An independent tool, not affiliated with, endorsed by, or connected to Nature, Nature Portfolio or Springer Nature, and not a substitute for real peer review. The name refers to the open project mumdark/nature-review-studio that this tool derives from — not to the journal Nature, to Nature Portfolio or to Springer Nature. It is a derived work of the open project mumdark/nature-review-studio, which publishes no licence file — its review taxonomy and vocabulary are used with credit and none of its corpus is redistributed. That project does not endorse this app.

See it first — six real manuscripts, six different verdicts

Every one is free, needs no sign-in, and replays a saved run at no charge.

Each example loads a full manuscript, measures it in this tab, and replays the review a model actually returned for it — the same fixture goes through the same renderer as a live run, so what you see is what you would get. Pick the situation closest to yours.

Or bring your own — drop a file, pick one, or paste the text

Drop a .pdf, .docx, .tex, .md, .txt, .odt, .rtf, .html or .epub here

The type is detected from the file's bytes, not its name — a PDF called paper.docx is read correctly and the mismatch is reported. Nothing leaves this tab in the free lane.

…or a link to a preprint — fetched server-side, signed-in only

Title — filled from the file, edit if wrong

Manuscript text — extracted, or pasted directly

Target journal (optional)

Notes for the panel (optional)

Structural report

Nothing measured yet. Load an example above, or drop a manuscript, and the full table appears here — free, in this tab, with no sign-in and no model call.

Free · measured in this tab

Structural report

Show only the rows needing attention

Structural report: one row per check, with the measured value, the expected target and a status.
Group	Section or check	Present	Measured	Expected	Status

Citation cross-check — read in both directions

Citation cross-check: citations that resolve to nothing, and reference entries the text never cites.
Direction	Marker	Measured	Expected	Status

How it works

Most of a desk check is measurement, not opinion

Whether the abstract is 340 words, whether Figure 4 has a caption, whether reference 17 is ever cited, whether a p-value ever appears beside an effect size, whether the ethics statement exists — none of that needs a model, and all of it gets a paper rejected. Drop the file and the browser reads it (PDF text layers, Word, LaTeX, OpenDocument, RTF, EPUB, HTML, Markdown, plain text), detects the type from its bytes, and prints the whole structural report as a table — one row per check, with what was measured, what was expected, and a status — before you sign in to anything.

The citation cross-check runs in both directions

Numeric, author–year and BibTeX-key styles are all parsed. Citations that resolve to nothing in the reference list get a row of their own; entries in the list that the text never cites get one too. The same pass inventories figures and tables, notes which are cited in the running text and which are never mentioned, measures reference-format consistency, and sniffs for emails, ORCIDs, DOIs, trial IDs and grant numbers you may want stripped for a double-blind venue.

The panel is metered, and it is held to the measurements

Your paper's method fingerprint chooses the reviewer set — a wet-lab mechanism paper gets three reviewers, a clinical cohort with a model gets five — and the panel returns an editor letter with one of four revision-level bands, per-reviewer major and minor concerns with inline evidence pointers, a consensus block limited to concerns two reviewers raised independently, and a revision task table. Then every figure, table, section, reference, p-value and sample size the review cites is checked back against your manuscript, and anything it invented is printed by name. Long manuscripts are clipped from the middle, never the ends, and the report says exactly what was cut.

The examples are not screenshots

Every example in the gallery ships as a fixture: the raw reply a model produced for that exact manuscript, replayed through the same parse, contract-enforcement and render chain a live run uses. Each one is checked mechanically before release — it must normalize with zero contract warnings and reconcile with zero unverified pointers, and its input hash must match, or it does not ship. Nothing in the gallery was written by hand to look like output.

Credit. A derived work of mumdark/nature-review-studio (https://github.com/mumdark/nature-review-studio): its 12-axis concern taxonomy, its manuscript-fingerprint→reviewer-set mappings, its closed strategy and action-status vocabularies, its four revision-level bands and its consensus-only output rule are what this studio renders and enforces. That project publishes no licence file, so this app is a clearly derived work with credit rather than a republication — none of its corpus, snapshots or reference documents are redistributed here, only the taxonomy and vocabulary. The source project does not endorse this app.

Non-affiliation. This app is not affiliated with, endorsed by, or connected to Nature, Nature Portfolio, or Springer Nature, or any of their journals, editors or staff. Its name is inherited from the open source project it derives from, mumdark/nature-review-studio, and the word “Nature” in it refers to that project rather than to the journal Nature or its publisher. Nothing it produces is a Nature journal review or any journal's official assessment, no journal has seen or approved it, and the revision-level band is this tool's own assessment rather than anyone's editorial decision. It is not a substitute for real peer review.