Manu­script review

The review is attached to the work, not to the title – that is where the editorial process lives anyway: how the manuscript arrived, the editor, the start and end of the review period, return postage, the internal assessment. The “AI review” tab sits in the work's detail view and is the machine-made precursor to exactly what the editorial team records there by hand today.

The second way in is the submissions list. It holds every review in the house with title and author, date received, editor, review deadline, status, recommendation, overall score, best profile, level, length and cost. Four key figures sit above it: open submissions and how many of them have no editor yet, reviews in progress and how many of those are full reports, deadlines falling within the next seven days, and what is waiting for the list meeting. An overdue review deadline shows in red in the row. A click leads into the work, not into a second detail screen – the result has exactly one home.

A review starts with four entries. The manuscript arrives as a file – PDF, DOCX, DOC, RTF or TXT, up to 200 MB – or from the media library, if the document is already held on the work. Then the publishing profiles to check against, pre-filled with the default profile. Then the scope: a quick check on a text sample or a full report on the entire text, each stating plainly what it is based on, how long it takes and what it is estimated to cost. And finally the language of analysis and your consent – a checkbox that names the provider. Without the checkbox nothing is transmitted and nothing starts. Before the run begins, the screen states the detected length in words and chapters; the cost estimate changes with every additional profile. The run works in the background with progress and a step description, and the screen may be closed.

The quick check result opens with the recommendation band: accept, consider or decline, an overall score from 0 to 100, the reasoning, and a sentence naming the profile the manuscript fits best. Below it, seven dimensions – hook and opening, style, reading appeal, tension, characters, plausibility and list fit – as stars, bars and a radar chart, each expandable with “What does this measure?” and the reasoning. Six cards summarize content and opening, setting and context, formal aspects, readership and market, quality and risks, and the selling point, right down to a pitch in seven words. The tension curve runs across the course of the manuscript, with average, peak and the share above six.

The full report extends that same record by five areas. Decision: overview, list-meeting paper and long-form report. Content and dramatic structure: summaries at four lengths – one sentence, one paragraph, one page, chapter by chapter – plot and dramatic structure with story arc and shape, the characters with role, goal, conflict, transformation and appearances, plus a manuscript bible with places, timeline, motifs, terms and a continuity check. Style and genre: extracts with quotation and explanation, a reading-appeal index, chapter analyses, slack passages and drop-off risks, plus the style X-ray. Market and fit: the ranking of profiles, reading level and genre conventions. Editorial and production: the notes, the estimated total effort and the next steps. If a single analysis step fails, the block concerned says so – the remaining values stand.

The fit is assessed in three stages. Profiles that fail on language, on the length range or on a ruled-out subject are set aside without any model run and remain visible together with the reason – which is already half of the rejection letter. The rest receive a coarse fit with a score and a brief verdict, and anyone who wants more detail requests the fine-grained fit for a single profile. It never starts by itself, because it costs the powerful model. Beneath the ranking it states how many profiles were checked and how many were excluded; profiles that were not checked are named rather than left out.

The editorial notes are the part that gets worked on: a list for notes, revisions, strengths, risks and sensitivities, sorted by priority from critical down to low, each entry with category, location in the text, quotation, reasoning, estimated effort and confidence. Every entry can be ticked off, discarded or reopened, with the person and the time recorded. A dialog moves recommendation, fit, open notes, strengths and key figures into the work's internal assessment – with a grade on the existing scale of 1 to 6, pre-filled and freely editable, with a source note and a text preview. After that, the text in the work is ordinary editorial work.

Several runs sit side by side as versions: “Request full report” extends the existing quick check, “Review again” starts a new run, with a different profile or a different model if you wish. The footer of every result states the model, the prompt version, the date of analysis and what the run cost. And every document the module produces carries the same sentence: an AI-supported decision aid that does not replace reading.