigned document.

**The hierarchy tools, in order of power:**

    Size → Weight → Position → Space → Colour

Use them in that order. Colour is the weakest and the most overused, and it is the one that fails for readers who cannot distinguish it. A hierarchy built on colour alone is not a hierarchy.

**The three-second test.** Show any page to someone who has not seen the document. After three seconds, ask what the page says. If they can state the primary idea, the hierarchy works. If they describe the layout, or name two things, or say "it's about satisfaction", it does not. This is the fastest and most reliable diagnostic in report design, and it can be run on a whole document in fifteen minutes.

## 9. Output format

**A. The design specification.** One to two pages, produced before pages are laid out, so the decisions are made once rather than per page.

- Reading mode and density contract, per section
- Grid, anchor positions, and the template family with each template's use
- Type specification: faces and their jobs, size scale, measure, leading, hierarchy steps, and the accessibility floor
- Spacing scale and the proximity rule
- Colour system: palette, semantic role of each colour, and what each means in a chart
- Chart specification: scale conventions, category order, labelling, gridline weight, greyscale requirement
- Disclosure specification: position, size and contrast for base, source, question reference and caveat, and the rule for what is body text rather than a footnote
- Accessibility baseline and whether a formal conformance claim is being made

**B. The page template set.** Each template shown with its anchor positions and an example.

**C. The designed document.**

**D. The design QA log.**

| Page | Issue | Category | Fix applied |
|---|---|---|---|

Categories: hierarchy, headline, chart consistency, disclosure, accessibility, spacing.

**E. Accessibility check record.**

| Check | Standard applied | Result | Exceptions and reason |
|---|---|---|---|

Contrast measurements, type size floor, colour-independence, heading structure, alt text, table headers, reading order.

**When the evidence is thin.** Design must never compensate for thin evidence, and this is the specific temptation of the skill. A page with one modest finding is a page with one modest finding: set it well, give it space, and let it be short. Do not add a second chart to fill the space, do not upgrade the headline to make the page feel worth its place, and do not add an image because the layout looks empty. A sparse well-set page reads as confidence. A padded page reads as padding, and per K4 §1 the padding is where fabrication enters a document that had none in it when it arrived.

## 10. Quality checks

Run before the document is presented. Sits on top of K4 §8.

1. Does every page pass the three-second test, and does every page pass the covered-headline test with a single answer?
2. Is every page headline a claim rather than a topic, true of its own page, and safe if quoted with no chart attached?
3. Does any headline claim causation, behaviour, trend or population reach the page's evidence does not support?
4. Is the base description, base size and question reference on every page carrying a figure, in the same position, above the type floor?
5. Is any material caveat set below body size or placed away from the finding it qualifies?
6. Does a colour mean the same thing in every chart, and does every segment keep its colour throughout?
7. Do comparable charts share a scale, and is any non-zero axis baseline signalled?
8. Does every chart survive greyscale, and is any meaning carried by colour alone anywhere in the document?
9. Does measured contrast meet at least 4.5 to 1 for body and small text, and 3 to 1 for large text and meaningful non-text elements?
10. Is the body measure within the legible range for the reading mode, and is leading set rather than inherited?
11. Are heading levels real and in order, are tables' header rows marked as headers, and do meaningful images carry alternative text?
12. Was whitespace reduced anywhere to accommodate content?
13. Is there any element on any page that carries no information?
14. Could a reader tell which analyst built which chart?
15. Has any claim, number, base or caveat changed during design?
16. Does the density of each section match the density contract stated for it?

