---
name: research-report-architecture
description: >
  Decides the structure of a research report before any of it is written: the
  chapter set, the chapter order, where methodology goes, what earns a place in
  the main document, and how much space each section gets. Use for "how should
  I structure this report", "what should the chapter order be", "give me an
  outline for the deck", "what goes in the appendix", "should methodology come
  first", "we have the findings, what shape should the report be".
category: 12 Report Design and Compilation
ref: 12.01
tier: 1
inherits: [K2, K3, K4, K5]
---

# Research Report Architecture

## 1. One-line description
Determines the structure of a research report before writing begins: what the chapters are, what job each one does, what order they run in, where the method sits, what earns main-document space, and how large each section is allowed to be.

## 2. What this skill is used for

**The research problem it solves.** Most weak research reports are not badly written. They are badly shaped, and the writing was never going to rescue them. The shape goes wrong in a small number of predictable ways. The report follows the questionnaire, so it ends with a chapter on demographics and no chapter on the answer. The largest chapter is the one with the most data behind it rather than the one that matters most, which means the reader spends most of their attention on the least important thing in the study. Methodology sits at the front for a reader who wanted the answer and at the back for a reader who needed to judge the evidence before believing it. Every chapter is a topic, so no chapter has a conclusion, and the reader assembles the argument themselves or does not assemble it at all. And most commonly, the structure is the same structure the team used last time, applied without asking whether this project has the same shape as that one. Structure is decided by default in most projects, and defaults are decided by the instrument, the template and habit. This skill makes the decision explicit, makes it early, and makes it an agreed artefact rather than something discovered halfway through drafting.

**Where it sits.** Reporting, at the front. After analysis is substantially complete and the findings are known well enough to be shaped, before compilation and before any prose exists. It is cheap here and expensive later: changing a chapter order on a page of notes takes ten minutes, and changing it in a written report takes two days and loses the source references.

**Typical use cases.**
- A completed study where the team is about to start writing and nobody has agreed what the chapters are.
- A project whose objectives cut across each other, so the objective list is not a usable chapter list.
- A report for two audiences with different reading behaviour, where one structure has to serve both.
- A tracking wave inheriting last wave's structure, where somebody has to decide whether to keep it.
- A study that produced far more material than the report can hold, where the main-document boundary has to be set on principle.
- A deliverable whose format is partly fixed by a tender, a regulator or a client template, where the remaining freedom has to be used well.

**Who uses it.** Research directors and project leads who own the deliverable's shape; senior researchers planning a write-up; consultants scoping a report against a decision; anyone who has been handed a pile of analysis and asked what the report should look like.

## 3. When to use it

- The analysis is complete enough that you know what the findings are, and writing has not started.
- The chapter set is currently either an objective list, a questionnaire order, or last project's structure, and nobody has tested whether it fits.
- The report serves a live decision and the structure needs to be built around it rather than around the research.
- Objectives overlap or cut across each other, so a one-objective-per-chapter mapping produces repetition.
- The report has two or more audiences who will read it differently, and the structure must accommodate both without becoming two documents.
- The material substantially exceeds the space available, and something has to be moved to the appendix on a defensible rule rather than by fatigue at 11pm.
- A team of several writers will produce sections, and they need a shared plan before they start or the sections will not join.
- A previous report in the same series exists and continuity is expected, so the decision is what to keep and what to change.

## 4. When NOT to use it

- **The analysis is not far enough along to know what the findings are.** Architecture designed over unknown findings is architecture designed from the questionnaire, which is the failure it exists to prevent. You do not need every cross-tab, but you need to know roughly what the study says. If you cannot state the top five findings, structure the analysis first (Category 06 and 07), not the report.
- **The task is assembling the report, not shaping it.** Once the structure is agreed, the work of inventorying inputs, prioritising the finding register, building the evidence map, reconciling conflicting numbers and drafting is **12.03 Research Report Compilation**. This skill hands 12.03 a plan and stops. It does not compile, it does not write prose, it does not adjudicate between two analysts' figures, and it does not build the conventions register. In the other direction, where 12.03 discovers during assembly that a planned chapter cannot be filled from the finding register, that is a structural fault and it comes back here rather than being papered over in the draft.
- **The question is how the pages look rather than what the pages are.** Grid, typography, hierarchy on the page, chart styling and template application are **12.02 Research Report Design**. Architecture decides that chapter three exists and what it establishes. Design decides how chapter three reads on a page.
- **The question is what the story is rather than how it is sequenced.** If the argument itself is unresolved, if you do not yet know what the study concludes, that is **11.01 Research Narrative Development**. Architecture arranges an argument. It cannot manufacture one, and a structure built over an absent narrative produces a well-organised list.
- **The deliverable is an executive artefact only.** A board note, a one-page summary or a short decision brief has a different construction logic and belongs to **12.05 Executive Research Reporting**. Applying full report architecture to a two-page deliverable produces a two-page deliverable with a table of contents.
- **The format is entirely specified and no structural freedom exists.** Some tenders, regulatory submissions and syndicated products fix the section set completely. Do not run a structural design exercise whose output cannot be used. Instead, place the argument inside the fixed format: an executive summary that carries the narrative, chapter openers that state conclusions, and a stated note that the order is prescribed. Say plainly that the structure was not chosen.
- **The report exists and needs checking.** Restructuring a finished report as a way of reviewing it destroys the original's decisions and puts a second version into circulation. Use **12.06 Research Report QA**, and restructure only if QA finds a genuine structural defect.
- **The conclusion has been fixed in advance and the structure is being designed to lead to it.** Where the chapter set has been specified by a stakeholder who already knows what the report will say, this is not architecture. Per K4 §4.2, name it, and offer either an honest structure or a clearly labelled assessment of the stated position against the evidence.

