Skip to content
Marginalia
  • Method
  • Sample report
  • Benchmark
  • Journals
  • Pricing
中文Free scan

Pre-submission reviewLife sciences

Read your paper the way Reviewer 2 will.

Upload a manuscript. In about 30 minutes you get the concerns referees are likely to raise — each tied to the sentence that triggers it, why it matters, and what to change.

Opening soon

Uploads open shortly. Until then, see what a full report looks like in the sample report.

The free scan returns a risk score and the most serious issue in full. The complete report is $29. Read a sample report

Manuscript · Draft 3p. 1 of 31

Research article

FIS7 loss drives mitochondrial elongation and synaptic decline in aging cortical neurons

L. Okafor, M. Brandt, S. Iyer, J. Whitcombe

Abstract

Mitochondrial dynamics shape synaptic health, yet the factors that couple organelle morphology to synapse loss remain unclear. Here we provide the first evidence linking mitochondrial shape to synaptic decline1, identifying FIS7, an outer-membrane protein, as a regulator of fission in cortical neurons. Knockdown elongated mitochondria and reduced spine density by 41% (n = 142 neurons; p < 0.001)2. FIS7 levels fell as neurons matured, and these data establish FIS7 as a driver of synaptic decline during aging3.

1 · Introduction

3 notes on this page

Fig. 1 — A page as a referee reads it. Hover or tap a note to find the sentence it flags — every finding quotes your manuscript verbatim.

of referees’ major concerns found on reviewbench
85%of referees’ major concerns found on reviewbench
from upload to a full report
~30 minfrom upload to a full report
reviewed independently, then merged
2 passesreviewed independently, then merged
checked against Crossref and Retraction Watch
42 refschecked against Crossref and Retraction Watch
§ 1Findings

Specific enough to act on tonight.

Each issue names where it is, quotes what you wrote, says why a referee will object, and proposes the smallest change that answers it. Quotes that cannot be found in your file are thrown out.

  • Ranked by severity. Major concerns first — the ones that decide revisions.
  • Grounded in your text. Every quote is matched against the manuscript.
  • Tracked across drafts. Mark issues addressed; re-run on the revision.

Major/Causal claim/Results · Fig. 3

The causal claim rests on a single shRNA with no rescue

“demonstrating that FIS7 is required for the maintenance of dendritic spines in mature neurons.”
Why a reviewer will raise it
Every loss-of-function result in Fig. 3 uses one shRNA. Reviewers routinely ask whether the phenotype is off-target, and “required” is stronger than knockdown alone can support.
What to change
Re-express an shRNA-resistant FIS7 and show spine density recovers, or repeat the key panels with a second, independent shRNA or a CRISPR knockout. Until then, soften “required” to “contributes to”.

Major/Statistics/Methods · Fig. 2D

Neurons, not animals, are treated as independent replicates

“n = 142 neurons from 3 cultures; p < 0.001, two-tailed t-test.”
Why a reviewer will raise it
Neurons from the same culture are not independent. Pooling them inflates n and can turn a culture-level effect into an apparently strong one — a frequent reason for a major-revision request.
What to change
Report one value per culture (or per animal) and test on those, or fit a mixed-effects model with culture as a random effect. State the biological n in every legend.

Fig. 2 — Two of seven findings from the sample report on a fictional neuroscience manuscript. Open the full sample

§ 2Method

Two referees who never compare notes.

A full report runs two independent review passes and merges them. On our benchmark the merge lifted major-concern recall by 8 points, from 70% to 78%, over a single pass — the step most people skip when they ask a chatbot.

Full report pipelineThe manuscript is parsed, then reviewed by two independent passes while references are checked. Every quote is verified against the manuscript, with a retry when a pass returns fewer than five findings. Findings are merged and ranked into the report, about 30 minutes after upload.Fewer than 5 findings → retry, up to 3×01ParseText, legends, references02aReview pass AIndependent referee02bReview pass BIndependent referee02cReference checkCrossref, Retraction Watch03Verify quotesUnfound quotes dropped04Merge & rankDeduplicated, by severity05Your reportQuote, reason, fix0:00~1 min~24 min~27 min~30 min
  1. 01ParseText, legends, references
  2. 02Two review passesRun independently, in parallel
  3. 02cReference checkCrossref, Retraction Watch
  4. 03Verify quotesUnfound quotes dropped
  5. 04Merge & rankDeduplicated, by severity
  6. 05Your reportQuote, reason, fix

Fig. 3 — The full-report pipeline. The free scan runs a single fast pass of steps 01–03 and shows the most serious finding.

Free scan in 3 minutes

A risk score for the whole manuscript and its single most serious issue, in full. No account needed to see it.

Full report in about 30 minutes

Every finding ranked, with quote, reason and fix, plus a line-by-line reference check. Emailed when ready; export as PDF.

Your file, then gone

Processed without retention by model providers and deleted from our storage 30 days after the report.

§ 3Revisions

Know what the next draft fixed.

Upload the revision and the report is re-run against the last one. Each earlier finding is marked resolved or still open, and anything the revision introduced is flagged as new — the same check a referee makes in round two.

Included with every report pack.

Draft 3Draft 4

4 resolved · 1 open · 1 new

  1. 01

    The causal claim rests on a single shRNA with no rescue

    Rescue added as Fig. 3E; spine density recovers to 94% of control.

    Resolved
  2. 02

    Neurons, not animals, are treated as independent replicates

    Re-analysed per culture with a mixed-effects model (n = 6).

    Resolved
  3. 03

    The novelty claim overlooks the Drp1 literature

    “First report” removed; Drp1 work now cited in ¶4.

