From $9 per review · Anonymity Check bundled free
Paper Review
A senior-reviewer-level red team for your manuscript. Catches the blind spots reviewers will flag before they do. Four-persona adversarial debate, live citation cross-check, figure-integrity scan. Delivered in under 10 minutes.
Sample report
What does the report look like? View sample
# Manuscript Review
## Critical Blind Spots
- **Hyperparameter search asymmetry favors the proposed method.**
Quoted: "We tuned learning rate and batch size on the validation
set; baselines used reported defaults."
Surfaced by: Methodology Critic, Statistical Skeptic (consensus).
Fix: Match search budget for each baseline; report the protocol
in the methods section.
- **Sample-size justification missing for Experiment 2.**
Quoted: "We collected 24 participants ... yielding sufficient
statistical power."
Surfaced by: Statistical Skeptic.
Fix: Provide an explicit power calculation with assumed effect
size and the source of that assumption. With n=24 and the
reported effect, post-hoc power is below 0.6.
## Data-to-Claim Contradictions
- Claim: "Our method beats all baselines by a wide margin."
Evidence: Table 2 shows the proposed method tied with Baseline C
within reported standard errors on 4 of 7 datasets.
## Rectification Checklist
- [A] Add hyperparameter search protocol to methods (page 5,
Section 3.2). Match search budget across all baselines.
- [A] Replace "wide margin" framing in abstract and Section 5
with what the data actually shows.
- [B] Add per-seed variance to all results tables.
- [B] Reference [17] does not resolve to a CrossRef entry,
re-verify the citation.
- [C] Figure 3's y-axis breaks 0-60 then jumps to 90; redraw on
a continuous scale.
## True Novelty Estimate
The core architectural contribution is genuinely novel and not
covered by [11] or [23]. The empirical comparison, as currently
framed, oversells the gap. Marginal advance: the proposed mechanism
is interesting but the published evidence does not yet warrant the
strength of the abstract's claim.
## Reference Verification Summary
47 of 52 references verified against CrossRef. 3 had weak title
matches (re-check [12], [17], [31]). 2 had no DOI and could not be
located by title, verify these are correctly cited.
## Citation Support Audit
Verdicts: 18 supported, 4 partially supported, 2 not supported by
abstract, 1 contradicted, 3 source unavailable.
| Claim | Cited ref | Verdict | What the source's abstract says |
|---|---|---|---|
| "Method X reduces error by 40%." | [14] | Contradicted | Abstract reports a 4% reduction, not 40%. Verify the figure. |
| "Prior work proved this is optimal." | [9] | Not supported by abstract | Abstract describes a heuristic; optimality may be in the full text. |
_5 citations with weaker claims were not audited (per-run cap)._
Reviewing more than one manuscript? Volume packs save 16-17% per review.
Nothing is kept after delivery. The review is deleted from our servers the moment you retrieve it, and if a result reads as low-quality we'll refund the price. Advisory only, you decide what to change.
How the review works
- Layer 1 — Figure-integrity vision scan. Every page is rendered and reviewed for digital manipulation concerns, figure-vs-text contradictions, and presentation issues (broken y-axes, missing error bars, color-only encoding). Hedged language; flags concerns to investigate, never accusations.
- Layer 2 - Citation support audit. Every reference is verified against CrossRef, then we go deeper: for the most load-bearing claims, we fetch the cited source's abstract and check whether it actually supports the claim you attached it to. Each finding shows your claim next to the source's own words, with a verdict: supported, partially supported, not supported by the abstract, contradicted, or source unavailable. Hedged language; verify each against the full text.
- Layer 3 — Adversarial four-persona panel. Methodology Critic, Statistical Skeptic, Data Integrity Officer, and Editor-in-Chief each red-team the paper in parallel. Findings are merged with a cross-persona consensus filter so issues raised independently by multiple personas surface first.
- Layer 4 — Rectification synthesis. A final pass produces the structured report you'll act on: critical blind spots, data-to-claim contradictions, an A/B/C rectification checklist, and an honest novelty estimate.
What you get
- One review of one manuscript (up to 20 MB PDF, ~100 pages).
- Markdown report you can download.
- Delivered in under 10 minutes via a polling status page.
Frequently asked questions
What does the review include?
Four layers in sequence: a Claude Vision pass over your figures and tables, a live CrossRef cross-check of every reference, an adversarial four-persona reviewer debate (Methodology Critic, Statistical Skeptic, Data Integrity Officer, Editor-in-Chief) with consensus filtering, and a synthesis pass producing critical blind spots, data-to-claim contradictions, a prioritised rectification checklist, and a true-novelty estimate.
How long does it take?
Typically 4-8 minutes for a 20-page manuscript. You don't need to keep the tab in the foreground — the status page polls until the review is ready and renders it inline.
What about my privacy? Is the manuscript stored?
Your manuscript is sent to Anthropic's Claude API for analysis. Anthropic retains inputs for up to 30 days for abuse monitoring, and does not use them for model training. Purplelink itself does not store your manuscript or the resulting review after delivery: the review is deleted from our servers the moment you retrieve it on the status page. Save the result locally before closing the tab.
What domains does it cover?
You pick one of five domain profiles at upload: Machine Learning, Biomedicine, Psychology and Social Science, Chemistry and Materials, or General. Each profile loads domain-specific attack vectors so the reviewers focus on the failure modes that matter for your field.
How deep is the citation check?
Two levels. First, every reference is checked for existence against CrossRef (dead DOIs, hallucinated or weakly-matched titles). Second, for the most load-bearing claims we fetch the cited source's abstract from OpenAlex, Semantic Scholar, or CrossRef and assess whether it supports your claim. This is an abstract-level check: a claim supported only in the full text is marked "not supported by abstract," not wrong - verify it against the source. When no abstract can be retrieved, the citation is marked "source unavailable" rather than guessed. On citation-heavy papers we audit the most important claims first and tell you how many were not audited.
Is this a replacement for peer review?
No. Treat it as a first-pass red team before you submit — a way to catch the obvious problems a real reviewer will flag before they do. The review is advisory only; you remain responsible for every decision about your manuscript.
What if the review misses something or makes a mistake?
AI reviewers can be wrong. The output flags concerns with hedged language and quotes the manuscript when challenging a claim, so you can verify every finding against your text. If a specific result reads as low-quality, email ben@purplelink.llc — we'll refund the price.
What file format?
PDF up to 20 MB. If you're submitting LaTeX, compile to PDF first (our free LaTeX-to-PDF tool works).
What about clinical or patient data?
Do not submit unredacted patient data. The manuscript will be sent to Anthropic's API with 30-day retention. For papers containing identifiable health information, redact before upload.