## 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 decision the report informs, who makes it, and when.** Structure is downstream of use. A report informing a live investment decision and a report providing an annual read on the same market have the same findings and should not have the same shape.
- **The agreed objectives**, in their final form. They set the completeness test at the end, and they are the first candidate chapter set even when they turn out to be the wrong one.
- **A statement of what the study found**, at the level of the top ten to twenty findings with their bases and confidence. Not the full finding register (that is 12.03), but enough that structure is designed over real content.
- **The audience, and how they will read.** Who receives it, who presents it, who will read only the summary, who will interrogate a specific number, and whether it will be read in a meeting, alone, or extracted into somebody else's deck.
- **The format ceiling.** Page or slide limit, medium, whether an appendix or data book is available, and any fixed sections imposed by a client, a tender or a regulator.

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

- **The agreed narrative or storyline** (from 11.01): converts architecture from a design task into a sequencing task, and makes chapter order a matter of following the argument rather than inventing one.
- **The previous wave's report or a predecessor in the same series**: shows what the audience is used to, what conventions carry, and where deviation will need explaining. Its risk is that it also supplies a structure nobody re-examines.
- **The questionnaire or discussion guide as fielded**: lets you see where the instrument order already matches a natural sequence, and where it does not, which is the check that stops instrument order being adopted silently.
- **Known stakeholder positions and prior beliefs**: determine where in the sequence a contradicting finding should sit, since a finding that overturns a held belief needs its evidence in front of it rather than behind it.
- **A record of how the previous report was actually used** (which sections were quoted, which were never opened): the single most useful and least often collected input to a structural decision.
- **The presentation plan**: whether the report will be walked through live, sent, or both. A document that will be presented can carry a different balance between narrative and reference than one that must survive alone.

## 6. Questions to ask before starting

1. **What decision does this report inform, and what does the reader have to believe to make it?** Determines the chapter set and the order. The chapters are the steps in the argument that gets the reader from where they are to that belief. *Default if unanswered:* structure around the objectives, and flag that decision-led structure requires an answer to this question.
2. **How will this actually be read, and by whom, in what order?** Determines where the answer sits, how much load the summary carries, and whether sections must be safe when extracted. A report read alone on a screen, a report walked through on a call, and a report mined for one number are three different structures. *Default:* assume the summary is read by everyone and the body by few, and build so the summary is self-sufficient.
3. **Is this one argument or several parallel findings?** The single most consequential structural question, and the one most often skipped. One argument gives a linear structure where each chapter depends on the one before. Several parallel findings give a modular structure where chapters are independent and order is by importance. Forcing parallel findings into a linear argument produces false connective tissue; forcing an argument into modules destroys it. *Default:* test it (step 3 in the methodology) rather than assume.
4. **Do the objectives map to chapters, or do they cut across each other?** Determines whether the objective list can serve as the chapter list. Objectives frequently overlap, and a chapter per objective then repeats the same evidence three times under different headings. *Default:* assume they cut across until you have tested the mapping, since the reverse assumption is the more expensive error.
5. **Does this audience need the method before the findings, or after?** Determines the placement of methodology and it varies genuinely by audience, not by convention. *Default:* a short method statement early, full method in the appendix, with the rule in step 5 applied.
6. **What is fixed and what is free?** Template sections, tender formats, tracking conventions, page limits. Determines the design space. *Default:* ask explicitly, because inherited constraints are usually assumed to be firmer than they are.
7. **What did the last report in this series do, and did it work?** Determines what continuity is worth and what deviation costs. *Default:* keep conventions that aid comparability, re-examine structure, and never carry a structure forward solely because it exists.

