| name | imrad-structure |
| description | Use when drafting, restructuring, or reviewing an empirical scientific paper. Covers the IMRaD format (Introduction, Methods, Results, Discussion): what each section contains, the rhetorical move set, common failure modes, and section-by-section editing checks. |
IMRaD structure
IMRaD = Introduction, Methods, Results, Discussion. The dominant structure for empirical scientific papers (life sciences, social sciences, much of CS/engineering). Theory/review papers use different conventions.
The hourglass shape
Introduction: broad ─────────┐
▼
Methods: narrow
Results: narrow
▲
Discussion: ─────────── broad ┘
Open broad, narrow to your specific question, do the narrow work, then widen out again to implications.
Process
Drafting discipline that separates writing from editing, and the intro from everything else:
Before drafting: the brainstorm contract
Answer eight questions, in this order, and write the answers into a short project_context.md. That file is the binding contract for the paper; later scope disputes get resolved against it. The phase order (problem → contribution → evaluation → positioning → narrative) matters: deciding the problem and evaluation before the positioning prevents solution-first bias.
- What is the problem, and why now?
- Who has this problem, and what do they do about it today?
- What is the contribution, in one sentence?
- What evaluation would convince a skeptic the contribution is real?
- What result would falsify the claim?
- What is the closest related work?
- What is the delta from it: a difference in capability, not in mechanism?
- What is the narrative spine: stakes → insight → evidence, in three beats?
Drafting order
Not front-to-back:
- Draft-0 Introduction: fast, imperfect; it sets the guardrails.
- Evaluation: write toward the evidence you actually have.
- Design / Methods
- Background
- Related Work
- Final Introduction: rewritten from scratch against the actual evidence. The intro is written twice, deliberately.
- Abstract: last, always.
Topic sentences first
Before writing prose under any section, write all its topic sentences and read them as a sequence. If the topic sentences alone don't make the argument, prose underneath won't save them.
Move sequences
Introduction: Stakes → Problem/Gap → Key Insight → Approach → Contributions → Results preview.
Evaluation: Setup anchoring → Head-to-head → Deep dive → Takeaway synthesis → Ablations → Robustness. The takeaway synthesis (each results subsection ending with a paragraph stating what the reader should now conclude) is the pattern present in accepted papers and absent in rejected drafts. Never end a subsection on a number.
Compression is a separate pass
Target cutting 30–50% from a first full draft, as an explicit final stage, never while drafting. Compressing as you go kills momentum and protects bad paragraphs: you won't cut what you just polished.
Section-by-section
Introduction (≈10–15% of body)
Job: motivate why this paper exists and what it contributes.
Standard four-move structure (the CARS model, Create A Research Space):
- Establish territory: cite recent work showing the area matters.
- Establish niche: identify a specific gap, contradiction, or unanswered question.
- Occupy the niche: state what this paper does to address that gap.
- Outline: 1–2 sentences previewing the paper's structure.
Common failures:
- "Background dump": pages of literature with no gap identified.
- Gap is invented or trivial ("no one has studied X in our specific setting").
- Contribution claimed but never delivered.
Methods (≈15–25%)
Job: enough detail for a competent reader to reproduce the study.
Order: participants/materials → procedure → measures → analysis plan.
Conventions vary by field:
- Past tense ("We sampled..."), not present.
- Passive vs active is field-dependent (biomed often passive; CS often active).
- Pre-registered hypotheses go here, not in Results.
Common failures:
- Hiding decisions ("after preprocessing": what preprocessing?).
- Reporting results inside Methods.
- Vague sample sizes or unjustified exclusions.
Results (≈25–35%)
Job: report findings. Facts only, no interpretation.
Order: descriptive stats → primary analyses (in hypothesis order) → secondary/exploratory.
Conventions:
- Every figure/table is referenced in the prose ("Figure 2 shows..."), and every result is briefly stated even if a figure shows it.
- Effect sizes accompany p-values where applicable.
- Negative results are reported as clearly as positive ones.
Common failures:
- Interpreting results in this section ("This suggests..." belongs in Discussion).
- Figures the prose never references.
- p-values without effect sizes or CIs.
Discussion (≈20–30%)
Job: interpret what the results mean.
Standard moves:
- Restate principal finding in one sentence (no new caveats yet).
- Compare to prior literature: what's consistent, what's surprising.
- Mechanism / interpretation: why might this finding arise?
- Limitations: honest, specific (not boilerplate).
- Implications: for theory, for practice, for future work.
- Conclusion: what should the reader walk away believing?
Common failures:
- "Limitations" section that lists generic caveats no one believes you take seriously.
- Repeating Results verbatim.
- Wild extrapolation beyond what was measured.
Quick editing checks
When reviewing or revising:
Variants and adjacent structures
- Theoretical / review papers: use thematic structure instead.
- CS systems papers: often Intro → Background → Design → Implementation → Evaluation → Related Work → Conclusion. Evaluation maps to Results+Discussion; Related Work may sit at end.
- Medical journals: often require structured abstracts; sometimes Methods are split into "Study design" and "Statistical analysis."
Related: [[abstract-writing]], [[citation-formatting]].
