brycewang-stanford NeurIPS Review Process
Use this skill to reason about what reviewers and ACs are likely to do with a submission. Do not use it to infer acceptance odds from folklore; use it to identify decision-relevant weaknesses.
Process model
- Reviews happen in OpenReview under double blind.
- Reviewers are asked to evaluate quality, clarity, significance, and the submission's declared contribution type.
- ACs coordinate reviewers, handle conflicts and quality issues, write or supervise meta-reviews, and make recommendations.
- Ethics concerns can be flagged and routed to ethics reviewers; severe cases can affect decisions.
- Authors who are also reviewers or ACs face reciprocal-reviewing obligations; gross negligence can create sanctions affecting their own submissions.
Reviewer mental model
Reviewers are overloaded cross-area specialists. They need to see:
- what the contribution is;
- why it is new relative to close NeurIPS/ICML/ICLR/ACL/CVPR/KDD-style work;
- whether the evidence supports the exact claim;
- whether limitations, safety, data, and reproducibility are handled responsibly;
- whether the paper can be trusted as a scientific artifact.
Diagnosis workflow
- Map each likely review concern to quality, clarity, significance, ethics, reproducibility, or fit.
- Predict which concerns the AC can use in a meta-review.
- Decide whether the problem is fixable by clarification, extra analysis, better framing, or re-routing.
- For author response, prioritize issues that change decision logic rather than issues that only improve tone.
Output format
[Likely review split] enthusiastic / borderline / skeptical
[AC-level issue] <one issue most likely to drive the meta-review>
[Ethics/reproducibility flags] <none or list>
[Response strategy] clarify / concede / add small result / reroute