# Academic papers: structure, critiques, templates

How a research article is ordered (IMRaD and its abstract), what the evidence says about that order, and what publisher templates enforce.

Steps: [frame](frame.md) · [protocol](protocol.md) · [sources](sources.md) · [review](review.md) · [state of the art](state-of-the-art.md) · [position](position.md) · [against](against.md) · [baseline](baseline.md) · [bibliography](../bibliography.md)

## Claims

| Claim | Grade | Where |
|---|---|---|
| IMRaD became the only pattern in four leading medical journals by the 1980s | measured | review.md 1 |
| IMRaD is not the default in many fields outside biomedicine | measured | review.md 1 |
| Journal endorsement of CONSORT raises reporting completeness, modestly | measured | review.md 2 |
| Endorsement of other guidelines showed no significant gain on nearly every item | measured | review.md 0 |
| Introductions follow three moves (CARS) | supported with caveat | review.md 3 |
| Structured abstracts hold more information than unstructured ones | measured | review.md 4 |
| Structured abstracts are easier to read and to search | supported with caveat | against.md 3 |
| Visual abstracts raise attention on social media | measured | review.md 4 |
| Visual abstracts or highlights improve understanding | convention, untested | position.md |
| Researchers read selectively, about 30 minutes per article | supported with caveat | review.md 5 |
| Experienced readers go to figures and results first | supported with caveat | review.md 5 |
| Readability of abstracts has fallen since 1881 | measured | review.md 6 |
| The paper misrepresents the process of discovery | supported with caveat | against.md 1 |
| Results-blind review cuts the positive-result rate from 96% to 44% | measured | state-of-the-art.md |
| Few shared notebooks reproduce their results (4.03%) | measured | against.md 5 |
| Templates fix layout, metadata and references, not the argument | supported with caveat | review.md 8 |
| IMRaD helps a reader find a result faster than another order | convention, untested | position.md |
| Reviewers confirmed 71.5% of 2019 NeurIPS code claims made on the checklist | measured | review.md 9 |
| Artifacts do not raise citations in a decade of evaluations | measured | review.md 10 |
| A declared data-sharing statement is a weak signal of actual sharing | measured | review.md 13 |
| An infographic and a plain summary give the same knowledge | measured | review.md 12 |
| At least 13.5% of 2024 biomedical abstracts were processed with a language model | measured | review.md 16 |
| ML documentation rose before the checklists were introduced | supported with caveat | review.md 9 |
| An automated reminder raised the share of articles with RRIDs from 33% to 65% at one publisher | supported with caveat | review.md 12 |
| Empirical evidence on criteria for plain-language summaries is scarce | supported with caveat | review.md 0 |
| 34.6% of 162 STM publishers had a public AI policy for authors in late 2023 | measured, descriptive | review.md 15 |
| No study located of STAR Methods, of release blog posts, or of authors' declared against estimated AI use | open | state-of-the-art.md |