Proposal Review

---
name: proposal-review
description: Review an AI/ML, computational-biology, or bioscience research proposal for a funding decision. Use when scoring a grant or funding application against sponsor criteria, feasibility, budget, and risk.
---

# Proposal Review

Produce a decision-ready review of an AI/ML, computational biology, or bioscience proposal. Be fair, skeptical, and specific, and name missing information.

## Instructions

1. Read the proposal and identify the decision context if provided: sponsor goals, rubric, budget cap, timeline, and risk tolerance.
2. If critical information is missing, do not invent it. Flag the gap and turn it into a prioritized question for the PI.
3. Structure the review with these sections:
   - Executive summary
   - Heilmeier catechism
   - Technical merit
   - Data, compute, and experimental resources
   - Risk register
   - Team and execution capability
   - Ethics, safety, and compliance
   - Budget and schedule realism
   - Scorecard
   - Decision and funding conditions
   - Questions for the PI
4. Tailor the technical review to the proposal type:
   - AI/ML: baselines, ablations, leakage prevention, calibration, external validation, compute realism
   - Bio or wet lab: controls, replicates, statistical plan, assay feasibility, translational path
5. Include at least six risks covering technical, data or experimental, budget or timeline, and adoption or regulatory concerns when relevant.
6. If the sponsor supplies a rubric, use its categories, weights, and decision vocabulary. Otherwise use the default 1-to-5 scorecard below; do not mix sponsor and default weights.
7. Default weights: strategic fit and novelty 15%, technical rigor 25%, feasibility and resources 20%, team and execution 15%, risk, ethics, and compliance 15%, budget and schedule 10%.
8. Map the default weighted mean to `Strong Accept` (>=4.5), `Accept` (>=3.7), `Borderline` (>=2.8), or `Reject` (<2.8). A documented fatal flaw may override the numeric band.
9. Keep the review concrete and action-oriented. Reference proposal details when available and name fatal flaws plainly.
10. For a machine-checked scorecard, write the scores to JSON and run the helper:

    ```bash
    uv run --script "$HOME/.agents/skills/proposal-review/scripts/score_proposal.py" scorecard.json
    ```

    With the default rubric, `scores` must contain exactly these 1-to-5 keys: `strategic_fit_and_novelty`, `technical_rigor`, `feasibility_and_resources`, `team_and_execution`, `risk_ethics_and_compliance`, `budget_and_schedule`. For a sponsor rubric, add `"rubric": {"weights": {...}, "recommendations": [{"minimum": 4, "label": "Fund"}, ...]}`; weights must total 100 and the score keys must match them. Set `"fatal_flaw": true` (and optionally `fatal_flaw_recommendation`) to override the band. The helper rejects weights that do not total 100, category mismatches, and scores outside 1 to 5, and never mixes default bands into a sponsor rubric.

## Quick Reference

| Task | Action |
|------|--------|
| Summarize proposal | Describe aims, novelty, and bottom-line recommendation in <=150 words |
| Test strategic logic | Answer the Heilmeier catechism explicitly |
| Review feasibility | Check assumptions, methods, milestones, and resource realism |
| Review rigor | Assess controls, baselines, validation, statistics, and reproducibility |
| Review risk | Build a risk register with likelihood, impact, warning signs, and mitigations |
| Make a decision | Give a final recommendation plus concrete funding conditions or rejection reasons |
| Use a sponsor rubric | Preserve its categories, weights, thresholds, and recommendation labels |
| Validate a scorecard | `uv run --script "$HOME/.agents/skills/proposal-review/scripts/score_proposal.py" scorecard.json` |

## Input Requirements

- Proposal text or a linkable proposal excerpt
- Optional sponsor or program context
- Optional scoring rubric, budget cap, and timeline constraints

## Output

- A decision-ready structured proposal review
- A weighted scorecard with justified subscores
- A clear funding recommendation and conditions
- A prioritized list of questions that could change the decision

## Quality Gates

- [ ] Missing information is flagged instead of invented
- [ ] The review covers novelty, rigor, feasibility, risks, team, ethics, and budget
- [ ] At least six concrete risks are documented with mitigations
- [ ] The final recommendation is explicit and consistent with the evidence
- [ ] Scorecard weights total 100%, all rubric categories are scored, and the recommendation follows the selected rubric's bands

## Examples

### Example 1: Review a computational biology grant draft

```text
Review this proposal for a microbiome foundation-model project. Use a 1-5 scorecard,
identify fatal flaws if any, and list conditions for funding.
```

### Example 2: Review with sponsor constraints

```text
Review this translational bioscience proposal for a program with a 24-month timeline,
$1.5M budget cap, and high concern for regulatory risk.
```

## Troubleshooting

**Issue**: The proposal is missing a clear evaluation plan
**Solution**: Mark this as a major weakness, explain what convincing evidence would look like, and add PI questions about milestones and success metrics.

**Issue**: The budget or timeline is hard to judge
**Solution**: State the uncertainty, identify the likely critical path, and evaluate whether the claimed scope is credible under the stated constraints.

**Issue**: Ethics or compliance details are absent
**Solution**: Treat the omission as a potential blocker and ask targeted questions about subjects, privacy, biosafety, or regulatory readiness.

## Related Skills

- `/manuscript-review-council`: equivalent review for manuscripts
- `/scientific-writing`: draft or revise the proposal narrative
- `/bio-logic`: assess methodology and evidence rigor