## 7. Step-by-step methodology

**What architecture is.** A written decision about the shape of a document, made before the document exists, and agreed by the people who own it. It has one output, the architecture document (step 9), and its value is that it forces the structural questions to be answered at the point where answers are cheap.

**The reference structure, and the standing warning.** A conventional full research report contains: cover and title; executive summary; research context and background; objectives; methodology; sample description; headline findings; thematic chapters; supporting evidence; implications; recommendations; conclusion; appendix. **This is a starting point for a decision, not a template to be filled.** It represents the accumulated default of the profession, which means it is a reasonable first guess and a poor final answer. Every element in it must earn its place in this specific report, and the order must be argued rather than inherited. The standing warning applies to the whole skill: **a team that produces the same structure for every project has stopped making the decision.** The tell is a report where the chapter set could be predicted from the template before anyone read the findings. If your last four reports had the same thirteen sections, at least three of them were the wrong shape.

**Step 1. Fix the decision and the reading behaviour.** Write, in one sentence each: the decision the report informs, who makes it, when they make it, and what they currently believe. Then write how the document will be read, concretely: who reads the summary only, who reads a single chapter, who will extract a page into their own deck, who will look for one number to challenge, whether it is presented live or sent cold. These two paragraphs govern every later decision and are the reason the same findings produce different reports for different clients. Where there are two audiences with genuinely different behaviour, name both and note which one the structure serves primarily, because a structure that serves both equally usually serves neither. *Correct result:* two short paragraphs that a colleague could use to predict most of your structural choices.

**Step 2. Restate objectives as questions and test the mapping.** Write each objective as a question the report must answer, in the client's vocabulary. Then test whether the questions map cleanly to chapters. Lay the top findings against the objective list and mark which objective each answers. Three patterns appear. **Clean mapping**: each objective is answered by a distinct body of evidence, and the objectives can serve as the chapter set. **Cross-cutting**: the same evidence answers two or three objectives, so a chapter per objective would repeat it; the chapters must then be built around the evidence bodies, with a coverage table showing where each objective is answered. **Uneven**: one objective carries most of the findings and another has almost none, so equal chapters would misrepresent the study. This test takes twenty minutes and prevents the commonest structural fault in research reporting, which is a report organised around the questions asked rather than around the answers found. *Correct result:* a mapping table of objective to evidence body, and a verdict of clean, cross-cutting or uneven, with the structural consequence stated.

**Step 3. Determine the shape of the story.** Decide whether the study produces one argument or several parallel findings, because they need different structures and the choice is not stylistic. **One argument** means the findings depend on each other: chapter two only makes sense once chapter one is believed, and the report builds to a conclusion. Structure it linearly, and the test is that no chapter can be moved without breaking the sense. **Several parallel findings** means the findings are independent answers to independent questions: three things are true about the market and neither depends on the others. Structure it modularly, order by importance, and let each chapter stand alone. **The mixed case is real and common**: a core argument with two or three independent findings attached. Handle it by building the linear argument as the spine and placing the independent findings after it as clearly separate sections, rather than interleaving them, which is what makes a report feel like it loses its thread halfway through. The diagnostic: try to state the report in three sentences. If the sentences need "because" and "therefore", it is an argument. If they need "and", it is parallel. If you find yourself writing "and therefore", you have a mixed case and should say so. *Correct result:* an explicit verdict, and the three-sentence statement that produced it.

**Step 4. Choose the ordering principle, and choose it from the argument.** Order the chapters by the logic the reader has to travel, not by the order the questions were asked, not by the order the work was done, and not by the order the data arrived. The four workable principles: **decision-led** (organised around the choice to be made, strongest with a live decision), **journey or chronological** (organised around a customer or process sequence, strongest when the experience is the subject), **segment or audience-led** (organised around groups, strongest when different teams own different audiences, and weakest at producing a single argument), and **thematic** (organised around the substantive themes the evidence produced, the default for parallel findings). Instrument order is not on this list as a design choice; it is on it only where the instrument already followed a natural sequence, or where a fixed format requires it. Choose one principle and hold it: a report that switches principle at chapter four reads as two reports stapled together, and readers feel the join without being able to name it. Then order within the principle by weight, leading with the finding that most changes what the reader believes. *Correct result:* a named ordering principle, a one-line justification, and a chapter sequence.

