Editorial Workflow & Modes
Our AI editorial service lets you upload a DOCX, EPUB, HTML, or MARKDOWN manuscript and select an editing style.
This page covers:
- Strict vs. Flexible Modes
- Zero-Temperature (“Deterministic”) vs. Warm Modes
- Editing Rubric Reference
- Before / After Examples
3) Quality Review — Tiers & What We Actually Check
Our editorial pipeline includes built-in heuristic verification and an optional Advanced QA layer. The Basic tier covers structural integrity and core quality guards with automatic retries on hard failures. The Advanced QA add-on adds a second, prompted LLM pass to pass/fail/patch sections and provide an LLM originality signal.
Quality Review Tiers (What to Expect)
| Tier | Goal | Price | How it works | Target quality* | What our Heuristic QA actually checks (per section) | Deliverables | Choose this if… |
|---|---|---|---|---|---|---|---|
| Basic Editorial (Single Pass + Heuristic QA) | Solid, publish-ready edit | Included | 1× AI editorial pass in your chosen edit type → automatic heuristic QA per section with up to 2 retries on hard failures. | ~80%+ | • Consecutive duplicate paragraph removal (auto-dedupe). • Non-consecutive duplicate detection (warn-only). • Fenced code blocks (```): hard error → retry. • Empty output after stripping markup/comments: hard error → retry. • Comment-contract validation for attached editor markers (must include data-start/data-end); if invalid → retry.• Length-ratio guard vs. source using word counts from HTML-stripped text: – Strict edit types (copy/line/proof/mechanical/fact-check/evaluation/sensitivity): 0.85–1.15. – Other edits: 0.70–1.30. Out-of-band → retry. • Auto-fix trivial HTML; validate HTML; malformed HTML → retry. • Safe write-back of cleaned .result.html when fixes apply.• Best-effort redline to compare source vs. result. | Edited manuscript (web preview + .docx with tracked changes or clean output), per-section QA status, notes where retries occurred, and best-effort redline artifacts. | Most manuscripts needing strong polish and formatting hygiene without a secondary LLM pass for deeper cross-section consistency. |
| Advanced QA Add-on (Double Pass) | Higher confidence & consistency | $19.99 add-on | Runs after Basic: a prompted LLM pass evaluates each section to ensure consistency, voice, pace, and tone are preserved and returns PASS / FAIL / PATCH: • FAIL ⇒ targeted re-edit of the section. • PATCH ⇒ minimal diffs applied to fix small misses/incorrect edits. • PASS ⇒ no change. Only suspect sections are re-run. Plus: an ‡LLM originality signal that asks the model to flag passages it perceives as potentially memorized/borrowed from known works; warn on low confidence; | ~90% | Everything in Basic plus model-guided pass/fail/patch on consistency, voice, pace, tone, and an LLM originality scan to surface potential memorization/borrowed text; high-confidence flags trigger hard-fail with regenerate/patch. | Everything in Basic plus a concise QA summary (what failed/patched/passed) and an originality flags report; patched sections included. | Longer works/series, voice-critical drafts, brand/technical sensitivity, or when you want stronger cross-section coherence without hiring a human editor. |
*Targets are directional, not guarantees. Source quality, length, and chosen edit type (copy, line, dev, etc.) influence outcomes. ‡ The originality signal is model-based and heuristic, not a legal determination of copyright.
Hard-fail auto-retry (both tiers): If a section fails critical checks (e.g., fenced code, empty output, malformed HTML, invalid comment contract, or out-of-band length ratios), our QA automatically requests a re-run (up to 2 times). Persistent issues are marked “completed_with_issues.”
1. Strict vs. Flexible Modes
| Mode Category | What It Means in Practice | Pick It When … |
|---|---|---|
| Strict | The AI keeps every paragraph intact—no splitting, merging, or moving things around. | You want a safe, surgical pass (e.g. Copy-edit, Proofread, Line-edit, Fact-check). |
| Flexible | The AI may merge or split paragraphs, shuffle order, or add connecting tissue if the rubric allows. | You need big-picture help (Developmental, Structural, Substantive edits). |
Quick tip: If you only need surface polish, choose a Strict mode.
For deep rewrites or re-organisation, pick a Flexible mode.
- Copy-editing / Proofreading (Strict · T=0): few or no comments; inline fixes dominate. Some models return zero comments and only tracked changes.
- Line Editing (Strict · T=0): occasional comments to justify re-phrasing or offer an alternative.
- Developmental / Structural / Substantive (Flexible · Warm): heavier commenting with rationale, re-order notes, and author queries.
- Model capability: Certain models attach DOCX comments at exact spans; others output a comment report we map to viewer side-notes (and to DOCX comments when possible).
- Safety & policy: Comments on flagged passages may be suppressed; see * *Content Safety**.
- Control: Toggle or cap comment density in Project Settings → Editor → Comments.
2. Zero-Temperature vs. Warm Modes
Some Strict modes run with the model’s temperature = 0 to guarantee the same, repeatable output every time you run them. Others allow a touch of creativity.
| Mode | Temp Setting | Rationale |
|---|---|---|
| Copy-editing, Proofreading, Line-editing, Fact-checking, Mechanical, Evaluation, Sensitivity | 0 | Prevents drift; fixes only what’s allowed by the rubric. |
| Content, Stylistic (still Strict) | Low (≈0.2) | Needs slight freedom for word-choice tweaks. |
| All Flexible modes | Medium (≈0.5) | Must explore alternate sentence/section options. |
3. Editing Rubric Reference
Below are the bullet-point instructions we feed the AI.
