Review here is AI-assisted and fully disclosed, and a named human editor reads and signs every decision. No accept or reject is ever issued autonomously by software. This page describes the process as it actually runs — we would rather be judged on a disclosed process than trusted on a vague one.
How a submission is assessed
- Intake and integrity screening. The submission is checked for completeness (missing items are named specifically, not as a generic block), and screened for integrity issues: references are verified against the scholarly record for existence and for whether they support the claims they are cited for, and files are screened for hidden instructions aimed at the review system; dedicated text-overlap and image-duplication screening will be in place before the journal's DOAJ application.
- Desk screen. Clear-cut cases — out of scope, no reproducibility package, fabricated references — are decided early, so authors are not kept waiting for a verdict the first pass already determines.
- Full assessment. A set of specialised, narrowly scoped AI assessments runs over the manuscript in parallel — covering methodological soundness, statistical rigour, simulation design, application and data quality, numerical consistency, software and code quality, writing, and the reference list. Each assessment is independent of the others and returns a structured, auditable verdict.
- Reviewer panel. Three reviewer personas — matched to the submission's topic and covering an application, a methodology and a software/reproducibility angle — each read the paper and write a referee-style report. These reviewers are openly synthetic: they carry visibly artificial names and every report is labelled as AI-generated. They first read the paper blind, then reconcile against the structured assessments, so their judgement is formed before they see anyone else's.
- Adversarial pass. A dedicated skeptic pass challenges lenient verdicts — the documented failure mode of AI reviewers is leniency, and we engineer against it.
- Reproduction. Where the article type carries runnable artifacts, the analysis is re-executed from the submitted package. No article is published before this step, and the published article states the outcome as its reproducibility badge.
- A named editor decides — and signs. The editor reads the assessments, the panel reports and the reproduction outcome, then confirms or overturns and signs the decision with their name. The human owns the decision; the pipeline drafts.
What authors receive
- Accept, minor or major revision: the complete detailed reviews — the structured assessments and the panel reports — free. This detail is what makes revisions succeed.
- Rejection (desk or after full review): a short, direct letter that develops one decisive issue in depth, concretely evidenced from the manuscript — something real to act on, not a form letter. The complete assessment report can additionally be purchased as optional feedback; buying it is entirely optional, never changes the decision, and is not offered where the rejection is for integrity reasons. Appeals are always free.
Revisions
Revised manuscripts return with a point-by-point response and are re-assessed against the prior round — the same editor re-adjudicates. At most two major-revision rounds: a paper that cannot converge in two rounds is rejected rather than kept on a treadmill.
Accountability
Every verdict is persisted in a structured, auditable record. AI-generated content in the review is always labelled as such (see the AI-use policy). Authors can contest any decision through the appeals procedure, which is heard independently of the editor who signed.