**Step 5. Place the methodology.** Method placement is a genuine audience decision and it has three defensible answers. **Before the findings**, in full, for an audience that must judge the evidence before it will accept the conclusions: academic and peer-reviewed contexts, regulatory submissions, expert or technical audiences, contested subject matter, and any situation where the method is itself in question. **A short statement early with the full method in the appendix**, for most commercial and organisational reporting: the reader gets sample, method, fieldwork dates and any material limitation in half a page, and the detail waits for whoever wants it. **After the findings**, for a senior audience reading for a decision, where a full method section at the front loses the reader before the answer arrives. Two rules override the choice. First, whatever the placement, **the disclosures required by K4 §7 travel with the findings they qualify, never only in the method section**: a base of 44, a non-probability sample, an unusual fieldwork period. A limitation in the appendix has been dropped, per K4 §4.3. Second, method placement never becomes method omission: a report whose method cannot be found is not a research report. *Correct result:* a placement decision with the audience reason stated, and a note of which disclosures are travelling with findings rather than sitting in the method section.

**Step 6. Define each chapter, and give it one job.** For every chapter write four things: a working title stating what it establishes rather than what it is about; a one-sentence statement of the job it does in the argument; the level of abstraction it operates at; and what the reader believes when they finish it that they did not believe when they started. **A chapter has one job.** A chapter doing two jobs is two chapters, or one chapter and a section that belongs elsewhere. The test is whether you can state the job without using "and" to join unrelated purposes: "establishes that the decline is concentrated in off-peak journeys" is one job; "covers usage and attitudes and pricing" is a container. The level-of-abstraction line matters more than it looks: a chapter that operates at market level should contain no paragraph about a button label, and mixed altitude inside a chapter is the most common reason a well-structured report reads as though several people wrote it. *Correct result:* a chapter table with title, job, altitude, and the belief the reader leaves with, in which no job statement contains an unrelated "and".

**Step 7. Size each section against its importance, not its evidence volume.** Assign a page or slide budget to every chapter, and set it by the chapter's weight in the decision. This is deliberately counter to the natural pull, and the natural pull is strong: **the section with the most analysis behind it grows to fit the analysis, which is how the largest chapter in a report ends up being the least important one.** A study that ran forty questions on usage and eight on the thing the client must decide will, left alone, produce a usage chapter five times the size of the decision chapter. Size deliberately: rank the chapters by their bearing on the decision, allocate the budget in that order, and then check whether any chapter is large because it is important or large because it is well-supplied. Where a chapter has more evidence than its budget, the surplus goes to the appendix and the budget holds. Where a chapter is important and thinly evidenced, the budget stays and the chapter says honestly what it can, per K3 §4.3, rather than being padded to its allocation. *Correct result:* a budget per chapter, ranked by decision weight, with a written note of any chapter where evidence volume and importance point in different directions.

**Step 8. Set the main-document boundary.** Decide what earns space in the main report and what goes to the appendix, on a rule rather than case by case. A workable rule: **the main document carries what the reader needs to follow and check the argument; the appendix carries everything else that was analysed.** By that rule the main document holds findings that advance or qualify the argument, the evidence directly supporting them with bases and question references, the material limitations, and the recommendations. The appendix holds full tables, the instrument as fielded, the complete method, secondary findings that answer an objective but do not advance the argument, and analysis that produced nothing. Three cautions. Nothing goes to the appendix because it is inconvenient, per K4 §4.1: contradicting evidence is main-document material by default. Nothing goes to the appendix because it is a caveat, per K4 §4.3. And the appendix is not a graveyard: everything analysed appears in the main document or the appendix, and material that appears in neither is a decision to be recorded, not a gap left silently. *Correct result:* an explicit boundary rule, a list of what is going to the appendix, and a note of anything cut entirely with the reason.

**Step 9. Write the architecture as an artefact and agree it.** Produce the architecture document (Section 9), circulate it, and get it agreed by whoever owns the deliverable before compilation begins. This is the step that gets skipped, and skipping it is why structural arguments happen in draft two when the report has already been written into the wrong shape. An agreed architecture also does something a verbal understanding cannot: it lets several writers work in parallel without producing sections that do not join, and it gives the project a defensible answer when a stakeholder asks late why their topic is not a chapter. **Flag the structural review points here**, per K5 §3.1: whether the primary audience is correctly identified, whether the decision weighting is right, and whether a finding that contradicts a stakeholder's stated position is placed where it will be read. Those are judgements about the organisation, not about the evidence, and they belong to a named human, per K5 §2.1 and §2.3. *Correct result:* a one to two page document, dated, agreed, and referenced by everyone writing.