## 11. Common failure modes

| Failure | How to recognise it | How to prevent it |
|---|---|---|
| **The wall of charts** | Page after page of equally weighted charts with topic headlines; no page has a point | Steps 5 and 7. One primary idea per page, headline states the finding, and delete the charts that do not carry one |
| **The slide that is a document** | Eleven-point type in a projected deck, or a slide layout used for a document read alone | Step 1. Decide the medium. Presentation work goes to 11.05 |
| **The inconsistent chart family** | Colour means different things in different charts; category order changes; three axis styles | Step 6, applied to inherited charts by rebuilding rather than patching |
| **The decorative image** | Stock photography or abstract graphics in the position the evidence should occupy | Data-to-decoration rule. An image earns its place by carrying evidence, with a caption saying what it shows |
| **The caveat in six-point type** | The base line is grey, small, and at the foot of the page; a limitation is a footnote | Step 8. If a reader would act differently knowing it, it is body text |
| **Colour-only encoding** | The chart is unreadable in greyscale or to a reader who cannot distinguish the two series | Step 9. Add a second channel: direct labels, position, pattern, shape |
| **Compression creep** | Margins shrink and type sizes drop page by page as content is added | Step 4. Remove content, never space. Any exception starts the drift |
| **Hierarchy by colour** | Emphasis is carried by a highlight colour and nothing else | Section 8. Size, weight, position, space, then colour |
| **The seventh template** | A page that does not fit any template, so a new one is made | The page usually carries two ideas. Fix at step 5 rather than adding a template |
| **Silent content change during design** | A headline shortened to fit; a base dropped to clean up a chart; a number rounded to fit a label | Guardrail 1 in Section 12. Design never edits evidence. K2 §7 |
| **Standards-by-eye** | Contrast and type size judged by looking rather than measured, by someone with good eyesight on a good screen | Step 9. Measure. Light grey on white fails routinely and looks fine to its author |
| **Brand system overriding disclosure** | A template with no place for a base line, so bases are omitted | Section 4. The template changes, per K4 §4.3 |
| **The AI-generated design opinion** | Confident typographic prescriptions with no reading mode, no density decision and no measured contrast | Steps 1 and 9. Every prescription depends on inputs in Section 5 |
| **Partial redesign** | Half the document follows the system, half does not; the result reads as carelessness | Section 4, last bullet. Apply the small high-value set completely rather than the whole system partially |

## 12. AI guardrails

Skill-specific only. Universal prohibitions are inherited from K4 and are not repeated here.

1. **Never change a claim, number, base description, quote or caveat while designing.** Design moves things and sets things. It does not edit evidence. A headline shortened to fit a column is a claim altered by a layout constraint, per K2 §7.
2. **Never write a headline that claims more than its page supports.** The pressure to make a headline land is aesthetic; the cost is evidential. Causation, behaviour inferred from stated preference, and trend from a single wave are the three that enter here, per K4 §3.2 and §3.3.
3. **Never reduce the size, contrast or prominence of a disclosure to improve a layout.** A base or caveat that has been designed into illegibility has been removed, per K4 §4.3.
4. **Never let a brand or template system determine whether a required disclosure appears.** Guidelines govern how a document looks. Where a template cannot hold a base line, the template changes.
5. **Never add a chart, image, icon or block of text because a page looks empty.** Format is not evidence, per K4 §1, and a padded page is where content gets invented in a document that arrived clean.
6. **Never encode meaning in colour alone**, in any chart, table, status indicator or callout.
7. **Never state that a document meets an accessibility standard without the checks in step 9 having been run and recorded.** An untested conformance claim is a fabricated verification.
8. **Never prescribe type sizes, measures or contrast without knowing the reading mode.** A prescription made without it is a guess presented as a standard.
9. **Never rebuild a chart in a way that changes what it shows**, including changing an axis baseline, rescaling a comparison, reordering categories in a way that implies a ranking the data does not support, or dropping a series to simplify the page.

