Grant Writing skill for Codex

A fundable proposal makes a reviewer with 10 applications and little time say “this is important, this will work, and this team can do it” from the first page, and gives them no compliance reason to stop reading. The solicitation is the contract: every required section, limit and criterion is met, and the narrative is organized around how reviewers will score it.

Authorship and funder AI policies (check first)

The applicant must remain the author of the ideas, aims and science. Funders increasingly regulate AI use. NIH, for applications due on or after September 25, 2025, does not consider applications substantially developed by AI to be the applicant’s original ideas (NOT-OD-25-132), and the same notice caps PIs at six applications per calendar year. Other funders and institutions have their own rules.

Before doing anything substantive:

1. Ask which funder and program, and read its current AI-use policy (and the institution’s).
2. Default to the roles that are safest under strict policies: compliance checking, structure and outline feedback, clarity editing of the applicant’s own text, budget arithmetic, consistency checks, and mock review. Draft new scientific text only if the user confirms the funder permits it, and mark it clearly for the applicant to rewrite.
3. Never invent preliminary data, citations, letters of support, effort commitments or institutional resources.

Budgets, rates and certifications are reviewed by the institution’s sponsored programs office (or the nonprofit’s finance lead). Say so; Codex checks arithmetic and consistency, not allowability rulings.

Workflow

- [ ] 1. Parse the solicitation into a compliance matrix
- [ ] 2. Fix the core: problem, gap, central hypothesis or goal, aims
- [ ] 3. Write or review the first page (Specific Aims / Project Summary / LOI)
- [ ] 4. Align each narrative section to the scored criteria
- [ ] 5. Build the budget table and justification; reconcile with the narrative
- [ ] 6. Mock review against the funder's criteria
- [ ] 7. Final compliance pass; hand to the institution for routing

1. Compliance matrix

Extract every requirement from the funding announcement (NOFO, solicitation, RFP or call) and the funder’s general guide into a table. This is the most-reused artifact in the process.

| # | Requirement | Source (section/page) | Limit / format | Where addressed | Status |
|---|---|---|---|---|---|
| 1 | Specific Aims | NOFO Sec. IV.2 | 1 page | aims.docx | draft |
| 2 | Research Strategy | NOFO Sec. IV.2 | 12 pages (R01) | strategy.docx | outline |
| 3 | Data Management and Sharing Plan | NIH DMS policy | 2 pages recommended | dms.docx | missing |
| 4 | Letters of support | NOFO | none listed as required | - | n/a |
| 5 | Font/margins | Application guide | 11 pt+, 0.5 in margins | template | ok |

Include: page and character limits, fonts and margins, required and prohibited sections, eligibility, budget caps, cost-sharing rules, review criteria, special requirements (human subjects, data plans, mentoring plans, broader impacts), deadlines including the institution’s internal routing deadline (often 5-10 business days earlier).

2. The core

Write these in one or two sentences each before any long section. If they are unclear, no amount of polishing helps.

• Problem: what is wrong or unknown, and who it matters to.
• Gap: why current knowledge or practice cannot solve it.
• Central hypothesis or goal: testable (research) or measurable (program).
• Aims or objectives: 2-4, each independently valuable, not dependent on another aim succeeding first.
• Payoff: what will be known or changed at the end, and what that enables.

Aim test: each aim has a verb that implies an outcome (“Determine whether…”, “Test the hypothesis that…”, “Establish…”), not only activity (“Study”, “Explore”). If Aim 2 cannot start unless Aim 1 succeeds, add an alternative path or restructure.

3. The first page

NIH Specific Aims (1 page) is the template most other funders resemble:

Para 1 - Hook and problem: significance in 3-5 sentences, ending with the gap.
Para 2 - Long-term goal, objective of this application, central hypothesis and its basis
         (preliminary data in one line), rationale.
Para 3 - Aims: bold one-line aim, 2-3 sentences each on approach and expected outcome.
Para 4 - Expected outcomes and impact: what changes in the field when this succeeds.

NSF Project Summary: three labelled parts (Overview, Intellectual Merit, Broader Impacts) within one page, written in third person. Foundation LOIs (1-3 pages): need, approach, population served, outcomes, budget ask, organizational capacity.