**Step 10. Stress-test before handing over.** Run four tests on the plan. **The sequence test:** read the chapter job statements in order as a single paragraph. Does it argue, or does it list? **The extraction test:** take any single chapter out of context. Does it still say something true and self-contained, or does it depend on a chapter the reader may not have read? For a modular structure both must be true; for a linear one, chapter one must be. **The completeness test:** every objective is answered in a named chapter or listed as unanswered, with no objective quietly homeless. **The reader test:** walk the structure as the person who has to make the decision, and ask at each chapter break whether they now know enough to act, or whether they are still waiting. If the answer is still waiting at chapter six of seven, the answer is arriving too late and the order needs inverting. *Correct result:* four tests run, with any structural change made now rather than during drafting.

## 8. Analytical framework

The architecture chain:

    Decision → Audience and reading behaviour → Argument shape
        → Ordering principle → Chapter jobs → Sizing → Boundary → Agreed plan

**Applying it.** Each arrow constrains what follows, and the constraint runs one way. A structure that cannot be traced backwards to a decision and an audience was chosen by habit, whatever it looks like. The practical use of the chain is diagnostic: when a structure feels wrong and nobody can say why, walk back up it. The fault is almost always at an early link, most often that the decision was never fixed (link one) or that the argument shape was assumed rather than tested (link three).

**The one-job test, stated as a framework.** For any candidate chapter:

    What it establishes → why the argument needs that → what the reader believes after it
        → what would be lost if it were deleted

A chapter that survives all four is a chapter. A chapter that fails the last one ("nothing would be lost") is a container, and its contents belong in an adjacent chapter or the appendix.

**The three structural archetypes.** **Linear argument**: chapters depend on each other, order is fixed, the report builds. Strongest when the study has one thing to say and the reader has to be moved. **Modular**: chapters are independent, order is by importance, each stands alone. Strongest when several unrelated questions were answered and different readers want different ones. **Spine plus modules**: a linear core with independent findings appended, explicitly separated. The most honest structure for the common case, and better than pretending the independent findings are part of the argument. Choose one, name it in the architecture document, and hold it.

## 9. Output format

**The architecture document.** One to two pages, dated, agreed before compilation begins.

**A. The frame**
The decision, the decision-maker, the timing, and what they currently believe. How the report will be read, by whom, in what order. Primary audience named where there is more than one.

**B. Objectives as questions, and the mapping verdict**
Each objective as a question, with the evidence body that answers it and a verdict of clean, cross-cutting or uneven.

**C. Structural decisions**
Argument shape (linear, modular, spine plus modules) with the three-sentence statement that produced it. Ordering principle with its justification. Methodology placement with its audience reason. Boundary rule for main document versus appendix.

**D. The chapter table**

| # | Working title (what it establishes) | Its one job | Altitude | Reader believes after it | Budget | Objectives answered |
|---|---|---|---|---|---|---|

**E. Appendix contents**
What is going there, and what is being cut entirely with the reason.

**F. Structural review points**
Per K5 §3.3. Typically: primary audience identification, decision weighting behind the sizing, and placement of any finding that contradicts a held stakeholder position.

**G. What the architecture assumes**
The findings it was designed over, and what would change if they change. An architecture designed over draft analysis is provisional and says so.

**When the evidence is thin.** The structure contracts rather than filling. A chapter with two findings becomes a section inside an adjacent chapter, and the chapter count drops. An objective with no evidence behind it gets no chapter; it gets a line in "what we could not establish", per K3 §5.2, and that is a more useful thing for the client to receive than a thin chapter. Where the study as a whole is thinner than expected, the correct architecture is a shorter report, not the same architecture with less in each box. **A structure must never create a slot the evidence cannot fill**, per K4 §1: the slot will be filled, and it will be filled from the strongest nearby fragment.

## 10. Quality checks

Run before the architecture is agreed. Sits on top of K4 §8.

1. Can the structure be traced backwards to a stated decision and a stated audience, or was it chosen by habit?
2. Is the argument shape (linear, modular, spine plus modules) stated explicitly, and tested by the three-sentence statement rather than assumed?
3. Does every chapter have one job, statable without an unrelated "and"?
4. Does every chapter title state what the chapter establishes rather than name a topic?
5. Is the ordering principle named, justified, and held throughout, with no switch mid-report?
6. Is every objective answered in a named chapter, or explicitly listed as unanswered?
7. Does the mapping test show whether objectives are clean, cross-cutting or uneven, and does the chapter set reflect that verdict?
8. Is any chapter large because of evidence volume rather than importance?
9. Is the section bearing most directly on the decision among the largest, and if not, is there a stated reason?
10. Is methodology placed by an audience reason rather than by convention, and can it be found at all?
11. Do the disclosures required by K4 §7 travel with their findings rather than sitting only in the method section or appendix?
12. Is the main-document boundary set by a stated rule, and is nothing in the appendix because it is inconvenient?
13. Does everything analysed appear in the main document, the appendix, or a recorded cut?
14. Would any chapter be impossible to fill from the findings that actually exist?
15. Does this structure differ from the last project's structure, and if it does not, has anyone tested whether it should?
16. Has the architecture been agreed by the person who owns the deliverable, in writing, before compilation began?

