Shape: academic report
The shape of a report written for a course, an internship, a master's degree or a lab. It is one skeleton for four kinds, each part marked with the kinds that need it, and every heading is in French and in English.
Its body is the report shape of reader-ergonomics, a solved problem written up step by step. This shape puts that body inside the frame an institution asks for: a title page, an abstract, the context, the references. The evidence is the reports review, and its baseline is the technical-report standard NISO 2010.
When to use it
| Code | Kind | Read by |
|---|---|---|
| PR | course project report, rapport de projet | the teacher who grades it |
| IR | internship report, rapport de stage | the school, and the host organisation |
| MT | master's thesis, mémoire de master | a jury, then a defence |
| TR | technical report, rapport technique | a lab, a company, readers of a preprint |
- Not for an article sent to a journal: use the research paper shape.
- Not for a single long proof: use the proof shape.
- When the course or the school prescribes its own headings, keep them and their order, and map each one to a part below. No study says otherwise, so the school's rule wins. Parts the school does not name, such as the introduction or the conclusion, go before and after its headings, never inside them.
- The defence (soutenance) is a talk, and belongs to reader-ergonomics-presentations.
The four kinds at a glance
Each part, and the kinds that need it (✓ needed, opt optional, · left out):
| Part | PR | IR | MT | TR |
|---|---|---|---|---|
| Title page | ✓ | ✓ | ✓ | ✓ |
| Abstract, in French and in English | opt | ✓ | ✓ | ✓ |
| Executive summary | opt | opt | ✓ | ✓ |
| Contents | past 20 pages | past 20 pages | ✓ | ✓ |
| Introduction | ✓ | ✓ | ✓ | ✓ |
| Host organisation, one page | · | ✓ | · | · |
| Mission and objectives | ✓ | ✓ | · | · |
| Literature and background | short, in the introduction | · | a chapter | short, in the introduction |
| Method, Results, Discussion | ✓ | ✓ | ✓ | ✓ |
| Reflection on the objectives | · | ✓ | · | · |
| Conclusion, standing alone | ✓ | ✓ | ✓ | ✓ |
| Recommendations and perspectives | · | ✓ | ✓ | ✓ |
| Contributions, when several authors | ✓ | · | · | ✓ |
| Use of AI tools, when the school asks | ✓ | ✓ | ✓ | ✓ |
| Acknowledgements, at the back | · | opt | opt | · |
| Glossary | · | · | ✓ | ✓ |
| References, appendices | ✓ | ✓ | ✓ | ✓ |
How each part maps to the evidence
| Part | What it does | Evidence | Grade |
|---|---|---|---|
| Title page | names the report by its finding, then the school's items: school, supervisor, host, jury, date | NISO 2010; EOST 2026 (example) | convention, untested |
| Abstract | question, method, result and limit, in about 200 words, in both languages when the school asks | NISO 2010; most examiners open the abstract, introduction and conclusion first (Mullins 2002) | convention; supported with caveat for the early reading |
| Executive summary | the answer and what to do, for a reader who reads nothing else | NISO 2010; a preference test (Silyn-Roberts 2002); a null pilot of a methods summary (Koetsenruijter 2022) | supported with caveat: no controlled test on reports |
| Introduction | the question, its scope and the plan; in a thesis, the contribution in one sentence | NISO 2010; examiners form a first impression early, usually by the end of chapter 2 or 3 (Mullins 2002) | supported with caveat: interviews with 30 examiners |
| Host organisation | one page of context, the rest in an appendix | TSE 2017 (example); UT 2024 | convention, untested |
| Mission and objectives | the objective and its success test before the method; for an internship, the learning objectives too | TSE 2017 (example); UT 2024 | convention, untested |
| Literature and background | for a thesis, a chapter that opens with its answer and ends each section on what it implies for the question | Holbrook 2007; the first impression often forms at the literature review (Mullins 2002) | supported with caveat |
| Method | the approach, why this one, what it cannot show, each step named by what it achieves | NISO 2010; examiners ask for a justified approach with its limits (Golding 2014); the step names come from the standard's worked examples | supported with caveat |
| Results | each figure beside the text that explains it | NISO 2010 | convention for reports; the figure placement is measured in reader-ergonomics |
| Discussion | the findings interpreted and tied to the literature, with no claim beyond the data | Golding 2014; NISO 2010 | supported with caveat |
| Reflection | for an internship, what was learnt against the stated objectives | UT 2024; UGA 2026 | convention, untested |
| Conclusion | repeats the answer and stands alone, with nothing new | NISO 2010 asks that it be read alone; readers say they read it (Pinelli 1982, self-report survey); examiners open it first (Mullins 2002) | supported with caveat: no controlled test of conclusion first |
| Recommendations | only when the report proposes action; in a thesis, the purpose restated and consolidated | NISO 2010; Bunton 2005 | convention, untested |
| Contributions | who did and wrote what, in a report by a team | the school's or the course's rule | convention, untested |
| Use of AI tools | whether and how AI tools were used, in the words the school asks for; for a journal, the research paper shape's declaration | the school's or the course's rule | convention, untested |
| Headings | chapter labels stay conventional; headings below them state the claim | Lorch 1996 (free recall, unfamiliar topics) | measured in that setting; untested for reports |
| Language and presentation | form corrected by the checks, not tuned past them | presentation errors make examiners read more critically (Golding 2014); graders agree poorly on form (Brown 2023, in its first round; Bodnar 2018) | supported with caveat |
| Length | the school's limit, stated on the title page; in a thesis, each chapter opens with a recap and ends with a bridge, since it is read in sittings | EOST 2026 (example); examiners spend three or four days on a thesis, spread over two to three weeks (Mullins 2002) | convention, untested |
Each key (Author Year) is listed in the bibliography. Of those keys, Mullins 2002, read in full, gives the reading order and the timing of examiners, from interviews with 30 experienced examiners. Golding 2014, read in full too, reviews 30 studies of thesis examination and adds that an examiner's first impression changes only slowly. It cites no study for that statement, so the statement is marked as the review's own.
