Existing reviews first, then the seminal work by strand; strand 6 is practice; the academic-papers review already holds IMRaD, structured abstracts, CARS and reporting guidelines, which are linked, not repeated. Each strand's findings are graded: measured, supported with caveat, convention (untested), not supported. Sources from outside journals take a score instead of a grade, in sources; a source below 10 of 20 is marked as an example of practice only.
0. Existing reviews
| none located on student reports |
E1 to E5 returned no systematic review, scoping review or MLR of report structure, report reading or internship-report writing |
the gap is stated in position.md |
| Nesi 2012 |
a corpus survey of student genres, not a review of reports |
used as the nearest map (strand 1) |
| Hartley 2004 |
structured abstracts, narrative survey of 31 studies |
linked from the academic papers review |
1. Genres of student writing (R1)
| Gardner 2013 |
seminal (corpus) |
2,858 assessed texts, undergraduate and taught postgraduate, England; 13 genre families built from the texts, not from teachers' views |
measured, descriptive |
| Nesi 2012 |
survey (book) |
the totals are in Gardner 2013 Table 4 and the counts by level in the tables under its Figures 1 and 2 (full text); the corpus has four levels; Nesi's own table number is not verified (book not reached) |
measured, descriptive |
Counts from Gardner 2013, Table 4 (all levels) and the tables under its Figures 1 and 2 (by level and by disciplinary group); counts depend on the corpus version. Table 3 of the published paper holds essay prompts, not counts:
| Essay |
1,237 |
| Methodology Recount |
361 |
| Critique |
322 |
| Explanation |
214 |
| Case Study |
194 |
| Exercise |
114 |
| Design Specification |
93 |
| Proposal |
76 |
| Narrative Recount |
75 |
| Research Report |
61 |
| Problem Question |
40 |
| Empathy Writing |
36 |
| Literature Survey |
35 |
- The corpus holds no research master's thesis and no French text; its upper level is taught master's writing.
- This review's mapping of the four report kinds onto the families is a proposal, not the book's. Project report: Methodology Recount, Design Specification or Research Report. Internship report: Narrative Recount or Case Study. Technical report: Research Report or Design Specification. Thesis: none.
- Table 2 of Gardner 2013 (pp. 37 to 40) gives each family a purpose, a brief generic structure and example genres. Examples: Methodology Recount has Introduction, Methods, Results, Discussion. Design Specification has purpose, design development and testing. Research Report has aim or question, investigation, links to other research. Narrative Recount recounts events with optional comments. Case Study describes a case with recommendations. The chapters of the book that define each family's stages were not read, so no move list from the book is given.
- A study of a transnational engineering programme is said to have found Narrative Recount, Design Specification and Exercise in all four of its disciplines. It is not verified: no such study was found when searched for, and only a search summary mentions it.
2. Standards (R2)
| NISO 2010 |
standard, read in full |
required: title section, abstract (about 200 words), contents, summary, introduction, methods, results and discussion, conclusions; conditional: recommendations (when the report suggests action), references, appendices, glossary, list of figures if more than 5 |
convention, untested |
| NISO 2010 |
same |
over 50 pages a separate executive summary is often prepared; an executive summary may replace the abstract, and then should not exceed 10 pages (a soft cap; the standard's sample outline shows 3 to 15 pages) |
convention, untested |
| NISO 2010 |
same |
the introduction names the audience and holds no findings; the conclusions section is written to be read on its own and adds no new material |
convention, untested |
| NISO 2010 |
same |
organisation moved in 30 years from content-based to user-based; there is no single format, so the standard gives alternatives by purpose |
convention, untested |
| NISO 2010 |
same |
figures and tables stand as close as possible after their discussion |
convention, untested |
| ISO 1986 |
standard, catalogue record only |
ISO 7144 covers the presentation of theses and similar documents; first edition 1986; the catalogue says it was confirmed in 2024 (search snippet, secondary); the text is paywalled |
