Edit Module
Most authors don’t have a professional editor on retainer. They’re doing their own editing between drafts, with maybe a beta reader and a freelance pass near the end if the budget allows. The problem is that real editing is a structured process — a developmental editor reads for structure first, then a line editor reads for prose quality, then a copy editor catches grammar and consistency, then a proofreader catches the last typos — and most authors working alone collapse all four jobs into a single “I’ll just read through and fix what I notice” pass that misses things at every layer. The Edit module is Ishvana’s attempt to simulate the structured process. Five phases, each addressing a different layer of the manuscript, run in order, with the agents handling the heavy lifting on the phases where a model is actually good at the job. The module won’t replace a real editor for a book you’re trying to publish professionally, but it will catch the kind of thing a tired author misses on solo passes, and it gets you measurably closer to a clean manuscript before you hand it off to a human editor for the work a human does best.
The phases unlock sequentially. You have to complete each one before the next unlocks, which sounds annoying but is actually the whole point — the sequential structure is what keeps the module from being “five unrelated tools in a tab.” That said, each phase can also be run independently if you know you only need one pass.
The five phases
Section titled “The five phases”| Phase | Type | What it does |
|---|---|---|
| Baseline | Automatic | Captures starting word count, paragraph count, sentence count, chapter detection |
| Developmental Edit | Agent | Big-picture feedback on pacing, character, plot, tone |
| Line Edit | Agent + Rules | Paragraph-level prose quality with ProseGuard |
| Copy Edit | Agent | Precise text suggestions — accept, reject, modify |
| Proofread | Rules only | Consistency scan for variants, punctuation, and capitalization |
-
Baseline analysis.
An automatic scan that establishes your manuscript’s starting metrics: word count, paragraph count, sentence count, and chapter detection from headings. No model, no rules, just counts. The baseline exists so later phases can report “this pass changed 47 paragraphs” or “word count dropped by 200” — numbers mean more when you have a starting point.
Baseline takes a few seconds and completes silently. You don’t interact with it — it just runs when you open the Edit workspace for the first time on a manuscript.
-
Developmental Edit. Agent
Hawken reads your manuscript as a first-time reader and produces structural feedback through a two-pass process. This is the most model-heavy phase in the module, and it’s where the biggest gap between “what an author catches on solo re-read” and “what a real developmental editor catches” usually shows up.
Pass 1: First read. For each chapter, Hawken produces a short report covering tone observations, pacing notes, character notes, plot notes, strengths, and concerns. This is the equivalent of a careful reader writing their impressions after finishing each chapter.
Pass 2: Margin comments. Hawken goes back through the prose and marks specific passages with typed comments. Each comment has a type (Expand, Cut, Rework, Contradiction, Pacing, Character, Plot, General) and a severity (critical, major, minor, suggestion). You get a concrete list of “here are specific sentences and paragraphs that need attention, with a note explaining why.”
Editorial letter. After both passes, Hawken writes a full editorial letter covering overall assessment, structure, character analysis, pace and tension, theme and voice, and a ranked list of top priorities. This is the document-level summary — the kind of thing a freelance developmental editor would charge $500-2000 to produce for a novel.
Before any of this starts, the system automatically takes a snapshot of your manuscript so the pre-edit version is always recoverable. Nothing the module does is destructive.
-
Line Edit.
A split-pane editing view — the editor on the left, the issue list on the right. Two rule engines run here:
- ProseGuard for rule-based prose quality (banned phrases, POV consistency, Machine Tells, filter words, em-dash limits).
- Lore consistency checks for places where your prose contradicts your Legendry.
Issues are marked in the editor with wavy underlines. Click an issue in the list and the editor scrolls to it. Fix it or mark it as acknowledged or false positive. Batch size is configurable so you can work through manageable chunks instead of being overwhelmed by a single massive issue list on a long chapter.
Line Edit is the phase where you’re working at the paragraph and sentence level. It’s structured the way a real line editor would work — one issue at a time, fix or flag, move on.
