Against

This file gathers the evidence against the claims of this review, and the limits of its sources.

1. The survey that backs summary-first

  • Pinelli 1982 is a 1982 survey of NASA Langley engineers and scientists plus members of three societies; it asks what readers say they read, not what helps them.
  • Its respondents are mostly researchers (about 70%) and do not include students, examiners or French readers.
  • Pinelli 1984, read in full on 2026-10-29, splits the answers by role: 205 technical managers and 633 non-managers. Their preferences were very similar on every format question, so the role makes little difference among these readers. Its counts are close to Pinelli 1982's, and it is probably the same survey split by role; the article does not say so, so this is an inference.

2. Interviews are opinion

  • Sheffield 2017 is seven people saying how they read; one reads "backwards", which shows a habit and not that the structure helps.
  • The same page describes a client who prefers detail to a summary, so summary-first is not universal.

3. Headings are partial

  • Lorch 1996 found that, in free recall, headings help recall of unfamiliar topics only; an examiner or a host supervisor who knows the field may gain nothing.
  • No test of informative against label headings was located, so the rule "headings state the claim" is a convention.

4. The standard against overview-first

  • NISO 2010 keeps findings out of the introduction and puts them in the summary and conclusions; a report that follows it and also puts the answer in the introduction would break it.
  • The standard says there is no single format, so it is also evidence against one template.

5. Genre evidence is English, undergraduate and taught

  • Gardner 2013 and Nesi 2012 cover assessed writing in England; there is no research thesis in the corpus and no French text.
  • Paltridge 2002 (30 theses, one Australian university) and Bunton 2005 (45 theses, one Hong Kong university) were read in full text.

6. Grids are one teacher's rule

  • TSE 2017 and EOST 2026 score below 10 of 20 and are examples of practice only.
  • INSPE 2021 is a teacher-training grid and may not carry over to engineering or AI.
  • Weights of 7 and 13 out of 20 are choices; no study ties them to quality.

7. Conclusion-first may cost something

  • Two experiments on bad-news letters, then e-mails (Jansen and Janssen 2010, full text; prints in 2011), found explanation-first perceived as more comprehensible (perceived, not tested). For a grader receiving a failed result, the effect is untested.
  • A cross-cultural study of product texts found no difference in comprehension between inductive and deductive order (not verified: the study was not located again and is not named).

8. Reader preference is not learning

  • A stated preference for summary-first or for figures is not evidence that readers understand more; no study of this was located for reports.

9. Method limits

The protocol lists the limits of the search itself: closed pages, no hit counts, the stopping rules. They are limits of how the review was made, not evidence against the shape.