convention, untested |
3. Theses and their examiners (R1, R3)
| Paltridge 2002 |
seminal (published advice against practice) |
thesis structures; the abstract names four main kinds of thesis (traditional simple, traditional complex, topic-based, compilation of research articles) in 30 master's and doctoral theses at one Australian university, against eight advice books |
supported with caveat (full text) |
| Bunton 2005 |
corpus study |
conclusion chapters of 45 PhD theses at one Hong Kong university, 44 with a concluding chapter; of these, 36 are thesis-oriented and move through introductory restatement, consolidation of the research space (all 36), practical applications or implications, recommendations for future research and a rare concluding restatement, and 8 are field-oriented with a different structure |
supported with caveat (full text) |
| Mullins 2002 |
interview study |
30 experienced examiners (at least five theses in five years) at five universities; criteria, expected standard, key judgement points, the examiner's role and outside influences. The criteria are themes, not a fixed list: scholarship and originality, a coherent argument, adequate quantity, self-criticism and freedom from sloppiness. Most examiners open with the abstract, introduction, conclusion and references, then read in full, and form a first impression usually by the end of the second or third chapter, often at the literature review |
supported with caveat (full text) |
| Kiley 2004 |
follow-up |
interviews with 26 inexperienced examiners and supervisors; the roughly 100 examiner reports belong to a separate study by Kiley (2004) cited there; they read as experienced examiners do (abstract, introduction, conclusion, references first), lean on institutional criteria and are unsure of the bounds of a very good or poor thesis. The paper prints 2004 (volume 41(2)); its copyright and DOI say 2005 |
supported with caveat (full text) |
| Holbrook 2007 |
corpus study |
examiner comment on the literature review in PhD theses; 501 candidates, 1310 reports (1307 in the quality analysis), five Australian universities; on average about one tenth of the text of a report is devoted to the literature review |
supported with caveat (full text) |
| Bourke 2013 |
survey |
353 PhD and 74 research-masters examiners (427), 12 indicators; the importance ranking is nearly the same at both levels |
supported with caveat (full text) |
- Mullins 2002 (full text) reports by interview the order examiners follow: most of the 30 open the abstract, introduction, conclusion and references first, then read in full. Blog advice says abstract and reference list first, unsupported. No test of the order's effect was located (F8).
4. Who reads a report, and how (R3)
| Pinelli 1982 |
government survey, read in full |
378 internal and 504 external engineers and scientists; components read: conclusion 97%, results and discussion 91%, summary 87%, introduction 85%, title page 82%, abstract 68% (combined, n = 882); mean order: title page, summary, abstract, introduction, contents, conclusion |
measured (self-report survey) |
| Pinelli 1982 |
same |
components used to decide whether to read: the summary came first (69% internal, 77% external), then conclusion, abstract, title page, introduction |
measured (self-report survey) |
| Pinelli 1984 |
journal article, full text |
a NASA Langley survey (questionnaires to 513 Langley and 600 society engineers and scientists, 74% and 85% returned); of 205 managers and 633 non-managers, a majority of both wanted a summary as well as an abstract, 87% read the summary and more read it than the abstract, and the summary is what both use to decide whether to read on; 69% of managers and 68% of non-managers preferred the third person (Table 8); figures inside the text |
measured, descriptive (stated preferences) |
| Sheffield 2017 |
transcript of seven practitioner and academic interviews |
the executive summary is called the most important part; one reader goes summary, conclusion, graphs and equations, then introduction |
convention, untested |
- Pinelli 1982 notes that its respondents were about 70% researchers and 30% technical managers and does not separate them.
- No survey of what a host company reads in an internship report was located (G11).
- Host-supervisor forms (search summary only) rate conduct and skills; they are not reports' readers' studies.