    Resolved
  4. 04

    Aging is claimed in the title but tested only in culture

    Title unchanged; still no in vivo time point.

    Still open
  5. 05

    Animal age and sex are missing from the Fig. 4 legend

    Added: 12-week males and females, both groups.

    Resolved
  6. N1

    The new rescue panel has no statistics reported

    Fig. 3E shows means only; add n and the test used.

    New in Draft 4

Fig. 4 — Draft 4 of the sample manuscript checked against its draft 3 report. Resolved findings are struck through; the open one is the reason to wait before resubmitting.

§ 4Evidence

We publish our hit rate. Including where it is low.

reviewbench takes eLife Reviewed Preprints, whose real referee reports are public, and asks how many of the referees’ major concerns our report raised too. Every model or prompt change is re-scored before it ships.

85%

major-concern recall, held-out set of 20 papers

Read the methodology
Concern type
Launch gate 85%
Evidence for main claimn=41
91%
Missing experimentn=36
86%
Statistics & methodsn=33
84%
Alternative explanationn=20
80%
Key information missingn=19
74%
Novelty & framingn=25
68%
50%60%70%80%90%100%

Fig. 5 — Share of referees’ major concerns also raised by the report, by concern type. Novelty and framing remain the hardest for a model to anticipate. Runs differ by about 2 points, so smaller changes are not reported as improvements. Updated 23 Oct 2026.

§ 5Reference check

The part that doesn’t depend on a model.

Each reference is looked up in Crossref and Retraction Watch. A DOI either resolves to the paper you cited or it doesn’t; a paper is either retracted or it isn’t.

Verified
38
Retracted
1
Mismatch
2
Not found
1
Reference check results
Ref.CitationStatusFinding
[7]Adeyemi K. et al. J Cell Sci 132, jcs2281 (2019)doi:10.5555/jcs.2019.2281VerifiedDOI resolves to the cited article.
[12]Zhang H.-L. et al. Cell Rep 18, 1102⁠–⁠1115 (2017)doi:10.5555/celrep.2017.1102RetractedRetraction notice issued; check whether the claim it supports still stands.
[19]Kowalski T. & Reyes A. J Neurosci 40, 512⁠–⁠527 (2020)doi:10.5555/jneuro.2020.0512DOI mismatchDOI points to a different paper by other authors.
[23]Hallström R. et al. Neuron 102, 81⁠–⁠96 (2019)doi:10.5555/neuron.2019.0081VerifiedDOI resolves to the cited article.
[31]Moreau P. Mitochondrion 58, 44⁠–⁠51 (2021)No DOI givenNot foundNo record in Crossref under this title, journal or year.

Table 1 — Five of 42 references from the sample report. Status is shown with an icon and a word, never colour alone.

§ 6Cost

A referee’s read for the price of a reagent.

  • Editing-firm pre-submission review

    $800–2,0001–2 weeks

    One human reader. Thorough, slow, expensive.

  • General chatbot

    $20 / monthMinutes

    No review framework; tends to praise. Does not check references.

  • Marginalia full report

    $29~30 minutes

    Two passes merged, quotes verified, references checked, recall published.

How the options compare
OptionPriceTurnaroundWhat you get
Editing-firm pre-submission review$800–2,0001–2 weeksOne human reader. Thorough, slow, expensive.
General chatbot$20 / monthMinutesNo review framework; tends to praise. Does not check references.
Marginalia full report$29~30 minutesTwo passes merged, quotes verified, references checked, recall published.

Table 2 — Editing-firm range from published pre-submission review price lists; chatbot at a typical individual plan. Checked 23 Oct 2026.

Credits never expire and work on any manuscript — a revision or a new paper. Refunded if a report has fewer than three substantive issues.

See pricing
§ 7Questions

Response to reviewers.

The questions researchers ask before trusting us with an unpublished paper.

How we handle manuscripts
Comment 1Will this replace peer review, or a colleague’s read?
Response

No. It is a self-check you run before submitting, so the obvious objections are dealt with before a referee sees the paper. It does not judge whether your science is right, and it does not guarantee acceptance.

Comment 2How do you know the findings are any good?
Response

We run every release against reviewbench: published eLife manuscripts whose real referee reports are public. We count how many of the referees’ major concerns the report also raised, and we publish that number — including the concern types where we are weakest.

Comment 3Will my unpublished manuscript be used to train a model?
Response

No. Model providers process it under zero-retention terms, the file is deleted 30 days after your report, and you can delete it sooner from your account. Every subprocessor is listed on the security page.

Comment 4Does it rewrite or polish my English?
Response

No. Language editing is a different job, and tools built for it do it better. Every finding points at a sentence you wrote and tells you what a referee will object to; the rewriting stays yours.

Comment 5What if the report finds nothing useful?
Response

If a full report contains fewer than three substantive issues, email us within 30 days and we refund it. No form, no argument.

Before you press submit

Fewer surprises in the decision letter.

The free scan takes about three minutes and needs no account. You see the most serious issue before deciding anything.

Opening soon

Uploads open shortly. Until then, see what a full report looks like in the sample report.

  • Nothing retained

    Model providers keep nothing and never train on it.

  • Deleted in 30 days

    Or sooner, from your account, in one click.

  • Subprocessors listed

    Who touches your file, and for what.

Marginalia

Pre-submission review for life-science manuscripts. A self-check before you submit — not a substitute for peer review.

Product

  • Method
  • Sample report
  • Free scan
  • Pricing

Evidence

  • Benchmark
  • Journal guides

Trust

  • Security & data
  • Privacy
  • Terms

© 2026 JINYAN LLC. Manuscripts are never used to train models.

Colophon — set in Source Serif 4, Inter and IBM Plex Mono.