Methodology

How Scoring Works

The Manusights diagnostic isn't a grammar checker or a generic AI summary. It's a structured pre-submission screening system built specifically for biomedical manuscripts targeting high-impact journals. This page explains what happens when you upload a manuscript, how each score is generated, how we verify our own output, and where the system's limits are.

We publish this because researchers deserve to know what's behind the score before they trust it.

The Six Scoring Dimensions

Every manuscript is scored across six dimensions on a 1–5 scale. These dimensions mirror what journal editors and peer reviewers evaluate during initial screening and formal review.

Originality

1–5

What it measures: Does the manuscript introduce a genuinely new finding, mechanism, or framework?

How it's assessed: Cross-references the stated contribution against recent literature in the same subfield. Flags derivative framing or incremental-only advances.

Importance & Broad Interest

1–5

What it measures: Would researchers outside the immediate subfield care about this result?

How it's assessed: Evaluates whether the framing connects the finding to field-level questions, not just niche methodology.

Strength of Claims vs. Evidence

1–5

What it measures: Do the conclusions follow from the data presented?

How it's assessed: Scans for overstatement patterns: 'demonstrate' without mechanistic proof, causal language from correlational data, missing controls for key claims.

Methodological Soundness

1–5

What it measures: Are the methods appropriate, reproducible, and correctly reported?

How it's assessed: Checks for sample size justification, statistical test selection, blinding/randomization reporting, and figure-data consistency.

Clarity & Presentation

1–5

What it measures: Is the manuscript well-organized and easy to follow?

How it's assessed: Evaluates abstract structure (Problem → Gap → Method → Finding → Impact), figure labeling, and whether the narrative builds logically.

Prior Work & Reference Integrity

1–5

What it measures: Does the manuscript properly contextualize itself, and are its references real and current?

How it's assessed: Identifies missing key citations, outdated references, and whether the literature gap is stated explicitly. Also runs automated integrity checks: DOI verification (catches hallucinated references), retraction screening, self-citation ratio, and citation recency profiling.

Score interpretation: A score of 5 means the manuscript is strong in that dimension relative to the standards of high-impact biomedical journals. A score of 1–2 indicates specific, addressable weaknesses that are likely to trigger desk rejection or major revision requests. Scores are calibrated against the expectations of journals like Nature, Cell, and Science, not against average publications.

How the Diagnostic Works

From upload to delivered report, here's what happens at each stage.

1

Document Ingestion

Your manuscript (PDF or Word) is parsed into structured sections: abstract, introduction, methods, results, discussion, figures, and references. Tables and figure legends are extracted separately for cross-referencing against claims in the text.

2

Multi-Pass Analysis

The manuscript passes through multiple analysis layers, each focused on a different dimension. Structural analysis runs first (section completeness, abstract format, word count), followed by deeper passes for claim-evidence alignment, methodological rigor, and novelty assessment.

3

Reference Integrity Check

Your manuscript's reference list is verified against CrossRef and PubMed. Every DOI is resolved to confirm it exists. Retracted papers are flagged. Self-citation ratio and citation recency are calculated. Separately, every reference our AI cites in its own report is also verified and hallucinated citations are removed before delivery.

4

Journal Fit Assessment

Based on the manuscript's scope, methodology type, and contribution level, the diagnostic suggests three journal tiers: Reach (ambitious but possible), Realistic (strong fit), and Fallback (high-probability acceptance). Each suggestion includes rationale tied to the journal's recent publishing patterns.

5

Prioritized Action Plan

Issues are classified into three priority tiers. Priority A: must fix before submission (will likely cause desk rejection). Priority B: should fix (reviewers will flag these). Priority C: worth improving (strengthens the manuscript but won't block acceptance). Each issue includes a specific recommendation and rationale.

6

Report Assembly & Quality Check

The six scored sections, journal fit analysis, and prioritized action list are assembled into a structured report. A final consistency pass ensures scores align with identified issues and that no section contradicts another. The report is delivered as a downloadable document within 24 hours.

Citation & Reference Integrity

We verify citations in two directions: the references our AI cites in its report, and the references in your manuscript. Both are checked against real academic databases in real time.

