Writing a useful review

Sooner or later you'll be asked to review. A good review helps two people: the editor, who must decide, and the authors, who must improve the paper.

A structure that works

1. Summary (3–5 sentences): what the paper claims and how. This shows the authors you understood it.
2. Strengths: what is genuinely good.
3. Major concerns: problems that affect whether the conclusions hold. Number them.
4. Minor points: clarity, typos, small missing details.
5. Recommendation (often in a separate box for the editor), with a short reason.

Make every concern useful

• Specific: point to the section, figure or line.
• Explained: why is it a problem?
• Actionable: what would fix it? An analysis, a control, a citation?

ExampleUnhelpful: "The evaluation is weak." Helpful: "The main comparison (Table 2) uses a single random seed and a baseline with default settings. Please report mean and spread over at least 5 seeds, and tune the baseline with the same budget, so readers can tell whether the 1.2-point gain is real."

Use everything you've learned

Claim versus evidence, baselines and ablations, error bars, leakage, overclaiming: the skills from Deep Paper Reading, Statistics and Designing Experiments are exactly what a reviewer needs.

Tone

Review the paper, not the people. Write what you'd be comfortable saying to the authors face to face.

RememberSummary, strengths, numbered major concerns, minor points, recommendation. Every concern specific, explained and fixable.