Research Foundations › Module 6: Planning a project
Writing a one-page proposal
Lesson 3 of 3 · 2 min
A one-page proposal makes your plan clear to you and to anyone who might help: a mentor, a lab or a funding body. Writing it often shows the gaps in your thinking before you lose weeks to them.
The parts
| Part | What to write | Length |
|---|---|---|
| Title | Specific, not clever | One line |
| Question | Your research question, and hypothesis if you have one | 1–2 sentences |
| Why it matters | Who cares about the answer, and why | 2–3 sentences |
| What is known | The 3–5 most relevant papers, and the gap you will fill | 3–4 sentences |
| Method | What you will do, with what data, and how you will measure the result | 4–6 sentences |
| Timeline | Main stages with weeks | A short list |
| Risks | What could go wrong, and your backup plan | 2–3 sentences |
| Ethics and data | Consent, privacy, licences | 1–2 sentences |
- Title
- What to write: Specific, not clever
- Length: One line
- Question
- What to write: Your research question, and hypothesis if you have one
- Length: 1–2 sentences
- Why it matters
- What to write: Who cares about the answer, and why
- Length: 2–3 sentences
- What is known
- What to write: The 3–5 most relevant papers, and the gap you will fill
- Length: 3–4 sentences
- Method
- What to write: What you will do, with what data, and how you will measure the result
- Length: 4–6 sentences
- Timeline
- What to write: Main stages with weeks
- Length: A short list
- Risks
- What to write: What could go wrong, and your backup plan
- Length: 2–3 sentences
- Ethics and data
- What to write: Consent, privacy, licences
- Length: 1–2 sentences
An example (shortened)
Title: Does low-light data augmentation close the day–night gap in leaf-disease classification?
Question: How much does accuracy drop on low-light photos for a classifier trained on a public leaf-disease dataset, and does adding synthetic low-light images to training reduce that drop more than adding the same number of normal images?
Method: Train a standard image classifier five times with different random seeds under three conditions: no extra images, extra normal images, extra synthetic low-light images. Measure accuracy on a separate test set of real low-light photos. Report the mean and spread.
Risks: Real low-light test photos may be hard to collect. Backup: use a held-out set of darkened photos, and state this limit clearly.
Tips
- Use plain words. If a smart first-year student cannot follow it, simplify it.
- Be honest about limits. Reviewers trust a proposal that names its risks.
- Ask for feedback before you start, not after.
RememberIf you can write your project on one page, you understand it. If you cannot, keep reading and thinking before you start running experiments.