5. What is measured about helping a report reader (R4)
| Lorch 1996 |
two experiments |
in free recall (Experiment 1, N = 80; familiarity set by norming ratings of 13 students), headings raised the number of unfamiliar topics recalled but not familiar ones, through topic access and not through recall of detail within a topic; in summarization (Experiment 2, N = 99) headings raised the number of topics included for familiar and unfamiliar topics alike, and mattered most for briefly discussed topics (full text) |
measured |
| Hartley 2004 |
narrative review, 31 studies |
structured abstracts are judged more informative and are typically longer; see the academic papers review |
supported with caveat |
| executive summary |
none found |
no controlled study of an executive summary was located (E3) |
open |
| conclusion first |
none found |
two experiments on bad-news letters, then e-mails (Jansen and Janssen 2010, full text; prints in 2011), report that readers perceive explanation-first as more comprehensible; perceived, not tested, comprehension (tested comprehension did not differ: 72% against 71% correct, Experiment 2 only); no test on reports (E5) |
open |
| length |
none found |
school rules treat excess length as a weakness; no study links length to outcome (F9) |
open |
6. Practice: French schools (R2, R3)
| TSE 2017 |
school grid, example of practice only (9.5) |
M1 internship report scored on 20: form 7 (instructions 1, title not just "rapport de stage" 1, layout and bibliography 2, graphics 2, writing 1), substance 13 (company 1, objectives 2, analyses 3, methods 3, comments on results 3, introduction and conclusion 1); asks 2-5 pages each on company, missions and conclusion with distance |
convention, untested |
| EOST 2026 |
school guide, example of practice only (9) |
the internship report is judged as an engineer's mémoire; minimum 50 pages with introduction, methods, results, conclusion, bibliography; company presentation and a cost annex are additions; a résumé in French and in English with at most five keywords; the supervisor must authorise the defence; jury of three; 45 minutes (30 + 15) |
convention, untested |
| INSPE 2021 |
university grid (12) |
M1 teaching mémoire: written part 15 of 20 beside 5 for conduct; substance asks a literature review that leads to the problématique and a method that tests the hypotheses; form asks syntax, spelling, plan, layout |
convention, untested |
| UGA faculty notes |
not retrieved |
a search summary reports a 250-word abstract and keywords in one faculty and a 35-page cap in the teaching mémoire guide; not verified |
open |
- No Grenoble M1 AI report guide was found (G4, G7).
- No CTI text fixing a report format was found (G6); formats are set by each school.
Second pass (2026-10-29): examiners, and summary first
Synthesis for the two questions the first pass left thin; its strand numbers continue review.md (7 and 8). Its strands are graded: measured, supported with caveat, convention (untested), not supported. A number in a strand is marked "(secondary)" when it was read in an abstract, a record, a search summary, or in another paper's report of it. Held elsewhere and linked, not reviewed again:
- Britt 2003 (claim first, one-sentence arguments) and Garner 2013: sister review reader-ergonomics-presentations.
- Jansen 2010 details, Nielsen 2006, Nildam 1998, Sternadori 2008, Hartley 2004: review of reading order in the shared repository.
- Pinelli 1982, Pinelli 1984, Sheffield 2017, NISO 2010: first-pass review.md.
P2a. Examiners and graders
7. Existing review of how examiners work
| Golding 2014 |
qualitative systematic review: 30 journal articles (ERIC and Google Scholar plus snowballing), 3504 examiner reports on about 1324 theses, 426 quality rankings, 698 examiners; English language, peer reviewed, master's, professional doctorates and PhDs; 11 practices; it does not mention PRISMA and gives no effect sizes (Golding 2014, full text, pp. 565 to 567; the totals are sums over the 30 articles) |
supported with caveat |
| Golding 2014 |
the written thesis carries the judgement: most examiners have decided before the oral, which confirms it (Golding 2014, full text, p. 565; it rests on one of the 30 articles and on three studies cited but not among the 30) |
supported with caveat |
8. What examiners look at first, and when they decide
| Golding 2014 |
