Grant Planning

Grant Planning

Plan the scientific argument before drafting application answers. Produce a decision document that explains why the problem matters, why this team can make unusual progress, and why the proposed scope is executable.

Load supporting skills selectively

• Read ../grant-writing/SKILL.md when the funder and call are known. Follow

its sponsor-type guidance while planning fit and impact.

• Read the solicitation, scoring criteria, application questions, and supplied

funder notes before generating stories.

• Before returning a polished planning document, read

../general-writing/SKILL.md and ../general-writing/references/eval.md.

• Use [planning-template.md](references/planning-template.md) for a substantial

planning document or a comparison of multiple stories.

Keep planning distinct from drafting

• Decide the problem, thesis, evidence, team logic, feasibility, and scope

before writing application prose.

• Use provisional language for unresolved design choices.
• Preserve competing stories when they represent genuinely different

scientific centers of gravity.

• Do not polish a weak story into apparent coherence. Surface the missing

premise, evidence, or team connection.

• Hand the selected plan to grant-writing only after its logic survives

comparison and risk review.

Build the evidence inventory

Read the relevant project files and source material. Record:

1. The funder's stated vision, scope, exclusions, review criteria, and budget.
2. The concrete problem or risk the call is trying to change.
3. The applicants' relevant prior results, methods, code, datasets, testbeds,

collaborators, and preliminary evidence.

1. The scientific assets each applicant contributes.
2. The intervention ideas that follow from those assets.
3. The available personnel, compute, time, access, and budget.
4. The facts that remain unverified and the decisions the applicants must make.

Distinguish source-backed facts, interpretations, hypotheses, and open questions. Do not promote an analogy into a result.

Download and contain arXiv sources

For literature-dependent planning, download the LaTeX source for each relevant arXiv paper when it is available. Prefer the source archive over relying only on the rendered PDF because the .tex, bibliography, and appendix files are easier to search and verify.

1. From the grant repository root, create the dedicated

Background/papers/ directory.

1. Extract each source archive into its own

-/ subdirectory. Record the arXiv ID, version, source URL, and local path in the evidence inventory.

1. Resolve the current version through

https://export.arxiv.org/api/query?id_list=, then download the reproducible, versioned archive from https://export.arxiv.org/e-print/v.

1. Reject archive entries with absolute paths, .. traversal, or links that

could write outside the paper's directory.

1. Keep searchable source files such as .tex, .bib, .bbl, .sty, and

source metadata visible to Git.

1. Add only paper-local binary media to the grant repository's .gitignore:

``gitignore # Binary media bundled with downloaded arXiv sources /Background/papers/**/*.pdf /Background/papers/**/*.jpeg /Background/papers/**/*.jpg /Background/papers/**/*.png ``

Do not ignore the whole Background/papers/ directory, and do not add repository-wide patterns such as *.pdf or *.jpg. If a source archive contains another large binary-media format, add a similarly scoped pattern under /Background/papers/ only. Never treat the downloaded source as proof by itself: inspect the relevant source passage and distinguish the paper's result from the proposal's interpretation.

Build the motivation spine

Answer these questions before naming a method:

1. What concrete failure, risk, or scientific blockage exists?
2. Who or what is exposed to the consequence?
3. Through what mechanism does the failure arise?
4. Why do current methods, benchmarks, or institutions miss it?
5. Why is the problem important and tractable now?
6. What changes if the project succeeds?
7. What minimum result would still change scientific understanding?
8. What would a clean negative result teach?

Prefer a causal account to a list of concerns. The motivation should make the research question feel necessary before the reader encounters the proposed intervention.

Combine prior work into a research advantage

For each relevant prior work, extract:

• the established result;
• the mechanism or concept it contributes;
• the concrete asset it supplies;
• the limitation that creates room for the proposed work;
• the claim it supports in the new plan.

Combine prior works through a scientific dependency, not a bibliography list. Show how one applicant's work defines the target, evidence, or environment while another applicant's work supplies theory, measurement, intervention, or evaluation.

Apply the removal test: explain what becomes materially weaker if either applicant is removed. Replace generic claims of complementarity with distinct responsibilities and shared scientific questions.

Establish feasibility

• Identify the strongest empirical foothold: a prior result, codebase, dataset,

testbed, theorem, or pilot.

• Define the first decisive experiment or analysis.
• Match the number of aims and interventions to the budget and duration.
• Separate required components from optional extensions.
• Name the main technical and scope risks without softening them.
• Give each risk a mitigation, fallback, or useful boundary condition.
• Explain why the available team and resources can complete the work.
• Treat milestones as scientific decision points, not activity lists.

Generate substantive story options

Make each story a coherent answer to one core problem. Do not create variety by renaming the same approach.

For each story, specify:

1. Working title and one-sentence thesis.
2. Core problem, stakes, and why now.
3. Scientific question and central hypothesis.
4. Why current approaches fail.
5. How the applicants' prior work combines.
6. Why this team is unusually positioned to execute.
7. Focused research plan at the level of aims and decisive tests.
8. Feasibility evidence and required resources.
9. Best-case, minimum valuable, and informative negative outcomes.
10. Main risks, mitigations, funder fit, and scope boundary.

Spend more space on motivation, team synthesis, and feasibility than on implementation details during planning. Include technical detail only when it changes the scientific thesis or credibility of execution.

Compare and select

Compare stories against:

• importance and clarity of the core problem;
• direct fit to the call;
• strength of preliminary evidence;
• necessity of the applicant combination;
• novelty of the scientific question;
• feasibility at the requested tier;
• generalizability of the result;
• risk of being judged out of scope;
• ability to produce a valuable result if the main hypothesis fails.

State the strongest story, the strongest backup, and whether parts of one story should become an aim or intervention inside another. Do not combine stories when doing so produces a list of disconnected methods.

Prepare the drafting handoff

End with:

• the recommended scientific spine;
• the evidence and claims safe to use;
• the role of each applicant;
• the essential aims and outcomes;
• the facts and choices still requiring confirmation;
• a map from the plan to the application questions.

Then invoke grant-writing to draft the application in the funder's required format. Preserve the planning document as the source of truth for the argument.

Source & license

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

• Author: msimchowitz
• Source: msimchowitz/writing-skills
• License: MIT

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