Peer review has always meant waiting. You finish a draft, hand it to your guide or a colleague, and the mechanical problems sit in the manuscript until someone with a week of their own work finds time to read it: a citation marker pointing at a source you deleted months ago, the same paper entered twice under two records, a sentence that claims rather more than the paper behind it does. Alfred can now sit on that reviewer list alongside the people you invited.

## What changed

- **Alfred is a reviewer, not a button.** Mark a manuscript Final, open the Reviewers panel, and add Alfred the way you add anyone else. It reads the whole manuscript and reviews it the way a reviewer does.
- **Citation integrity, record by record.** Markers pointing at a source that no longer exists, the same paper cited twice under two records, DOIs that do not resolve, a manuscript that cites sources but carries no reference list, and citation styles that drift between two forms. For a reference list you typed by hand, it also finds entries you never cited and in-text markers with no matching entry.
- **Claims read against the sources they cite.** For a claim that cites a source, Alfred reads the source and says one of three things: the source supports the claim, the claim goes further than the source does, or the source does not support it.
- **Argument and method, one section at a time.** Claims asserted without support, conclusions the evidence does not reach, contradictions between sections, design choices left unjustified, limitations that are missing, assumptions never stated.
- **The writing checks you already had.** Spelling that drifts between two forms, jargon with a plain equivalent, genuinely wordy phrasing, phrasing that reads as AI, typography, and abbreviations used before they are defined.
- **A triage panel, not a flood of comments.** Findings arrive grouped by category and by severity (must fix, should fix, consider) rather than landing in your threads. Accept the ones worth keeping and each becomes a real comment thread anchored to its passage. Dismiss the rest and the manuscript is untouched.
- **Numbered rounds you can compare.** Every run is a round, stamped with the manuscript version it read. Round two sits beside round one, so you can see what you fixed.
- **A coverage panel for what was not checked.** Alfred can only check a citation when it can reach the source: the paper is in your library, or it is open access, or its abstract is reachable. When none of those hold, the citation is listed as unchecked and Alfred produces no finding about it.
- **No seat, and no score.** Alfred does not take one of your paid reviewer seats, and the review carries no score and no verdict. It is a review to read and judge, like any other.

## Why

The expensive part of peer review is not the reading, it is the waiting, and a good deal of what comes back was mechanical all along. A supervisor's one careful pass should not be spent on a DOI that does not resolve, or on a sentence that claims more than the paper it cites. Those findings can be produced before anyone human opens the file, which leaves the reader with a draft where the only things left are the ones that need judgement.

That is also why the coverage panel exists. Alfred reviews the citations it can actually read. A paper behind a publisher's paywall, with no copy in your library and no abstract to work from, is a paper Alfred has not read, and the review says so instead of producing a confident finding about a source it never saw. This is the same discipline as the originality check, which searches open-access literature and the open web and states plainly that it does not search paywalled journal archives. A review you can trust is one that tells you where it stopped looking.

Run it before you take the draft to your guide or the journal, not instead of them.

Alfred as a reviewer is available on the Pro and Lab plans: 3 review rounds a month on Pro, 10 on Lab.