## 11. Common failure modes

| Failure | How to recognise it | How to prevent it |
|---|---|---|
| **The template applied** | The chapter set could have been written before anyone read the findings; it matches the last four reports | Step 1 and the standing warning. The reference structure is a starting point for a decision |
| **Questionnaire order by default** | Chapters map one to one onto sections of the instrument, and the report ends on demographics | Step 4. Order from the argument; instrument order only where a stated exception applies |
| **The objective-per-chapter reflex** | The same evidence appears in three chapters under three headings | Step 2. Run the mapping test before assuming objectives are chapters |
| **The inflated evidence chapter** | The largest chapter is the one with the most cross-tabs and the least bearing on the decision | Step 7. Budget by decision weight, then check every chapter against evidence volume |
| **The container chapter** | A chapter whose job statement needs "and"; deleting it would lose nothing | The one-job test in Section 8, fourth question |
| **The topic title** | Chapter titles are noun phrases: "Usage", "Awareness", "Pricing" | Titles state what the chapter establishes. Test by asking whether the title can be disagreed with |
| **Mixed altitude inside a chapter** | A market-level chapter containing a paragraph about a screen label; the report reads as several authors | Set altitude per chapter at step 6 and hold it at drafting |
| **The switched principle** | The report is decision-led to chapter three and segment-led thereafter | Name one principle at step 4 and hold it. Where two are genuinely needed, use spine plus modules and separate them visibly |
| **Method as an obstacle course** | Six pages of methodology before the first finding, for a senior audience reading for a decision | Step 5. Placement is an audience decision with three defensible answers |
| **The buried answer** | The conclusion arrives at chapter seven of eight | The reader test at step 10. If they are still waiting near the end, invert |
| **The appendix as a hiding place** | Contradicting evidence and material caveats are in the appendix | Step 8, third caution. K4 §4.1 and §4.3 |
| **The AI-generated complete outline** | A plausible thirteen-section structure produced without the findings, the decision or the audience | Structure is a decision over known content. Refuse to produce one before step 1 and step 2 have real inputs |
| **The slot that demands content** | A planned chapter with no findings to fill it, discovered at drafting | Step 10, completeness test, and check every chapter against the findings that actually exist |
| **Structure by committee** | Every stakeholder's topic became a chapter | Chapters are argument steps, not stakeholder territory. Coverage of an interest is not the same as a chapter for it |
| **Silent inheritance** | Last wave's structure carried forward with no one having reconsidered it | Question 7 in Section 6. Continuity of conventions, not automatic continuity of structure |

## 12. AI guardrails

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

1. **Never produce a report structure before the findings, the decision and the audience are known.** A structure generated from a brief alone is a template with the project's words in it, and it will be filled from whatever material is nearest. Ask for the inputs in Section 5 or say what cannot be decided without them.
2. **Never create a chapter, section or slot the available evidence cannot fill.** Format is not evidence, per K4 §1. An empty slot in an agreed architecture is a standing invitation to fabricate.
3. **Never present the reference structure in Section 7 as the answer.** It is the profession's default and the starting point of a decision. Reproducing it unchanged, for any project, is a failure of the skill even when the result happens to be right.
4. **Never let the structure hide a finding.** Placing contradicting evidence in an appendix, in a late chapter nobody reaches, or inside a chapter about something else is a structural form of the omission prohibited by K4 §4.1.
5. **Never separate a material limitation from the finding it qualifies through structural placement.** Per K4 §4.3, a base size of 44 is disclosed where the finding appears, whatever the method section does.
6. **Never infer the decision, the audience or the stakeholder's prior position when they have not been supplied.** These govern every structural choice. State them as assumptions at the point they bite, per K5 §5, or ask.
7. **Never size a section from the volume of material available.** Importance to the decision sets the budget; surplus material goes to the appendix.
8. **Never design a structure to lead to a predetermined conclusion**, including one a stakeholder has stated as expected. Per K4 §4.2, that is information about the stakeholder, not about the world.
9. **Never claim a structure is decision-led, journey-led or thematic without the ordering actually following that principle.** The label is a commitment, and a report labelled decision-led that runs in questionnaire order misleads everyone downstream.

