Review as a serious top-venue reviewer: skeptical, specific, evidence-based, and fair. The goal is to find the objections that would block acceptance before external reviewers do.
Read First
references/repository-contract.mdreferences/peer-review-policy.mdreferences/claim-audit.mdreferences/cs-methodology-evaluation-policy.md
Workflow
- Identify target venue or assumed review culture.
- Read the paper/proposal/manuscript claim-first: abstract, introduction, contributions, experiments, related work, conclusion.
- List acceptance-critical claims and what evidence supports each.
- Check the paper against nearest prior work and target venue expectations.
- Attack novelty, significance, correctness, methodology, clarity, ethics, and reproducibility.
- Separate fatal flaws, major weaknesses, minor issues, and presentation issues.
- Produce reviewer-style scores only when the user asks or the venue requires them.
- Save reusable findings in
reports/reviews/or pages derived fromwiki/templates/reviewer-concern.md.
Review Lanes
Use multiple lanes even when writing one consolidated review:
- editor lane: venue fit, contribution threshold, audience, desk-reject risks
- methodology lane: design, datasets, baselines, metrics, statistics, ablations
- domain lane: related work, terminology, prior art, missing comparisons
- adversarial lane: strongest counterargument, overclaiming, alternative explanations
- reproducibility/ethics lane: code/data access, artifact quality, privacy, licenses, dual use
If the user asks for a simulated panel, keep findings from lanes separate before synthesis so the final decision cannot invent or duplicate reviewer concerns.
Review Standard
Every major criticism should include:
- exact claim or section
- why it matters
- evidence from paper/source/run
- what would fix or weaken the concern
- severity: fatal, major, moderate, minor
- confidence and what would change the judgment
Do Not
- Be performatively harsh without evidence.
- Suggest impossible experiments without labeling them as optional.
- Let writing polish hide methodological weakness.
- Accept citation volume as proof of related-work coverage.
- Edit the manuscript while acting as reviewer; produce a report and revision roadmap.
- Let an adversarial finding disappear during synthesis. If it is wrong, explain why.
Output Contract
For deep reviews, write:
reports/reviews/<review-id>.md- executive summary and recommendation
- strengths with evidence
- weaknesses grouped by severity
- methodology/reproducibility assessment
- missing related work and gap-positioning issues
- questions for authors
- prioritized revision roadmap