Where academic convention and the standard disagree
| Conflict | What wins | Why |
|---|---|---|
| A story told in time order against the result first | the abstract, the summary and the conclusion carry the result first and stand alone; the body may follow time when time is the object, as in an internship | readers and examiners look at the summary and the conclusion first (Pinelli 1982, Mullins 2002); tests of the order of a whole text are mixed (Spyridakis 2000, Spyridakis 2002, Jansen 2010) |
| No findings in the introduction against an early verdict | in a thesis, the introduction ends with the contribution in one sentence; the result itself stays in the abstract | examiners form a first impression early, usually by the end of chapter 2 or 3 and often at the literature review (Mullins 2002), and it changes only slowly, a review's statement with no study cited (Golding 2014) |
| A long literature chapter against using each idea where it is needed | a chapter for the thesis, which examiners expect; folded into the introduction for project and technical reports; none for an internship | Holbrook 2007; the internship forms grade no literature (UT 2024) |
| Thanks at the front against the answer first | acknowledgements at the back | NISO 2010 makes them optional |
| Topic labels against sentence headings | labels for chapters, claims below them | Lorch 1996 |
| Impersonal against active voice, and length | the school's rule | untested either way; surveyed NASA readers preferred the third person (Pinelli 1984), a stated preference, not a test |
The principles it applies
| Part | Principle |
|---|---|
| Abstract, summary, a conclusion that stands alone | overview first |
| Method and results as steps named by what they achieve | worked examples with subgoals |
| Objectives and terms defined before the method uses them; a glossary | define before use |
| A figure for each process and relation, beside its text | draw relations, figure next to its text |
| Results kept apart from discussion | one mode per document |
| Headings that state the claim, on at most three levels | signal the structure, shallow outline |
| Detail moved to appendices | size is a prompt |
Skeleton
Copy this into a new file, or start from the LaTeX version, academic-report.tex. Each part says which kinds need it; delete the parts your kind leaves out, then number those left; LaTeX numbers its sections itself.
# Title that states the finding
Author, school and degree, course or host organisation, supervisor, date. The school's length limit, if it sets one.
## Résumé · Abstract
> All kinds but PR, where it is optional. The question, the method, the result and its limit, in about 200 words.
> The same text in French and in English when the school asks, with at most five keywords.
## Synthèse · Executive summary
> MT and TR; optional for PR and IR. The answer and what to do, for a reader who reads nothing else, in one page.
## Sommaire · Contents
> MT and TR; PR and IR past 20 pages.
## Introduction
> All kinds. The question, its scope and the plan of the report. MT: end with the contribution in one sentence.
> PR and TR: the background the question needs, briefly, here rather than in a chapter.
## Contexte · Host organisation
> IR only. One page: what the organisation does, and where the internship sat in it. The rest goes to an appendix.
## Mission et objectifs · Mission and objectives
> PR and IR. The objective and the test that says it is met; IR: the learning objectives too.
## État de l'art · Literature and background
> MT only, as a chapter; PR and TR keep it short, in the introduction. Examiners often form their first impression
> here, and give about a tenth of their report to it (Mullins 2002; Holbrook 2007, both of doctoral theses). Open with a
> one-paragraph answer: what is known, what is missing, and where this thesis stands.
### Champ et recherche · Scope and search
> What the chapter covers and leaves out, and how the sources were found: databases, terms, dates.
### Thèmes · Themes
> One section per theme, not per source, each ending on what it implies for the question.
### Le manque · The gap
> What no source settles, stated as a claim this thesis can test.
### Position · Position
> Where this thesis stands, and the method the gap calls for; it leads into the method chapter.
## Méthode · Method
> All kinds. The approach, why this one, and what it cannot show. One step per subsection, named by what it achieves,
> the figure of the process first.
## Résultats · Results
> All kinds. One subsection per result, its headline sentence first, the figure beside the text that explains it.
## Discussion
> All kinds. What the results mean, how they compare with the literature, and their limits; no claim beyond the data.
## Retour réflexif · Reflection
> IR only. What was learnt, against the learning objectives stated under Mission and objectives.