## 13. Best-practice principles

1. **Design determines what the reader receives, so it is part of the argument, not a coat of paint on it.** A page whose hierarchy is wrong communicates the wrong thing however good the words are.
2. **Almost all the value comes from a few decisions held without exception.** Measure, leading, type scale, one idea per page, headline convention, disclosure placement, colour meaning. Everything else is preference.
3. **A headline that states the finding is the highest-return change available.** Most readers read the headlines and nothing else, and most research headlines are empty.
4. **Whitespace is structure.** It is the strongest tool for showing what belongs with what, and the first thing sacrificed by people who think of it as empty.
5. **When a page is too full, remove content.** Never compress space. Compression creep is how documents become unreadable, one reasonable decision at a time.
6. **One idea per page, and the covered-headline test proves it.** If a colleague names two things, one is being ignored by every reader.
7. **Consistency is worth more than sophistication.** A plain document applied consistently reads as competent. A sophisticated one applied inconsistently reads as careless, and the carelessness transfers to the analysis in the reader's mind.
8. **Accessibility is a baseline that improves the document for everyone.** Higher contrast, larger type, direct labelling and real heading structure make a document better for readers who have no impairment at all.
9. **Measure contrast and type size. Do not judge them.** They will be judged by the person who chose them, on the screen they chose them on, which is the least reliable available test.
10. **A colour must mean one thing throughout a document.** A segment that changes colour between chapters costs the reader real effort on every page thereafter.
11. **Density is a contract, and it must be chosen.** Documents that fail to choose become too dense to read and too incomplete to search.
12. **Never let the layout ask for content.** A page with room for a fourth bullet will get one, and the fourth bullet was never in the evidence.
13. **The document is finished when a reader can predict where to find anything after two pages.** Not when it looks impressive.

## 14. Worked example

*Fictional scenario, used for illustration only. The organisation, pages and figures below are invented for the purpose of demonstrating method.*

**INPUT.** A university's student services department has a compiled 34-page report on why undergraduates disengage from academic support services. Claims are settled and QA'd. The document is currently a slide deck built by three researchers: 31 chart pages, topic headlines throughout, four different chart colour schemes, base lines in seven-point grey, three stock photographs of students on grass, and a brand template with a fixed logo band that occupies the top-left quarter of every page. The report will be read on screen by a faculty committee, printed for a subset of them, and individual pages will be extracted into departmental meetings. An accessibility requirement applies.

**PROCESS.**

*Step 1, and the first real decision.* The density contract is set as a narrative main document of 18 pages plus a dense 20-page data appendix, replacing 34 pages of undifferentiated middle density. This immediately resolves half the layout problems, because the pages that were dense because they were reference material stop competing with the pages that were dense by accident.

*Steps 2 and 3, and a template conflict.* Grid set, five templates defined. Then the brand template conflict: the fixed logo band sits exactly where the headline should go, which is why every page in the original has its headline in the second position and none of them read as headlines. Checked against the brand system, the band is mandatory on the cover and section openers and conventional elsewhere. It is reduced to a footer band on body pages, and the client is told what changed and why. Type specification: two families, four sizes, measure fixed at 62 characters for the narrative pages, leading at 1.45.

*Step 5, and the deletion.* The covered-headline test is run on all 31 chart pages. Eleven pages produce two answers, meaning they carry two ideas. Six of them split into two pages; five reveal that the second idea is a secondary cut that belongs in the appendix. Nine further pages produce the answer "I don't know", and on inspection seven of them have no finding at all: they are pages showing a distribution because a distribution existed. Those seven are cut. The main document falls from 31 chart pages to 18, and nothing was lost, which is the usual outcome of this test and the one people find hardest to believe in advance.

*Step 7, and the overreach caught.* Headlines rewritten as claims. One rewrite is corrected in review: "Poor timetabling drives students away from support services" was written from a chart showing that students who reported timetable clashes used services less. The design pressure was real, since the claim version is a better headline. It is causal language the cross-sectional design does not license, per K4 §3.2, and it becomes "Students reporting timetable clashes use support services less than half as often". A second headline, "First-year students avoid drop-in sessions", is amended to carry its base on the page, since the base is 71 and the finding is directional.

