Upload
Send a PDF. You get a fixed price from its length and a list of quick fixes before you pay. Sign in with your ORCID iD or an emailed link; the account is made when you pay.
Scientific review service
Know what a referee will say, before you submit.
A referee-grade report on your manuscript, with every claim graded by how well it is supported and a line under the decision saying what would reverse it. Priced from the manuscript before you pay.
How it works
Upload, get your report, revise
The review is automated. A person is involved once, when a result is published.
Get your report
Each claim is graded by how well it is supported, with the kind of evidence behind the grade. Every report passes a release gate before you see it, and every decision carries the condition that would reverse it.
Revise, then publish if you choose
Upload the next draft and see what is fixed, still open or new. When you are ready, publish a result that anyone can check. An editor signs every published result.
A ReviewBot report is scientific criticism of a manuscript. It is not a journal decision, an endorsement, or a guarantee of correctness, and nothing here is presented as the judgment of a human peer.
What a grade means
Four grades, four kinds of evidence
Every claim in your paper gets one grade. Every grade names the evidence it rests on. There are no scores and no percentages.
Proven
Machine-checked proof, matched to the paper by a named editor at publication.
Recomputed
Numbers rerun from the shared data and code.
Conditional
Holds if a stated assumption holds.
Open
Not yet supported. An open claim is not a wrong claim; it is one the evidence does not reach.
The evidence behind a grade is one of four kinds
- Reviewed proof
- The argument was read and checked step by step.
- Exact finite computation
- A statement with finitely many cases was checked exhaustively, or symbolically, with no rounding.
- Numerical calculation
- A number was recomputed, with the precision and tolerance stated.
- Proof-assistant kernel check
- A formal statement was compiled by a proof assistant, and an editor confirmed it says what the paper says.
Under every decision there is one more line: what would reverse it. The experiment, the proof repair, or the assumption that would change the verdict is written down with the verdict.
How we check
Every claim here comes with a proof you can run. And we can check yours too.
A review is only worth what can be verified behind it. So nothing is reported as checked unless the check can be rerun: the inputs, the code, the exact statement tested and the result, kept with the report and tied to the exact version of the manuscript by its hash.
The same standard applies to results we publish ourselves. A published result carries its data, its logs, its checksums and a command that reruns every check on your own machine. If it cannot be rerun, it is not published.
- Frozen inputThe manuscript and every attachment are hashed before anything is read. The report refers to that version and no other.
- Located findingsEvery finding points to the page, theorem, equation or line it is about. A suspicion is never written up as a defect.
- Reproducible checksA numerical or exact check records its inputs, precision, tolerance and software version, so it can be rerun.
- Independent reproductionWhere a result matters, it is recomputed separately before it is called verified. The report says which findings were.
- A record you can keepEach round is logged with the hashes of what went in and what came out. You can show later exactly what was reviewed and when.
Price
Three tiers, one fixed price each
The price is set from your manuscript's length before you pay and does not change. Nothing is added at checkout.
Tier 1
Essential
A$8.90
Targeted scientific review
The main claim, its assumptions and the common weaknesses; a map of the central claims; selected checks; the three to five revisions that matter most. Ready in hours.
Tier 2
Research
A$39
Targeted scientific review, broader
Every claim graded; derivations located in the source; the weakest link tested adversarially; a claim dependency graph; a comparison with your previous draft. Ready in days.
Tier 3
Advanced
from A$249
Full ReviewBot review when the gate passes
The deepest protocol: an independent second model, extended formal and data checks, a comprehensive revision comparison. Ready in days.
Tier 1 and Tier 2 are targeted scientific reviews. Only a Tier 3 run that passes the release gate is delivered as a full ReviewBot review; one that does not is delivered and charged as a targeted review. Two independent reviews can be bought at any tier for about 1.9 times the price. Prices are an Australian launch experiment in Australian dollars, GST included where it applies, and may change; the price you are quoted is the price you pay.
Published results
An archive where the review sits beside the paper
When a paper passes, its author can publish it here under their name, with the review and every grade beside it. Anyone with a qualified account can then request a deeper review of any published paper, and that result is published too, whichever way it falls.
There are no comments. A critique is a paper, submitted and reviewed to the same bar, and the two papers show each other. Null results are kept with the rest. The archive opens with the first published results.
- Each entry showsAuthor, version, the PDF's hash, field of research, result status, and the status of every tier: passed, revision requested, or not run.
- Two actions on every paperRequest a Tier 3 review. Respond with a paper.
- One person, onceAn editor checks that a formal proof says what the paper says, and signs the publication. Reports themselves are released by the gate.
- Not a verdict on your paper elsewhereA result here does not decide any journal's or repository's decision, and a claim that was not assessed is shown as not assessed, never as a pass or a fail.
- A sealed engineThe review engine has no route to the internet. The only way in is your manuscript; the only way out is the report.
- Never used for trainingYour manuscript is read to review it and for nothing else.
- Your consent, three timesOnce to review. Once more, only if you choose to publish. A third, off unless you tick it, for research opportunities arising from the manuscript to be recorded.
- Kept for a fixed period, then deletedManuscripts and reports are deleted after a fixed period following your last round. You can delete them earlier. Published results stay until you withdraw them.
- Who sees whatYou see your own submissions. An editor sees rounds that are up for publication. The operator sees system logs, not manuscripts.
Confidentiality
Your manuscript stays yours
Unpublished work is sensitive. The service is built so that the fewest systems and people see it, for the shortest time, and so that you can show afterwards exactly what was reviewed.
About
Who is behind it
ReviewBot is built by Emeritus Professor Pablo Moscato, University of Newcastle, Australia, who introduced memetic algorithms in 1988.
The platform is built and operated by Luke Martin, Newcastle. The site is a preview: the service opens with a closed pilot, and the name and address may change.