paper-redteam

Paper Redteam

Purpose

Simulate a serious pre-submission review before real reviewers see the paper. This skill should be strict, evidence-grounded, and useful: it should expose decision-critical weaknesses and produce a fix plan, not just generic feedback.

When To Use

• Before submitting to ICLR, NeurIPS, CVPR, ICML, ECCV, AAAI, TMI, or arXiv.
• After major rewrites.
• Before rebuttal planning.
• When deciding whether a paper is ready or needs another evidence pass.

Inputs

Required:

• target venue and year;
• paper context;
• available draft sections;
• venue checklist;
• claim-evidence matrix or enough draft text to create one.

Recommended:

• novelty-scout report;
• related-work search results;
• experiment tables;
• limitation section;
• anonymization status.

Reviewer Roles

• R1 Technical: correctness, method clarity, assumptions, algorithms, theorem/proof details.
• R2 Novelty: related work, positioning, contribution strength, overlap risks.
• R3 Evidence: experiments, baselines, metrics, ablations, statistics, reproducibility.
• R4 Presentation: abstract, introduction, figures, writing clarity, organization, page pressure.
• AC Meta Review: decision-critical synthesis, venue fit, desk-reject risks, fix prioritization.

Procedure

1. Load venue rules. Check page/anonymity/template/checklist constraints first. Desk-reject risks should not be mixed with ordinary writing advice.

2. Run evidence audit. If no claim-evidence matrix exists, create a compact one for title, abstract, introduction, and results.

3. Run novelty/related-work check. Use novelty-scout or mark novelty evidence missing.

4. Review by role. Each reviewer should produce strengths, major weaknesses, minor weaknesses, questions, and required fixes.

5. Separate fatal, major, and minor issues. Fatal issues block submission. Major issues affect likely review outcome. Minor issues improve clarity.

6. Produce fix plan. Rank by decision impact and effort.