Reviewers are matched to a manuscript by an advisory tool that shortlists candidates by subject-area overlap, availability and current review load, and hard-excludes anyone with a disqualifying conflict of interest (recent co-authorship, a shared institutional affiliation, or self-nomination). The shortlist is advisory only - a human editor decides who is actually invited.
Reviewers are expected to respond to an invitation within a few days either way, disclose any conflict of interest and decline or recuse themselves where one exists, maintain strict confidentiality of the manuscript from the moment it's received until publication or conclusion of the process, complete reviews within the agreed window (a prompt heads-up if a deadline can't be met is far more useful than a missed one discovered later), and evaluate manuscripts solely on scholarly or creative merit, free from bias toward the author's identity, institution, nationality or seniority.
Manuscripts under review are confidential documents. Reviewers must not share, discuss or disclose a manuscript outside the review process, must not use unpublished manuscript content for their own research, and must never submit a manuscript to any third-party AI tool for any purpose, including drafting a review - this is a firm confidentiality rule, not a preference. It applies to a reviewer's own ad-hoc use of external tools, which places a confidential manuscript outside the journal's control entirely; it is separate from the journal's own editorial tooling, which operates under the publisher's agreements and is described in the peer-review policy. For double-blind (Track B) manuscripts, reviewers must not attempt to identify the author and should tell the editorial office rather than proceed if they believe they have.
Recommendation options are accept / suggested revisions / not accepted (Track A) or accept / minor revisions / major revisions / reject (Track B). The recommendation selected is what the editorial office actually acts on - comments alone are not read as a substitute. If acceptance depends on the author making specific changes, that is a revision recommendation, not accept with a caveat in the comments; if a concern is serious enough to mention, it should be serious enough to be reflected in the recommendation itself. Reviewers who suspect plagiarism, data fabrication, undisclosed AI-generated content or other misconduct are expected to say so explicitly and flag it to the editorial office directly, not merely lower their recommendation.
With a reviewer's consent, completed reviews can be credited to their ORCID record as verified review activity - this never identifies which manuscript was reviewed or reveals any content.