Research Writing

Research & Academic Paper Writing

Guidance for conducting research effectively and writing clear, rigorous papers. Covers the full lifecycle: from choosing a problem through exploration and understanding, to distilling findings into a well-structured paper.

When this skill applies:

• Planning or executing a research project
• Writing or revising any section of a research paper
• Preparing conference submissions or journal articles
• Designing experiments or evaluating evidence quality
• Working on .tex, .Rmd, .qmd, or academic .md files
• Structuring a narrative around empirical findings

---

Quick Reference: The Research Process

Research follows four stages. Identify which stage you are in — each has a different north star and failure mode.

| Stage | North Star | Key Activity | Common Failure | |-------|-----------|--------------|----------------| | 1. Ideation | Find a good problem | Read literature, talk to mentors, identify gaps | Picking a problem that's too vague or already solved | | 2. Exploration | Gain information | Run exploratory experiments, follow curiosity, keep a highlights doc | Being too perfectionist; not casting a wide enough net | | 3. Understanding | Confirm/reject hypotheses | Design targeted experiments that distinguish between explanations | Insufficient skepticism; confirmation bias | | 4. Distillation | Compress into clear, rigorous claims | Write, refine evidence, communicate | Treating writing as an afterthought; not starting early enough |

If stuck, ask: "Which stage am I actually in?" Often you think you're in Stage 3 but you're still in Stage 2.

See references/research-process.md for the full framework including transition criteria and prioritization techniques.

---

Quick Reference: Paper Writing Process

Write papers using a compress-then-expand approach:

1. Compress — Identify 1-3 concrete claims your paper makes. If you can't state them in one sentence each, you haven't distilled enough.
2. Bullet-point narrative — Write the introduction as bullets: what's the context, what's the contribution, why should anyone care?
3. Bullet-point outline — Expand to full paper structure. For each section, list what it must communicate.
4. Results first — Collect key experimental results and make draft figures before writing prose.
5. First draft — Flesh bullets into prose. Use an LLM to get unstuck on individual paragraphs if needed.
6. Edit ruthlessly — Get feedback from others. Polish figures. Cut anything that doesn't serve the narrative.

See references/paper-writing-guide.md for detailed guidance on narrative crafting, claim types, and the iterative expansion process.

---

Paper Anatomy Cheatsheet

Abstract (6-sentence formula)

1. Establish the subfield/context (one sentence)
2. State the motivation or open question
3. State your key contribution
4. (Optional) Clarify key definitions or scope
5. Summarize the strongest evidence (one sentence per major claim)
6. Why it matters — broader implications or standard of evidence achieved

Include at least one concrete metric or result.

Introduction (6-paragraph structure)

1. Context and the key question — why does this matter?
2. Technical background — what's established, what tools exist?
3. Your key contribution, with nuance
4. The most critical evidence supporting it
5. (Optional) Additional claims or secondary contributions
6. Impact, implications, and a bullet list of specific contributions

Main Body

• Background: Define terminology and crucial techniques the reader needs
• Methods: Explain your approach and why it's appropriate
• Results: Specify experiments and results, with subsections per experiment type
• Alternative: If claims require different types of evidence, give each its own section rather than a monolithic methods-then-results block

Figures

• Ask: "What information should the reader take away from this figure?"
• Annotate key data points; use color, size, and shape to compress information
• Include clear axis titles, legends, and self-contained captions
• Avoid red/green as the sole distinction (colorblind accessibility)
• Combine the most important graphs into a single Figure 1
• Create explanatory diagrams for complex methods — high effort but high payoff

Discussion & Limitations

• Acknowledge limitations honestly — competent reviewers see through hype
• Discuss broader implications, future work, and reflections
• Conclusions sections are often redundant if the introduction is clear — skip if so

Related Work

• Explain how your work differs from and builds on prior work
• Place as the penultimate section unless it plays a crucial motivating role
• Contextualizing within the literature signals competence and builds trust

See references/paper-anatomy.md for section-by-section templates and detailed guidance.

---

Evidence Quality Checklist

Before finalizing any claim, verify:

• [ ] Distinguishes between hypotheses — Could an alternative explanation produce the same result?
• [ ] Statistically robust — For exploratory work, use p < 0.001 (not p < 0.05). Consider noise and sample size.
• [ ] Not cherry-picked — Track which analyses were planned vs. post-hoc. Report both.
• [ ] Ablation-tested — For complex methods, show which components actually matter.
• [ ] Diverse evidence — Multiple independent lines of evidence are far more robust than one strong result.
• [ ] Red-teamed — Actively try to break your own narrative. Ask: "If I'm wrong, where would I see it?"

Quality over quantity: one compelling, well-controlled experiment beats five mediocre ones.

See references/evidence-and-experiments.md for the full evidence quality framework.

---

Common Pitfalls

1. Overclaiming — State what the evidence actually supports, not what you wish it supported. Acknowledge limitations up front.
2. Unnecessary complexity — Simple techniques applied carefully often beat complex ones. Write in plain language; use jargon only when it adds precision.
3. Not starting the write-up early enough — Switch to distillation mode ~1 month before any deadline. Writing clarifies thinking — it's not just packaging.
4. Illusion of transparency — You know your work intimately; readers don't. What seems obvious to you will be confusing to them. Test this by having someone else read your draft.
5. Weak baselines — Compare against the strongest reasonable baseline, not a strawman. Reviewers will notice.
6. Post-hoc storytelling — Distinguish between what you predicted and what you discovered after looking at results. Both are valuable, but conflating them undermines trust.

---

Key Principles

• Inform, not persuade — Write to communicate truth, not to sell. The goal is contributing to shared knowledge.
• Compress ruthlessly — Readers take away at most a few sentences from your paper. Choose those sentences carefully.
• Red-team your own narrative — Assume you made a mistake somewhere. Where is it? What would disprove your claims?
• Writing is thinking — The distillation process itself deepens understanding. If you can't explain it clearly, you may not fully understand it yet.

---

Research Mindsets

Three mindsets that distinguish productive researchers:

1. Truth-seeking — Actively resist confirmation bias. At every stage, ask "What would change my mind?" and test it.
2. Prioritization — Time is the scarcest resource. Have a clear north star, write goals down, do weekly reviews. Be willing to drop promising-but-not-great directions.
3. Moving fast — Tight feedback loops are everything. Run the cheapest experiment that gives information first. Build good tooling. Audit where your time actually goes.

See references/research-mindsets.md for detailed frameworks on each mindset, including prioritization techniques and research taste development.

---

Additional Resources

| File | Contents | |------|----------| | references/research-process.md | The 4-stage research model, transition criteria, Steinhardt's decision framework | | references/research-mindsets.md | Truth-seeking, prioritization, moving fast, developing research taste | | references/paper-writing-guide.md | Narrative crafting, claim types, compress-then-expand process, novelty | | references/paper-anatomy.md | Section-by-section templates: abstract, intro, body, figures, discussion, related work | | references/evidence-and-experiments.md | Evidence quality, statistical rigor, red-teaming, ablations, baselines |

---

Adapted from Neel Nanda's research process sequence and ML paper writing guides (2025).

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: stchakwdev
• Source: stchakwdev/research-writing-skill
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.