## 13. Best-practice principles

1. **Structure is a decision, and most reports are shaped by a decision nobody made.** The default was set by the template, the instrument or the last project. Making it explicit is most of the value here.
2. **The reference structure is a starting point, not a destination.** It exists so you have something to argue with. If you have not argued with it, you have not designed anything.
3. **A chapter has one job.** Everything else about chapter design follows from this, including titles, altitude and length.
4. **Order from the argument, not from the instrument.** The order the questions were asked is a fact about your fieldwork and of no interest to the reader.
5. **Size against importance, not against evidence volume.** Left alone, the report grows in proportion to the analysis behind each section, which reliably promotes the least important chapter to the largest.
6. **The three-sentence test decides the argument shape.** "Because" and "therefore" mean linear. "And" means modular. Both means say so, and separate them visibly.
7. **Method placement is an audience judgement, not a house rule.** An expert audience needs it before it will believe anything. An executive audience reading for a decision needs the answer first. Both are correct for their reader.
8. **A limitation belongs to its finding, wherever the method section sits.** Structure must never be the mechanism by which a caveat gets separated from the number it qualifies.
9. **Write the structure down and get it agreed.** An unrecorded structure is not agreed, it is assumed, and the assumption will differ between the three people writing.
10. **Design the structure so the reader can stop early and still be right.** Most readers will. The reader who reads only the summary should not end up with a different conclusion from the reader who reads everything.
11. **What goes in the appendix should be decided by a rule, in daylight, before you are tired.** Boundary decisions made at the end of a project are made by exhaustion and they systematically favour the convenient.
12. **A structure inherited from the previous wave should be re-examined even when it is kept.** Continuity is often right. Continuity nobody chose is a habit wearing continuity's clothes.
13. **If you cannot state the top five findings, you cannot design the report.** Architecture over unknown findings is architecture from the questionnaire.

## 14. Worked example

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

**INPUT.** A national health charity commissions research into why volunteer retention has fallen across its regional branches. Inputs to the architecture decision: a five-objective brief; sixteen top-line findings from a survey of 900 current and lapsed volunteers plus 22 depth interviews; a 30-page limit; a previous report from three years ago structured as one chapter per objective; and two audiences, the executive team deciding whether to fund a new regional volunteer coordinator role, and the regional managers who will be asked to change how they onboard.

**PROCESS.**

*Step 1.* The decision is a funding decision made by the executive in eight weeks. What they currently believe, on record from the brief, is that retention has fallen because volunteer expectations have changed generationally. Reading behaviour differs sharply between the two audiences: the executive will read the summary and the recommendations, and will look for the cost case; regional managers will read the chapters about their own practice and will resist any implication that the problem is local. Primary audience: the executive, because they hold the decision. The managers' needs are served by making two chapters extractable and self-contained rather than by restructuring the whole report around them.

*Step 2, and the first structural finding.* The five objectives are rewritten as questions. The mapping test shows a cross-cutting pattern: objectives two, three and four ("what do volunteers value", "what causes lapse", "what would bring lapsed volunteers back") are all answered by the same body of evidence about the first three months of a volunteer's tenure. A chapter per objective would present that evidence three times. The verdict is cross-cutting, and the chapter set is therefore built around evidence bodies, with a coverage table showing where each objective is answered.

*Step 3, and the judgement call.* Is this one argument or several? The three-sentence test produces: "Retention loss is concentrated in the first three months. It is driven by unstructured onboarding rather than by changing expectations. Branches with structured onboarding do not show the decline." That is "because" and "therefore", so it is linear. But there is a fourth finding, on volunteer travel costs, that is genuinely independent and materially important to the funding decision. Rather than force it into the argument, which would require a causal link nothing measured, the structure becomes spine plus one module, with the travel-cost finding placed after the argument as a clearly separate section. Recorded in the architecture document so the writer does not later try to join them.

*Step 4.* Ordering principle: decision-led. The executive must decide whether to fund a coordinator role, so the sequence runs: where retention is being lost, why, what distinguishes the branches that do not lose it, what that implies for the coordinator proposal, and then the independent cost finding.

*Step 5, and a note.* The executive audience does not need method before findings. But the study will be contested by regional managers who have a stake in the conclusion, so a bare method statement will not hold. Resolution: half a page early covering sample, method, fieldwork dates and the two material limitations, full method in the appendix, and the base for the branch comparison (11 branches, small) disclosed on the chart itself rather than in the method section, per K4 §4.3.