4. Align to review criteria

Organize headings and topic sentences to mirror how reviewers score.

Funder	What reviewers score	Write so they can find
NIH research grants (simplified framework, due dates from Jan 25, 2025)	Factor 1 Importance of the Research (significance, innovation), Factor 2 Rigor and Feasibility (approach), Factor 3 Expertise and Resources (sufficient or not, no numeric score)	Explicit “Significance”, “Innovation”, “Approach” headings; rigor, power analysis, pitfalls and alternatives per aim
NSF	Intellectual Merit and Broader Impacts, each judged on the same five review elements	Literal headings “Intellectual Merit” and “Broader Impacts”; broader impacts with specific activities, audiences and measures
Foundations	Fit with mission, need, outcomes, evaluation, capacity, sustainability	Logic model: inputs, activities, outputs, outcomes; how success is measured

Approach section per aim: rationale, design, methods, expected results, potential pitfalls and alternative strategies, timeline. Reviewers punish missing pitfalls more than admitted risks.

5. Budget and justification

Build the budget as a table (spreadsheet or CSV) first, then write the justification line by line from it. Check:

• Personnel: effort as person-months (NIH) or percentage, salary base, fringe rate from the institution’s agreement; salary caps where the funder sets one.
• Equipment vs supplies: the capitalization threshold comes from the institution and, for US federal awards, 2 CFR 200 (raised to $10,000 per unit in the 2024 revision; older awards and some institutions differ). Confirm with sponsored programs.
• Travel, participant support, subawards, consultants, other direct costs, each tied to an activity in the narrative.
• Indirect (F&A) costs: negotiated rate and base (often modified total direct costs), or the funder’s cap. Do not assume; ask for the rate agreement.
• Year-over-year escalation consistent with the institution’s policy.

Justification pattern per line:

**Graduate research assistant (12 person-months/year, Years 1-3): $38,400 + fringe.**
Will run the mouse behavioral cohorts in Aims 1 and 2 (approx. 240 animals/year) and
maintain the analysis pipeline described in Approach 1.3.

Reconcile: every person, instrument, trip and subaward mentioned in the narrative has a budget line, and every budget line is justified by narrative activity. Recompute totals with a script or spreadsheet formula, not by eye.

6. Mock review

Score the draft as a tired, expert, skeptical reviewer using the funder’s actual criteria and scale (NIH 1-9 where 1 is best; NSF Excellent to Poor). For each criterion list strengths, weaknesses (labelled major or minor), and the fix. Common major weaknesses:

• Aims depend on each other; Aim 1 failure collapses the project.
• Hypothesis is not testable, or the methods do not test it.
• No preliminary data for the riskiest step.
• Approach lacks power analysis, controls, or rigor elements the funder requires.
• Scope too large for the budget and period.
• Broader impacts are generic (“we will publish and present at conferences”).

7. Final pass

- [ ] Every compliance-matrix row is "ok" with the section that satisfies it
- [ ] Page limits and formatting verified in the final PDF, not the word processor
- [ ] Terminology, aim numbering and names consistent across all documents
- [ ] Budget totals match forms; justification matches narrative
- [ ] Citations complete and formatted per funder; no hyperlinks where prohibited
- [ ] Biosketches, letters and data plans present and current

Submission is done by the institution’s authorized organizational representative or the applicant through the funder portal. Codex never submits; it hands over the package and the compliance matrix.

Gotchas

• Writing for a general audience on page one: reviewers are experts, but often in an adjacent field. Define the gap in their terms without textbook background.
• Recycling a previous application without updating to the current NOFO and review framework; criteria and page limits change.
• Generic broader impacts or evaluation plans copied between proposals.
• Ignoring the internal routing deadline and missing the real submission window.
• Resubmissions: the introduction must answer each prior critique specifically.

Verification

• The compliance matrix has no open rows; limits checked on the rendered PDF.
• A reader can state the problem, gap, hypothesis and aims after reading only page one.
• Budget arithmetic recomputed; every narrative resource has a line and every line a reason.
• The mock review lists no unresolved major weakness, or the user has accepted it knowingly.
• Report what Codex drafted versus edited, so the applicant can comply with disclosure rules.