Mithila Review
Transparency

How we review: humans decide, AI assists

Mithila Review uses a human-led, AI-assisted editorial pipeline. Agents help us read a global slush pile, check science, and draft feedback — but human editors make every decision. Here is exactly how it works.

Human-ledAI-assistedRubric publicNo auto-rejection

Our commitments

  • Agents assist, humans decide. No language model ever auto-rejects a writer’s work. A human editor makes and signs every accept/reject.
  • Blind review. Identifiers are stripped before scoring; editors see anonymized work until a decision is made.
  • You can opt out. Request a human-only review and your work will not be scored by agents.
  • AI-use disclosure, both ways. We ask writers to disclose AI use, and we disclose ours.
  • Fairness audited. We periodically check agent scores against human decisions for skew by country, gender, and language, and publish aggregate results.
  • Data ethics. Submissions are private, deleted on request, and never used to train third-party models without consent.

The pipeline

  1. Intake & validation — completeness, word count, anonymization, rights and AI disclosure.
  2. Fit triage — an advisory scorecard against the rubric below.
  3. Science check — scientific claims checked against open sources (OpenAlex, Crossref, arXiv), with citations; for fiction, responsible extrapolation rather than textbook accuracy.
  4. Ethics & sensitivity read — flags for harm, stereotyping, appropriation, and extractive “poverty/inspiration porn.” Humans review every flag.
  5. Similarity check — against our archive and the open web.
  6. Human editorial review — editors read the dossier and the work, discuss, and decide; a second reader for borderline and accepted work.
  7. Decision letter — drafted from the human decision, reviewed by an editor. Rejections include constructive critique: one strength, one growth area.
  8. Publication — accepted work is published open access with a DOI and machine-readable metadata.

The rubric (advisory — humans decide)

CriterionWeightWhat it asks
Craft & voice25%Language, structure, image, character, emotional truth.
Science / idea engagement20%Meaningfully engages a scientific idea, method, or future (or rigorous worldbuilding for fantasy).
Hopeful agency20%Empowerment, resistance, compassion, constructive futures with real stakes — hopepunk, not naive optimism.
Perspective & contribution20%Adds an underrepresented/global viewpoint or a fresh take.
Accessibility15%A non-specialist can follow and feel it.

Built on free, open tools

The pipeline runs on free infrastructure — version-controlled records, event-driven compute, free-tier language models, and open fact-checking APIs — so the journal stays lightweight and the process stays inspectable. We are open-sourcing the pipeline as a contribution to responsible AI in publishing.

This page is itself a science-communication artifact: a plain-language account of how AI can assist — but never replace — human editorial judgment. Questions or appeals: .