-
Copy Edit. Agent
The core copy edit pipeline. This is the phase that produces precise text suggestions — specific “replace this with that” recommendations that you accept, reject, or modify. Three passes under the hood:
Pass 1: Research (optional). Gathers context from your project lore and optionally from external research sources. The research pass exists because a copy editor working on historical fiction might need to check facts against real-world references, and Hawken’s research integration handles that when needed. Most fiction doesn’t need this pass and can skip it.
Pass 2: Ingestion. Hawken reads each document and produces reading notes — tone, character voice, pacing analysis, narrative thread identification, and arc tracking across documents. This is the model building a mental map of the manuscript before making any suggestions, so the suggestions that follow are grounded in the whole book rather than one chapter in isolation.
Pass 3: Editing. Hawken generates specific suggestions. Each suggestion has a precise location (paragraph, character offset), the original text, the suggested replacement, a rationale, a type (grammar, phrasing, consistency, style guide, tone, pacing, dialogue), and a severity (error, warning, suggestion).
Review and apply. Suggestions appear in the Copy Edit panel. For each one, you can accept, reject, modify, or bulk-operate across the whole set. Bulk accept-all is dangerous on a long manuscript — review individually unless you really trust the output.
Style guides. A style guide is a set of project-specific rules that get injected into the copy edit process. Ishvana ships with Chicago Manual of Style and AP Style presets, and you can create custom per-project guides — “my project uses single quotes for internal thoughts,” “never hyphenate ‘email’,” “capitalize ‘the Legendry’ as a proper noun,” whatever rules matter for your book.
-
Proofread.
A rule-based consistency scan with zero model involvement. Deterministic — same input always produces the same output. Catches the specific class of issues that models are bad at and humans are inconsistent at:
- Spelling variants. 13 common pairs like grey/gray, color/colour, judgment/judgement.
- Punctuation inconsistencies. Em-dash spacing, ellipsis style, quote style (straight vs. curly).
- Capitalization inconsistencies. “The Kingdom” vs. “the kingdom,” “Magic System” vs. “magic system,” etc.
For each inconsistency detected, you pick your preferred variant and corrections are applied as global find-and-replace operations. This is the last-pass polish that turns an almost-finished manuscript into something that reads as if a professional copy editor touched it.
Publishing readiness
Section titled “Publishing readiness”A composite score from four sources tells you how close your manuscript is to publication. The score is a rough diagnostic, not a grade — 95% isn’t a guarantee of publishability, and 60% isn’t automatic disqualification — but it’s a useful signal for “am I ready to ship this, or do I have more work to do?”
| Source | Default weight | What it measures |
|---|---|---|
| ProseGuard | 25% | Rule-based writing quality — how many violations remain |
| Lorekeeper | 25% | Lore accuracy — how many unresolved consistency issues remain |
| ML Health | 20% | Lore ML pipeline health — coverage, maturity, validation scores |
| Edit Completion | 30% | How many edit module suggestions were addressed vs. ignored |
Weights are configurable per-project. Full transparency is the point — you see the score breakdown, which categories were checked, and per-category recommendations for what to fix next.
The readiness score feeds into the Bookmaker’s publishing gate. By default, Bookmaker will warn you if you try to compile a manuscript with a low readiness score — not to block the compile, just to make sure you know what you’re shipping.
The workflow, in practice
Section titled “The workflow, in practice”A typical pass through the Edit module for a finished draft:
- Open the Edit workspace. Baseline runs automatically.
- Developmental Edit — read the editorial letter Hawken produces. Take a week if you need to. Make structural changes in the editor based on the feedback.
- Return to Edit module. Line Edit — work through ProseGuard and lore issues one chapter at a time. Most issues resolve in seconds per fix.
- Copy Edit — run the pipeline, review suggestions. Accept the obviously-correct ones, reject the obviously-wrong ones, modify the ambiguous ones.
- Proofread — run the consistency scan, pick preferred variants for each flagged pair.
- Check the publishing readiness score. If you’re below where you want to be, the breakdown tells you which category needs more work and you go back to that phase.
- Move to Bookmaker to compile the manuscript.
The whole pipeline for a 90,000-word novel takes somewhere between a day and a week depending on how much structural work the developmental edit flags. Most of the time is in the developmental and line edit phases because those are where the real decisions happen. Copy edit and proofread usually run in an hour or two each.