Simulated Peer Review

Academic referee simulation · Test release

Simulated Peer Review

Constructive AI feedback to help authors make their papers clearer, better supported, and more robust under criticism—not a pass/fail grade.

The report identifies strengths, tests reasoning and evidence, flags gaps or ambiguities, and suggests proportionate repairs. It places its emphasis on the philosophy of mind, language, perception, and knowledge, and on artificial intelligence, but papers, essays, and informal drafts across subjects are welcome and held to the same standards. This service is not affiliated with or endorsed by an academic journal.

The review aims to assess arguments impartially: it examines the thesis, distinctions, inferences, supporting evidence, and serious objections rather than rewarding agreement with a particular philosophical position.

Each report applies nine basic checks:

  1. Factual accuracy — are empirical and historical claims supported?
  2. Citation accuracy and hygiene — do the citations fit the claims they are meant to support?
  3. Logical structure — does the conclusion actually follow from the premises?
  4. Misattribution — are other thinkers’ positions represented fairly?
  5. Unargued assumptions — are load-bearing premises defended or at least flagged?
  6. Equivocation — does a key term shift its meaning mid-argument?
  7. Missing literature — are the obvious opposing views and relevant work engaged?
  8. Scope discipline — does the paper deliver what it promises, and no more?
  9. Scholarly apparatus — are references, notes, and definitions adequate to the claims?

The instructions were calibrated by testing them on established philosophical papers and revising them in light of published guidance for journal referees. Calibration changed the instructions; the underlying AI model was not retrained. The service remains experimental.

The report includes a summary, a simulated ACCEPT, REVISE, or REJECT recommendation, blocking issues, major revisions, and minor issues, strengths, and an account of what was and was not checked. These are advisory simulated recommendations, not grades, certificates of truth, or actual publication decisions. This is an argument-and-scholarship review, not a copyediting service.

This is AI-generated feedback, not review by human scholars. It can be mistaken or biased and does not independently verify external sources or citations. You can see the prompt here. Its findings are arguments to weigh, not rulings to accept; you remain the judge of what to revise.

Reference-paper assessments show how the same scholarly rubric was applied to a growing collection of established philosophical papers and other reviewed submissions. They illustrate the kinds of strengths and revision needs a report can identify—not pass/fail benchmarks, grades, or rankings. Even influential papers can receive substantial criticism. These assessments used the internal subscription model route; this public service uses OpenRouter. Both provide manuscript-only assessments without independent external source verification. The comparison page records the reviewed versions and limitations.

Checking review availability…

01 Your writing

TXT, Markdown, text-based PDF, or DOCX · maximum 5 MiB. Your original file stays in your browser; requesting a review sends the extracted text to the AI service. PDF extraction is limited to 200 pages. No OCR: for scanned PDFs, paste text you have extracted yourself.

Review and edit the extracted text before submitting. Formatting, tables, footnotes, and PDF reading order may change during extraction. Nothing is silently truncated.

0 / 40,000 words · 0 / 320,000 UTF-8 bytes

02 A little context (optional)

Up to 200 characters.

Any genre or subject · up to 80 characters.

You can name a target journal for a provisional venue-fit assessment, or ask about a disputed premise, an important distinction, or the strongest objection. Current journal policies are not independently checked. Up to 800 characters.

Optional publication consideration

Checking editorial submission availability…

Reviews run automatically. The site owner sees your work only if you opt in here and it meets the editorial-submission criteria. Nothing is published without your separate approval.

How submission and publication work

Automated review, separate from the site owner. This review tool is an automated service running on Cloudflare, with AI-generated reports supplied through OpenRouter. The site owner provides the service but does not personally conduct or supervise individual reviews. The site owner does not read review-only manuscripts or reports. Work is sent to the site owner only if you separately opt in to submit it for possible publication and it meets the editorial-submission criteria.

Writers are welcome to submit papers for possible publication on this site, where they can be discussed and debated. Publication consideration focuses on the philosophy of mind, language, perception, and knowledge, including consciousness, meaning, and artificial intelligence; the review service itself welcomes writing across subjects.

  1. Choose review only, or opt in. The form below includes an optional, unchecked box: “If this piece meets the editorial-submission criteria, send it for possible publication.” To opt in, provide your author name and email address and separately consent to private editorial submission. Your publication choice and contact details are not sent to the AI reviewer.
  2. Receive a simulated referee report. This is AI-generated feedback, not review by human scholars. Automatic forwarding requires a report listing no blocking issues, no major issues, and no more than five minor issues. The system checks the report’s structure and issue counts; if it cannot establish eligibility, it does not submit the paper automatically.
  3. Eligible, opted-in papers are submitted privately. The exact text reviewed, its report, and your contact details are sent to the editor through the email service. The result panel tells you whether the email service accepted the submission, whether the report did not establish eligibility for automatic forwarding, or whether delivery could not be confirmed. Save your report and submission package before leaving.
  4. Publication is a separate decision. The site owner considers only work explicitly submitted for possible publication. Meeting these criteria makes a paper eligible for consideration; it does not guarantee publication or certify its accuracy. The site owner may decline it, request revisions, or invite publication. Authors need not agree with the site’s philosophical positions.
  5. Agree before publishing. Authors retain copyright. Permission to publish, the final text, byline, any accompanying review, and publication terms must be agreed separately with the author. Nothing is published automatically.

Privacy: Review-only manuscripts and reports are not stored by the application or emailed to the editor. Opted-in, eligible submissions are processed by Cloudflare and Resend and retained in the editorial mailbox and email-provider systems under their policies. They are not added to the research library or site chatbot. Submission is not newsletter signup. Please do not submit confidential material or writing you lack permission to share.

For questions, withdrawals, or an unconfirmed submission, use the contact form; include the submission reference if available.

03 Before you send

Your original file stays in your browser. Only the writing visible above and your optional context are sent, after consent, through Cloudflare and OpenRouter to the selected AI model. For normal reviews, the application does not store manuscripts or review reports, and writing is never stored in an application database or used for retrieval-augmented generation (RAG). For authors who opt in to publication consideration, your contact details, evaluated manuscript, and referee report are sent via Cloudflare and Resend to the editor's private mailbox; Resend and mailbox systems retain delivery records under their retention policies, but manuscripts are not retained in an application database or RAG system. No writing is saved in cookies or local storage. This page has no page-view analytics.

Requests use zero-data-retention (ZDR) routing. That describes a routing requirement, not an independent guarantee about provider practices or a broader promise about model training. Do not include sensitive personal information or material you are not permitted to share.

Privacy, fair use, and pilot limits

To manage the service's budget, counts-only usage notifications to the site owner record admitted reviews, completion status, publication-submission delivery status, and daily budget reservations. These emails never include your writing, report, title, contact details, or network identifier. Notification records are deleted from the application within 48 hours; count-only emails and delivery records follow the email provider's and mailbox's retention policies.

Abuse prevention stores a daily keyed hash of your network address and usage counters. Active records are scheduled for deletion within 48 hours; provider backups and infrastructure logs follow their own retention policies. This is not guaranteed anonymity. Cloudflare Turnstile performs a security check; network providers necessarily process connection information.

The pilot allows up to two reviews per day per network identifier. A shared IP address shares that quota. There is a cap on AI use per day for the project; reviews may pause when that cap is reached. Availability or an individual allocation is not guaranteed.

Save your own report before leaving; there is no account or report history.

Read the exact review prompt and user-message template in a new tab before deciding.

Long papers may take up to ten minutes to review. Do not close the page while it is running.