examiners form an initial judgement within the first one or two chapters, sometimes while skimming the abstract, contents, introduction and conclusion (Golding 2014, full text, p. 568; no primary study cited); the "treat" and "endurance test" phrases are Carter 2008, p. 367 (23 Auckland faculty in informal panels), whose moment is the first few minutes of a first flick through the thesis |
supported with caveat |
| Golding 2014 |
the first impression has inertia: after a good start examiners read relaxed and overlook faults, after a poor start they read critically (Golding 2014, full text, p. 568; no primary study cited) |
supported with caveat |
| Golding 2014 |
Golding gives no ordered reading sequence and reports the skim of abstract, contents, introduction and conclusion as "sometimes" (full text, p. 568), but a reading order is given by three of the 30 studies: Carter 2008 (a first flick through the contents list, first and last chapters, conclusion and abstract), Mullins 2002 and Kiley 2004 (rows below) |
supported with caveat |
| Golding 2014 |
examiners need 2 to 4 full days, spread across 2 to 3 weeks (Golding 2014, full text, p. 568; a blend of Carter 2008, p. 369, two full days for several examiners with an extreme of two weeks, and Mullins 2002, p. 376, three or four days over two to three weeks), read in the evening when tired and in chunks; Golding, p. 569, advises signposts and summaries because Chapter 1 may be forgotten by Chapter 5, with no source cited |
supported with caveat |
| Mullins 2002 |
30 experienced examiners at five universities, interviewed on criteria, expected performance and critical judgement points; most open with the abstract, introduction, conclusion and references, then read in full; a first impression usually forms by the end of the second or third chapter, often at the literature review, and examiners decide early whether the thesis is likely to be hard work; it takes three or four days full time over two to three weeks |
supported with caveat (full text) |
| Kiley 2004 |
inexperienced examiners follow a similar reading process to experienced ones (abstract, introduction, conclusion, references first) but differ in their use of criteria, in the bounds of good and poor work and in reference to their own student experience; Golding 2014 reports practice as broadly consistent (secondary) |
supported with caveat (full text; Golding 2014's reading secondary) |
9. What criteria examiners apply
| Golding 2014 |
examiners favour a coherent argument, critical engagement with the literature, a justified approach carried out as announced, interpreted findings, and a publishable contribution |
supported with caveat |
| Golding 2014 |
presentation errors annoy and distract examiners, who lose confidence and read more critically, hunting for faults (Golding 2014, full text, pp. 568 to 569; one quotation from Johnston 1997, p. 339, the rest uncited) |
supported with caveat |
| Golding 2014 |
consistent recommendations for 96% of 804 theses (Golding 2014, full text, p. 567; the 96% is derived from the 4% that Holbrook 2008 prints, next row); fewer than 1% of examiners recommend a fail (p. 568, citing Lovat, Holbrook and Bourke 2008, p. 70) |
supported with caveat |
| Holbrook 2008 |
804 theses and 2121 examiner reports from eight Australian universities, 2003 to 2004: only 33 of the 804 theses (4%) had one or more discrepant reports, and 37 of 2121 reports (under 2%); recommendations were compared with the university committee's decision and with each other, a discrepancy being lenient (accept, minor correction) against severe (revise and resubmit, fail); the content of discrepant and agreeing reports was mostly similar (Table 3, p. 42, no percentage) |
supported with caveat (full text) |
| Holbrook 2007 |
501 candidates, 1310 reports, five Australian universities: about one tenth of the text of a report is on the literature; four indicators (working understanding, critical appraisal, connection to findings, disciplinary perspective); a barely passable thesis may show less, but critical appraisal is pressed |
supported with caveat (descriptive content analysis of reports; full text) |
| Bourke 2013 |
353 PhD and 74 research master's examiners rated 12 indicators in five areas (contribution, literature review, approach and methodology, analyses and results, presentation); the order of importance was essentially the same at both levels; the 12 indicators were strongly related to the holistic quality judgement, mainly for the PhD. It is a survey, not an editorial |
measured, descriptive (full text for the indicators and their order; abstract read for the link to the holistic judgement) |