*Step 6 and 7, and the sizing correction.* Six chapters defined, each with a one-job statement. The initial page budget, drafted by evidence volume, gives eight pages to volunteer motivations, which produced the most survey questions and the most cross-tabs, and four pages to the branch comparison, which is the chapter that actually bears on the funding decision. Sized against decision weight instead: branch comparison rises to eight, motivations falls to four, and the surplus motivation analysis goes to the appendix. This is the single largest structural change made in the exercise, and it was invisible until the budget was set deliberately.

*Steps 8 to 10.* Boundary rule stated. The finding that contradicts the executive's stated generational belief is placed in chapter two, early and with its evidence in front of it, rather than in the recommendations where it would arrive as an assertion. Flagged `RESEARCHER REVIEW RECOMMENDED` per K5 §2.1, since whether to lead with a finding that contradicts the sponsor's stated position is an organisational judgement the study does not contain. The sequence test reads as an argument. The reader test shows the executive knows enough to decide by the end of chapter four, which is accepted: chapters five and six add the implications and the independent cost finding rather than the answer.

**OUTPUT.** A two-page architecture document: spine-plus-module shape, decision-led order, six chapters each with a one-job statement and a decision-weighted budget, a coverage table mapping five objectives across the six chapters, method placed early in brief with full detail appended and one base disclosed on-chart, an appendix list, and two structural review points. Agreed by the project director before compilation began, and handed to **12.03**.

## 15. Advanced usage

**Two audiences, one document.** Where a report must serve a decision-making audience and an operational one, do not build two structures. Build for the primary audience and make the operational audience's chapters extractable: self-contained, with their own bases and context, so a chapter lifted into a regional meeting still stands. Where the two audiences need genuinely different conclusions emphasised, the honest answer is one report plus a separate short artefact from **12.05**, not a compromised structure that under-serves both.

**Fixed formats with residual freedom.** Tender responses, regulatory submissions and syndicated products fix the section set. The remaining freedom is almost always in three places: the executive summary, the opening paragraph of each prescribed section, and the ordering within sections. Put the argument there. State in the front matter that the section order is prescribed, so a reader knows they hold an index rather than an argument.

**Tracking series.** Separate the two things that get carried forward. Conventions (base definitions, segment names, terminology, comparison rules) should be carried forward deliberately, because they are what makes waves comparable. Structure should be re-examined every wave, because the story changes even when the questionnaire does not. A tracking report that keeps the same chapter order for six waves is usually reporting the instrument, not the market.

**When the study answers a different question from the one commissioned.** Occasionally the evidence is strongest somewhere the brief did not point. Do not silently restructure around it, and do not bury it to protect the commissioned objectives. Build the architecture to answer the commissioned objectives, place the unexpected finding where its importance warrants, and raise the divergence explicitly with the client, marked `RESEARCHER DECISION REQUIRED` per K5 §3.1. Whether the report is re-scoped is a commercial and relationship judgement, not a structural one.

**When no structure works.** If every candidate structure produces either repetition or false connection, the usual cause is that the study contains two studies. Say so, and propose two documents or one document with a stated internal division, rather than an elegant structure that conceals the seam.

## 16. Skill chain

**Recommended previous skills:**
- **08.05 Insight Prioritisation and Sizing.** Hands over which findings matter most and by how much, which is the input the sizing decision at step 7 runs on.
- **11.01 Research Narrative Development.** Hands over the storyline, which converts architecture from design into sequencing.
- **08.03 Implication Development.** Hands over the implications, which determine whether the report needs a separate implications chapter or whether implications sit at the end of each chapter.

**Recommended next skills:**
- **12.03 Research Report Compilation.** Takes the agreed architecture and fills it, testing every chapter against the finding register and returning any chapter that cannot be filled.
- **12.04 Research Evidence Integration.** Takes the chapter set and the boundary rule, and determines which evidence streams supply which chapter with provenance intact.
- **12.02 Research Report Design.** Takes the chapter set, the altitude decisions and the page budgets, and turns them into a visual and typographic hierarchy.
- **12.05 Executive Research Reporting.** Takes the architecture where a separate executive artefact is also owed, so the two are built from one structural decision rather than diverging.

**Runs well alongside:**
- **11.02 Executive Summary Development**, since summary construction and report architecture are the same decision viewed at two scales.
- **12.06 Research Report QA**, which checks the finished document against the architecture it was supposed to have.
- **K5**, at the structural review points this skill reliably generates: primary audience identification, decision weighting behind section sizing, and placement of a finding that contradicts a stakeholder's stated position.

---
A Yazi Supplied Skill and resource.