## Conclusion
> All kinds. The answer again, standing alone: a reader who opens the report here understands it. Nothing new.
## Recommandations et perspectives · Recommendations and perspectives
> IR, MT and TR, only when the report proposes action. MT: the purpose restated and the contribution consolidated.
## Contributions
> PR and TR written by more than one author. Who did and wrote what.
## Déclaration d'usage de l'IA · Use of AI tools
> All kinds, when the school or the course asks. Whether and how AI tools were used, in its words.
## Remerciements · Acknowledgements
> Optional, IR and MT, at the back so that the answer comes first.
## Glossaire · Glossary
> MT and TR. Every term defined at its first use, and listed here.
## Bibliographie · References
> All kinds, in the school's style; every entry with a DOI or a stable URL.
## Annexes · Appendices
> Detail a reader can skip: the host organisation's longer presentation, raw data, code listings.In a thesis, open each chapter with two lines recalling where the argument stands, and end it with one line leading to the next: an examiner reads it over several days (Mullins 2002).
Worked example
A technical report (TR) on a real choice made in this family. It has the parts a TR needs, numbered once the parts it leaves out are deleted.
A complete master's thesis example awaits a thesis published under an open licence, CC BY or public domain, that can be reproduced in part. The reports review's sources hold none, so the thesis keeps the skeleton and its literature sub-skeleton above. The decision it records is reader-ergonomics-presentations' decision 0007.
# Graphviz drew a five-box diagram with no overlap, where Mermaid and D2 did not
Technical report, reader-ergonomics-presentations, 29 October 2026. Two authors: the maintainer and an AI assistant.
## Résumé · Abstract
Cinq outils ont dessiné la même figure, quatre boîtes et trois flèches sur trois niveaux, avec une palette et une
police : Mermaid, Graphviz, D2, TikZ et un script Python. Graphviz seul l'a placée sans défaut ; une seule figure a
été testée. Mots-clés : diagrammes en code, placement de graphes, Graphviz.
Five tools drew the same figure, four boxes and three arrows on three levels, with one palette and one font. Graphviz
alone placed it with no defect; one figure was tested. Keywords: diagrams as code, graph layout, Graphviz.
## Synthèse · Executive summary
Draw diagrams of relations with Graphviz and keep each `.dot` beside its SVG; keep TikZ for LaTeX documents and
Mermaid for GitHub comments. The cost: GitHub does not draw `.dot`, so each diagram is two files kept in step by a check.
## Sommaire · Contents
1 Introduction · 2 Method · 3 Results · 4 Discussion · 5 Conclusion · 6 Recommendations · Contributions ·
Use of AI tools · Glossary · References
## 1. Introduction
The family's figures were typed in characters. A first draft of a figure shape chose Mermaid because GitHub draws it
from source, not because of how it reads. Question: which tool draws a small diagram of relations most cleanly?
Answer, on one figure: Graphviz.
## 2. Méthode · Method
- One figure: where reader-ergonomics-topic sits, four boxes, three arrows, three levels.
- Five tools: Mermaid 11.12 from the command line, Graphviz 2.43, D2 0.7.1 with its ELK layout, TikZ under LuaLaTeX,
and the family's graph_diagram.py.
- The same three colours, one per kind of repository, and the same font, Inter, in each.
- Each rendered to PNG at one width and read side by side for overlaps, broken names, and labels away from their arrow.
## 3. Résultats · Results
| Tool | Defects seen | Placement |
|---|---|---|
| Mermaid | a name broken across two lines; a level label on an arrow | automatic |
| Graphviz | none | automatic |
| D2 (ELK) | two arrow labels printed over each other | automatic |
| TikZ | none | by hand |
| graph_diagram.py | greyscale by design; an arrow crossed its label | by hand |
## 4. Discussion
Graphviz was the only automatic layout with no defect. Mermaid and Graphviz both use a layered layout, so the
difference here came from their defaults, not their algorithm. With larger diagrams, crossings matter most to a reader
(Purchase 1997), which favours a layout built to reduce them. Limit: one figure, read by one reader, with no measure of
reading speed.
## 5. Conclusion
On a small diagram of relations, Graphviz gave the cleanest automatic layout of five tools.
## 6. Recommandations et perspectives · Recommendations and perspectives
- Graphviz by default; TikZ inside LaTeX; Mermaid where GitHub must draw a diagram in a comment.
- Test again on a figure of about twenty nodes before generalising.
## Contributions
The maintainer chose the tool from the comparison. The assistant drew the five versions and wrote the report.
## Déclaration d'usage de l'IA · Use of AI tools
An AI assistant wrote the five sources, rendered them and drafted this report; the maintainer read the figures and
decided.
## Glossaire · Glossary
- **Layered layout:** nodes placed in ranks so that arrows point one way, each rank ordered to reduce crossings.
- **Diagrams as code:** a figure written as text and drawn by a program.
## Bibliographie · References
- Purchase (1997). Which aesthetic has the greatest effect on human understanding? Graph Drawing, LNCS 1353,
248-261. https://doi.org/10.1007/3-540-63938-1_67