Our Report's Citations

Every DOI and PMID the diagnostic cites is resolved against CrossRef and PubMed. Citations that fail verification are removed before the report reaches you. No hallucinated reference in our output.

Your Manuscript's References

Every reference in your bibliography with a DOI or PMID is independently verified. The diagnostic flags five categories of reference problems:

Hallucinated References

DOIs that don't resolve in CrossRef. This is the #1 signal that AI was used to draft the manuscript. Editors are actively screening for this.

Retracted Papers Cited

Cross-checked against CrossRef retraction metadata. Citing a retracted paper can trigger immediate desk rejection and raise research integrity concerns.

Duplicate References

Detected by DOI match and fuzzy title comparison. Duplicate entries signal sloppy bibliography management and waste limited reference slots.

Self-Citation Ratio

Calculated against manuscript author names. Ratios above 25% are flagged. Editors at top journals treat excessive self-citation as a manipulation signal.

Citation Recency Profile

Year distribution of your bibliography: last 5 years, 5–10 years, and 10+ years. Fields with fast-moving literature expect recent citations.

Validation Against Real Peer Review

We validate the diagnostic by running it blind against published manuscripts that have transparent peer review files available. The diagnostic is generated without access to reviewer comments, then compared issue-by-issue against what reviewers actually flagged.

21 / 32

major reviewer concerns matched across 3 validated papers, plus 10 additional issues reviewers did not raise

Nature Communications — WNK1 Kinase (Immunology)

Protocol

Blind run against a published manuscript with transparent peer review. The diagnostic was generated without access to reviewer comments, then compared issue-by-issue.

Result

11 of 17 major reviewer concerns matched

2 additional issues identified that reviewers did not raise

Known Limitation

Missed one rescue-experiment request requiring deep kinase subfield expertise

Read full case study →

Nature Communications — Neuroblastoma (Oncology)

Protocol

Same blind protocol. Diagnostic generated independently, then compared against published reviewer reports.

Result

5 of 5 major concerns matched

1 additional issue identified

Known Limitation

Small sample (5 concerns); high match rate may reflect straightforward methodology issues

Read full case study →

eLife — Skin TRM T Cells (Immunology)

Protocol

Blind run against an eLife paper with publicly available reviewer assessments under eLife's open review model.

Result

5 of 10 major concerns matched

7 additional issues identified that reviewers did not raise

Known Limitation

eLife's review model emphasizes different criteria than traditional journals; some misses reflect format differences rather than analytical gaps

Read full case study →

What the Diagnostic Does Not Do

Transparency about limitations is part of the methodology. Here's what the diagnostic is not designed to handle.

  • \u2717Replace peer review. The diagnostic is a screening layer that catches issues before submission. It doesn't replicate the judgment of a domain expert who has spent 20 years in the subfield.
  • \u2717Guarantee acceptance. No tool can predict whether a journal will accept your paper. Scores indicate manuscript readiness, not editorial decision outcomes.
  • \u2717Judge deep mechanistic novelty. Whether a particular kinase pathway finding is truly novel requires subfield-specific knowledge that goes beyond what any automated system can reliably assess. This is where our expert human reviewers add value.
  • \u2717Store your manuscript. Manuscripts are processed through Anthropic's zero-retention API and are not stored after your report is delivered. No training on your data. No retention.
  • \u2717Work outside biomedical sciences. The diagnostic is calibrated for biomedical and life sciences manuscripts. Humanities, social sciences, and engineering papers will receive less accurate assessments.

Data Handling & Privacy

Zero-retention processing

Manuscripts are processed through Anthropic's API with zero data retention enabled. Your document is not used to train any model.

No storage after delivery

Once your report is generated and delivered, the manuscript is deleted from our processing pipeline. We don't keep copies.

Encrypted in transit

All uploads and report deliveries use TLS encryption. Your manuscript is never transmitted in plaintext.

NDA available

For expert human reviews, a formal NDA is signed on every engagement. For AI diagnostics, the zero-retention architecture provides equivalent protection.

See it in action

Upload your manuscript and get a free readiness preview. No account required.