*Steps 8 and 9.* Base lines move to a fixed position above the fold of the page at body-adjacent size, measured at 7.1 to 1 contrast. Two material caveats move from footnotes into body text next to their findings, per K4 §4.3. The accessibility pass finds three failures: a two-series line chart distinguished only by colour, at a contrast ratio between the series that fails in greyscale, fixed by direct labelling both lines at their end points; a "traffic light" table encoding status only in cell colour, fixed by adding the word; and heading levels applied as bold body text rather than as real headings, which broke navigation on export and was invisible on screen. The three stock photographs are removed. Two fieldwork photographs of the actual drop-in space are retained, each with a caption stating what it shows and no claim about what it proves.

**OUTPUT.** A two-page design specification; five page templates; an 18-page narrative document with a 20-page appendix at reference density; every headline a claim; a chart family with one colour meaning throughout and every chart legible in greyscale; disclosures at a fixed position and measured contrast; two caveats promoted to body text; an accessibility check record with three fixed failures and one exception noted; and a design QA log recording the seven pages cut and the one causal headline corrected.

## 15. Advanced usage

**Building a reusable team template.** The value of these decisions multiplies when they are made once for a team rather than once per project. A team template needs three things a project design does not: a written specification a new joiner can apply without asking, a small enough template family to be memorised, and a stated exception process, because an undocumented exception becomes the new default within two projects. Include the disclosure specification in the template file itself, so a base line has somewhere to go by default rather than by discipline.

**Documents that must serve screen, print and projection.** Do not average them. Design for the primary mode, then check the other two against defined tolerances: type that survives print at the intended size, contrast that survives a projector's reduced dynamic range, and a layout that survives being read at half width on a phone. Where the three genuinely conflict, produce two versions from one content set rather than one compromised version, and hand the projected version to **11.05**.

**Working under a brand system you did not choose.** Establish early what is mandatory and what is convention, because the two are usually conflated. Almost every research disclosure conflict resolves in the disclosure's favour, and it is easier to make that case before pages exist. Where a mandatory element genuinely prevents a required disclosure, escalate it as a compliance matter rather than solving it typographically.

**Retrofitting a document that cannot be rebuilt.** Where charts exist only as images and there is no time to rebuild them, apply the highest-value subset: headline convention, one idea per page, consistent disclosure placement, and the removal of decorative material. State plainly that the chart family remains inconsistent, rather than making a partial attempt at consistency that reads as error.

**When the client's stated preference conflicts with legibility.** A request for smaller type to fit more on a page, or for a caveat to be made less prominent, is a request to reduce what the reader receives. Per K4 §9, say what cannot be done and why, offer the strongest honest alternative (an appendix, a shorter document, a second version), and mark it `RESEARCHER DECISION REQUIRED` per K5 §3.1 where the disclosure itself is at stake.

## 16. Skill chain

**Recommended previous skills:**
- **12.03 Research Report Compilation.** Hands over a compiled report with settled claims, headlines, bases and caveats, which is the precondition for design.
- **12.01 Research Report Architecture.** Hands over the chapter set, altitudes and page budgets, which become the density contract and the template family.
- **11.04 Data Visualisation and Chart Selection.** Hands over the chart type for each finding, which this skill then makes consistent as a family.

**Recommended next skills:**
- **12.06 Research Report QA.** Takes the designed document and checks it, including its presentation and accessibility, against the standard it claims.
- **12.05 Executive Research Reporting.** Takes the designed report where a separate single-page or board artefact is also owed.
- **11.05 Research Presentation Development.** Takes the content where a spoken presentation version is required, which is a different design problem and not a resize.

**Runs well alongside:**
- **11.02 Executive Summary Development**, since summary design and page hierarchy are the same problem at two scales.
- **13.03 AI Output Verification**, run over the designed document to confirm no claim moved during layout.
- **K4 §4.3 and §7**, which set the disclosure requirements that step 8 implements, and **K5**, at the review points design generates: any change to a brand system, and any request to reduce the prominence of a disclosure.

---
A Yazi Supplied Skill and resource.