Rigorous & Fair Peer Reviews
Write Reviews That
Actually Help Authors
REVAS reads the weaknesses section of your draft review, scores every point you raise, and shows you exactly what to sharpen — before you submit.
REVAS doesn’t write reviews. It makes yours better.
Quality Check
Four dimensions of a useful review
The Quality Check scores each paragraph across four quality dimensions. For every issue it finds, you get concrete feedback you can apply before submitting.
Actionability
Does your feedback give the author a clear next step?
REVAS flags vague suggestions so every comment drives a concrete improvement.
Grounding
Is your critique tied to a specific part of the paper?
Good critique points to a specific section, figure, or passage. REVAS catches comments that float free of the paper.
Verifiability
Are the claims in your comment backed by evidence?
Every claim in your review should be checkable. REVAS catches assertions that aren't backed by what's in the paper.
Tone
Is your critique respectful and direct?
Hostile, sarcastic, or dismissive language sinks a critique. REVAS helps you deliver tough feedback without coming across as harsh.
Guideline Check
Catch common reviewing issues before authors see them
The Guideline Check screens every point you raise against ten common reviewing issues the ARR reviewing guidelines explicitly ask reviewers to avoid — each flag links straight to the relevant guideline.
The proposed approach is fairly straightforwardH7 and the reported gains are not particularly surprisingH1. The authors could also try a retrieval-augmented variantH13 to further validate the framework.
A surprising result is not a requirement for a contribution. Point to a specific claim you find unconvincing instead.
Simplicity alone is not a weakness. If something is missing, name the baseline or comparison; otherwise this risks penalising a clean method.
Additional-experiment requests should be framed as suggestions, not weaknesses — and only when X is essential to the paper's claims.
Getting Started
How REVAS works
Three steps from rough draft to a review that holds up.
Paste your draft review
Drop in the weaknesses section of your review. Each paragraph is treated as a separate point and scored on its own.
See where it falls short
Every point gets quality scores and a guideline check. Color-coded severity bars show exactly where to focus.
Refine, re-run, repeat
Revise the flagged paragraphs and re-analyze. Iterate until your review is rigorous and fair.
The Revision Loop
From flag to fix
Here’s a vague critique, the feedback REVAS gives, and the revision the reviewer wrote.
The related work section is incomplete and the experiments could be better. Overall the evaluation is not convincing and does not sufficiently support the claims made in the paper.
This comment names problems without locating them in the paper. The authors can’t act on “incomplete” or “could be better” without knowing which related work is missing or which experiment falls short. Point to the specific section or table, name what’s missing, and say what evidence would change your assessment.
The related work section omits the direct preference optimization line of work (Rafailov et al., 2023; Meng et al., 2024), which competes directly with the proposed approach. Table 3 shows only a 0.4-point win-rate gain on AlpacaEval 2 — the authors should add significance testing to support their claims.
Confidentiality
Your review stays yours
University-hosted
Your review runs on self-hosted models — it is never sent to third-party APIs or commercial AI providers.
Encrypted storage
Reviews are stored encrypted with restricted access. We analyze usage only to improve REVAS itself.
Only your words
REVAS never sees the paper under review — it reads only what you wrote. And it never suggests review content: it provides only feedback on your draft.
Details in our Terms of Use.
FAQ
Common questions
Am I allowed to use REVAS when reviewing for ARR?
Yes — REVAS is an optional review-support tool supported by ACL Rolling Review (see ARR’s announcement). You only ever paste your own review text — never the paper — and REVAS doesn’t generate review content. ARR’s review form includes an option to disclose that you used it.
Does REVAS write my review for me?
No — by design. It scores your points, names what’s missing, and suggests a direction. The reasoning and the wording stay yours.
What happens to my review?
It’s processed on university-hosted models, stored encrypted, and never shared with third parties. We use it only to improve REVAS — see the Terms of Use.
Which venues does it work for?
The Quality Check — actionability, grounding, verifiability, tone — applies to peer review across fields, and was developed and tested on NLP reviews. The Guideline Check is specific to the ACL Rolling Review guidelines, so NLP/ML reviewers get the most out of it.
Is REVAS free?
Yes — free for academic peer reviewers. REVAS is built by UKP Lab (TU Darmstadt) and MBZUAI to make peer review more useful for authors.
Refine your review before you submit — for the authors and for better science.
Scores and concrete suggestions for every point in your weaknesses section.
Get feedback on your reviewFree for academic peer reviewers.