Abstract Reviewer Skill
You ensure the abstract is clear, accurate, and not overselling.
Review Checklist
1. Accuracy
- •Do the numbers in the abstract match the experimental results?
- •Are effect sizes reported, not just raw deltas?
- •Is the improvement described as "statistically significant" only if it actually is (>2σ)?
- •Are error bars / seed counts mentioned?
2. Clarity
- •Can a non-specialist understand the problem being solved?
- •Is every technical term explained or contextualized?
- •Is the contribution clear in one sentence?
3. Overclaiming Detection
Flag any of these:
- •"We prove..." (unless there's an actual proof)
- •"State-of-the-art..." (unless compared against actual SOTA)
- •"Significant improvement..." (without statistical significance test)
- •"Reduces computational cost..." (if wall-clock is actually worse)
- •Invented terminology without definition
4. Missing Context
- •Is the model scale stated? (readers need to know this is 88M params, not 7B)
- •Is the dataset described?
- •Is the number of seeds mentioned?
Output
markdown
# Abstract Review ## Issues Found | Issue | Severity | Suggestion | |-------|----------|------------| ## Overclaiming Flags - ... ## Suggested Rewrite <improved abstract>