Research proposal

A PhD research proposal does two jobs at once: it shows the applicant can think like a researcher (frame a real problem, propose a credible approach, see the obstacles), and it demonstrates specific fit with the target lab. A proposal that could be sent to any lab in the field fails the second job. The aim here is a proposal that sits precisely at the intersection of where the lab is heading and what the applicant can credibly do — concrete enough to be evaluated, modest enough to be feasible in a PhD, and clearly the applicant's own thinking.

Step 1 — Load the lab and the applicant

Read the professor's profile at knowledge-base/professors/<slug>.md, focusing on Research agenda & open problems and recent publications — the proposal must connect to the lab's trajectory, not its past. Read the applicant's profile/profile.md, profile/research-statement.md (if present), and profile/cv-master.md for the methods and results they can credibly build on.

If no professor profile exists, run professor-analyzer first; a proposal not anchored in the lab's real open problems is just a generic essay. Note any program constraints the user gives (page/word limit, required structure) and follow them.

Step 2 — Find the intersection

The core intellectual work: identify a problem that is (a) genuinely open in the lab's agenda, (b) something the applicant's background gives them a credible angle on, and (c) scoped to a PhD. Avoid both extremes — a problem the lab has already solved, or a moonshot no student could make progress on. Prefer a specific, well-motivated question over a broad theme. It's fine to propose one main thrust with one or two extensions rather than a sprawl.

Step 3 — Draft the proposal

Use this structure unless the program specifies otherwise. Keep it tight; most PhD proposals are 1–3 pages.

# <Working title>

## Motivation & problem
The specific problem, why it matters, and the gap in current work (cite real papers —
the lab's and the field's). 1–2 paragraphs.

## Background & related work
What's been done, including the lab's relevant work, and what's still open. Show you've
read the literature, including the professor's. Cite specific real papers.

## Proposed research
The core idea and approach. Concrete enough to picture the first experiments. Break into
1–3 aims/phases. State hypotheses and what success would look like.

## Methods & feasibility
How you'd actually do it, what you'd build on (your own skills/results + the lab's
resources), and why it's achievable in a PhD timeframe.

## Risks & alternatives
The main ways it could go wrong and your fallback directions. Including this signals
research maturity — proposals that pretend nothing can fail read as naive.

## Fit with the lab
Why *this* lab: the specific resources, datasets, methods, or expertise that make the
project viable here and not just anywhere. Tie to the professor's open problems by name.

## References
Real citations only.

Step 4 — Keep it honest and the applicant's own

Cite only real papers, and never claim preliminary results the applicant doesn't have — "preliminary experiments show X" is a fabrication unless X actually happened. Where the proposal would benefit from a specific the applicant must supply (a dataset they have, a result from their MS thesis, a concrete prior finding), mark it with [brackets] rather than inventing it. The proposal should read as the applicant's thinking sharpened, not as generic AI prose — match the voice in their research statement.

Step 5 — Save and hand off

Write the proposal to knowledge-base/applications/<id>/proposal.md. Summarize for the user the core idea, why it fits this lab, and which [brackets] they need to fill. For a submission-grade document (PDF/Word with formatting), hand off to the pdf or docx skill. If the proposal surfaces a strong angle, suggest it can seed the outreach-email hook or the application-materials SOP so the application tells one coherent story.

Guardrails

Follow shared/references/ethics.md. The proposal represents the applicant's research judgment to people deciding whether to mentor them for years — fabricated results or a misrepresented grasp of the literature are both dishonest and easily caught in interview. Aim for a strong, real, feasible draft the applicant can stand behind and defend.