| Chetcuti 2022 |
examiner reports at one university (four faculties, 2017-2018): explicit regulation criteria plus implicit personal ones; they look for a contribution to knowledge, critical and analytical work, and no typographical or grammatical errors; expectations differ by faculty |
supported with caveat (abstract read) |
| Sharmini 2015 |
online survey at one New Zealand university (896 supervisors invited, 62 complete responses, all with experience of examining publication-based theses); 86% of the respondents rated top-ranked and international peer-reviewed journals as significantly or very significantly influential (the base of that percentage is not printed) |
supported with caveat (full text) |
| Hodgson 2020 |
synthesises the literature on examiner expectations into a framework and a tool for supervision; its criteria were not reached |
convention, untested (abstract only) |
10. How reliable the grading is
| Williams 2018 |
internal and external markers of 324 master's theses in one psychology department (MA Psychology, MSc Psychology, MSc Applied Psychology) at the University of Canterbury correlated 0.39 on the grade-points they recommended, close to the figure for journal reviewers |
measured (full text) |
| Bodnar 2018 |
8 capstone design reports, 2 trained raters, 5 dimensions (context, methodologies, results, sources, syntax): overall ICC 0.850; syntax lowest at 0.567; in the comparison rubric the genre dimension had ICC 0.101 |
measured (small) |
| Brown 2023 |
first validation round: 6 instructors, 4 reports, 15 constructs: abstract 0.950, results 0.839, discussion 0.861, conclusions 0.816, references 0.902 rated excellent; technical writing 0.194, format 0.058, figure and table format 0.138 rated poor; total score 0.950 |
measured (small) |
| Reddy 2010 |
review of empirical rubric studies: of three studies on achievement, two found higher performance with rubrics and one (N = 16) none, which the authors call inconclusive; three validity studies found the clarity and appropriateness of the language of rubrics a central concern; the three studies that report inter-rater reliability tend to show a relatively common interpretation of student work |
supported with caveat (review; full text) |
- In Brown's first round and in Bodnar, graders agree on the content sections and on the total, and disagree most on writing and format constructs. Brown's second round (revised rubric, four graders) reversed this: ICC 0.00 (poor) for abstract, results and conclusions, while the format constructs became excellent; the total stayed 0.94. Samples of 4 and 8 reports are small.
11. How schools weight the report (practice)
| UT 2024 |
master's internship at Twente: report 60%, performance at the host 20%, supervisor's observations 20%; the report is scored on nine equal elements: context, tasks and roles, learning objectives, reflection on reaching them, research problem or assignment and results, contribution to future ambitions, contribution to the organisation, logical structure with references, correct academic language |
convention, untested (12 of 20) |
| UT 2024 |
by arithmetic on that form, context is one ninth of the report (about 7% of the final mark), and structure plus language together two ninths (about 13%) |
convention, untested |
| ULB 2025 |
30% of the mark from the host supervisor and 70% from the jury on report and defence; a report under 50% of the points is redone; no report structure is stated |
convention, untested (10.5 of 20) |
| UGA 2026 |
M1 STAPS stage unit: report 75% and defence 25%, also in the resit; a reflective analysis is obligatory; eight weeks full time. It is not the M1 AI unit |
convention, untested (10 of 20) |
| BTS 2010 |
example only (9 of 20): report coefficient 0.5 against 2 for the oral, so the report coefficient is one sixth of the total; the scientific-content criteria are marked as evaluated on the report and the oral, so the report's real share is not isolated |
convention, untested (example) |
- No Grenoble M1 AI report guide was found (G16).
- The report weighs 60% to 75% of a mark in the three university sources (UT 60%, ULB at most 70%, UGA 75%); BTS 2010, an example only, gives one sixth (17%). The shares measure different things (internship mark, unit mark, exam); none rests on a measure.
- Module guides at Nanterre, Caen, Strasbourg, Leiden and Wageningen were seen in search summaries only and are not used.
P2b. Summary first, conclusion first
12. Existing reviews
| E7, E8, E9, E12, E13 |
no review, meta-analysis or MLR of summary-first or conclusion-first for reports; the reviews found concern text-structure instruction, not the order of a finished text |
open |
| Koetsenruijter 2022 |
the nearest design to an executive summary test is a pilot with a summary box on one section (below) |
measured, weak |
13. Controlled tests
| Jansen 2010 |
two experiments on Dutch bad-news letters and e-mails: readers rated the explanation-first version more comprehensible and agreeable and its writer more competent and empathic, and were more inclined to comply (full text; Experiment 1 N = 450, Experiment 2 N = 574; the paper prints in 2011). Tested comprehension did not differ: 72% against 71% correct, Experiment 2 only (Table 5, p. 56). It holds for bad news; it is not a test of a summary or of reports |
measured (ratings) |
| Jansen 2013 |
e-mail against voice mail, direct against indirect structure: the indirect order (explanation, then bad news) was valued more, but only in e-mails |
measured (ratings; abstract read) |
| Spyridakis 2000 |
Japanese readers recalled more from expository text with its thesis at the end than at the start; preference unclear |
measured (abstract read) |
| Spyridakis 2002 |
American readers performed equally with either order; Japanese readers again recalled more from the thesis-last text |
measured (abstract read) |
| Davey 1990 |
107 middle-school students: main-idea comprehension was best with the topic sentence first, then last, then absent; a field-dependent subgroup did better with it last, on the best-title task only |
measured (abstract read; children) |
| Bridge 1984 |
college readers, passages with or without an initial topic sentence: better title generation and recognition with it in Experiment 1; no difference in Experiment 2, where generating the main idea helped memory |
measured (abstract read) |
| Suchan 1989 |
262 naval officers: a "high-impact" memo took 17% to 23% less time to read, with better recall and less rereading, than a bureaucratic memo of the same content (478 and 482 words). The style bundles nine changes, the bottom line first among them, and the authors measured their cumulative effect, so order is not isolated; comprehension was recall only |
measured (full text; bundled styles, recall only) |
| Silyn-Roberts 2002 |
744 engineers in training courses chose between a text ending on the main item and a version with a document summary first and section summaries: all chose the second; all read the summary, then the section summaries, then the detail. A stated preference with no test of comprehension |
supported with caveat (abstract read) |
| Koetsenruijter 2022 |
nonrandomized pilot, 35 advanced medical students reading a quantitative report; a summary box on the methods drew attention but did not change time spent on the methods, and the arm with the box rated the section less helpful (4.1 control against 2.9 intervention on 1-10 as printed; 22 and 13 students; the gap is not significant, P = .09). A summary of one section, not an executive summary |
measured, weak (abstract read) |
| Kieras 1978 |
seven-sentence artificial passages: putting the main-event sentence first raised the share of readers choosing it as the best title from about .55 to .67 (35 volunteers); content mattered more than position; bad order lengthened reading and lowered recall, more under memory load (Experiments 2 and 3) |
supported with caveat (full text) |
| Kieras 1980 |
a statement of the theme resembled the intended theme sentence more when that sentence opened the passage (mean similarity 2.48 against 2.11 on a 0 to 5 scale, 40 students), and in two-topic passages the topic mentioned first was chosen more often (96 students); the effect is modest, with edited and artificial texts, and none on reading time |
supported with caveat (full text) |
- Linked, not reviewed again, because the sister review holds them. Britt and Larson 2003 (full text) found claim-first arguments of one sentence read about 0.7 to 1.1 s faster, only when the claim was clearly marked. Claims were recalled better than reasons in either order. Reason-first arguments tended to be recalled in claim-first order. Garner 2013 found sentence headlines beat topic phrases on essays; that is not verified here.
- The direct tests of the news inverted pyramid (Nildam 1998, Sternadori 2008) are in the shared review and disagree (not verified here).
14. Practice
| Army 2020 |
AR 25-50, effective 2020-11-10: effective writing is understood "in a single rapid reading"; two requirements are the main point at the start (bottom line up front) and the active voice; the memo guidance cites no study |
convention, untested (14 of 20) |
| Pyramid Principle (G21) |
answer-first for consulting documents; every page found is a practitioner guide and none cites a test |
convention, untested |
| Executive summary (G18) |
the claim that decision makers read it first rests on guides; the only support is self-report (Pinelli 1982, first-pass review) and a stated preference (Silyn-Roberts 2002) |
convention, untested |
15. What the pool does not hold
| a controlled test of an executive summary on comprehension, time or decision |
none located (E7, G18) |
open |
| a controlled test of conclusion first against conclusion last in a report |
none located (F16, F18) |
open |
| a test of putting the thesis abstract or the contents first for examiners |
none located; examiners are studied by interview and report analysis only |
open |
| a link between report length and outcome |
none located |
open |