Each mode only does what its rubric allows.
Copy-Editing (Strict · Zero-Temp)
- Correct spelling, punctuation, basic grammar.
- Fix obvious word-choice errors.
- Never change sentence structure unless it’s a clear error.
- Preserve voice & tone. No added content.
Proofreading (Strict · Zero-Temp)
- Hunt for typos, doubled/missing words, spacing issues.
- No stylistic rewrites—only fix what’s obviously wrong.
- Flag factual errors in a comment; don’t correct them.
- Minimal, laser-focused changes.
Line Editing (Strict · Zero-Temp)
- Improve flow, clarity, cadence, concision.
- May break long sentences or combine choppy ones.
- Never add or remove story beats.
Fact-Checking (Strict · Zero-Temp)
- Verify names, dates, stats.
- If uncertain, leave the sentence and add a “⚠ Possibly incorrect” comment with a source link.
- Correct only objectively wrong facts.
Mechanical Editing (Strict · Zero-Temp)
- Enforce house style: capitalisation, serial comma, units, italics for ships, etc.
- No voice or structural changes.
- Keep fixes to the smallest possible span.
Evaluation Editing (Strict · Zero-Temp)
- Score the chapter 0-5 for clarity, style, structure, accuracy.
- Return the text untouched, followed by a short report.
Sensitivity Reading (Strict · Zero-Temp)
- Flag insensitive language or stereotypes.
- Suggest respectful alternatives in comments.
- Only replace a word when the fix is obvious (e.g., outdated slur).
Content Editing (Strict · Warm)
- Ensure content is cohesive and on-topic.
- Word-choice and tone tweaks allowed; no structural moves.
Stylistic Editing (Strict · Warm)
- Match a target style or genre (e.g., academic, YA fantasy).
- Adjust diction and rhythm accordingly, without altering plot beats.
Developmental Editing (Flexible · Warm)
- Big-picture: pacing, character arcs, theme.
- May suggest scene moves, cuts, or additions in comments.
- Rewrites sparingly, focusing on structural coherence.
Substantive Editing (Flexible · Warm)
- Blend of line-level polish and structural tweaks.
- Improve logic and transitions; add bridging sentences if needed.
- No inventing facts.
Structural Editing (Flexible · Warm)
- Re-order paragraphs or sections for logical flow.
- Insert headings/sub-headings where helpful.
- Sentence-level polish only if structure demands it.
4. Before / After Examples
All samples below are plain text—the format you see in the web preview or DOCX download.
4.1 Copy-Edit (Strict, Zero-Temp)
| Original | Edited |
|---|---|
| The quikc brown fox jumps over teh lazy dog. | The quick brown fox jumps over the lazy dog. |
4.2 Line Edit (Strict, Zero-Temp)
| Original | Edited |
|---|---|
| She walked into the room. It was big and had many chairs. The windows were open in the hot afternoon sun. | She stepped into the spacious room, rows of chairs gleaming beneath the sweltering afternoon sun streaming through open windows. |
4.3 Structural Edit (Flexible, Warm)
| Original (3 paragraphs) | Re-organised (2 paragraphs) |
|---|---|
| 1. John woke early and stretched. 2. He ate breakfast slowly, savoring each bite. 3. Then he plunged into the bustling city streets, determined to make the day count. | 1. John woke early, stretched, and savoured a leisurely breakfast. 2. Refuelled, he plunged into the bustling city streets—determined to make the day count. |
What You, the Author, Will See
- Upload your Manuscript and choose an editing mode.
- Watch your email for updates as the manuscript moves through the workflow.
- Download the revised Manuscript (with tracked changes or clean copy).
- Review comments in Word — they appear exactly where the AI left them.
Strict modes guarantee every paragraph from your upload is still there (only tweaked).
Flexible modes may merge or split paragraphs, but never lose your original ideas.
Frequently Asked Questions
Q: What AI model does Bookcicle use?
A: By default, we automatically route your manuscript to the most suitable model. Studio (Ultimate-AI) subscribers
can manually choose from any of our supported models.
Q: How long does the service take?
A: Most edits complete in under an hour. During peak demand, turnaround may extend up to 24 hours.
Q: Does AI replace a human editor?
A: No. AI accelerates the editing process and flags potential issues, but final stylistic judgment remains with
you (or your professional editor).
Q: What if my manuscript is very long?
A: We accept up to 200 000 words per job. For larger works, please contact us about splitting into multiple tasks.
Q: What file formats can I export?
A: We currently offer exports in .docx, .html, and .md formats.
Q: What happens if content violates policy?
A: Any sections that violate our AI usage policy will be returned unchanged. Please remove disallowed material
before uploading.
Q: What formatting requirements should my document meet?
A:
- Manuscripts should follow a standard editorial format (content only) to minimize processing errors.
- Maximum length: 200,000 words.
- Maximum file size: 3 MB.
- Remove title pages, outlines, images, back matter, or appendices to reduce file size.
- Use clear section breaks or chapters (e.g., Heading 1/Heading 2 in DOCX,
#/##in Markdown). - While we’ll attempt to skip front/back matter, providing a content-only manuscript ensures the best outcome. Documents without logical breaks may be rejected.
Disclaimer: AI edits can contain inaccuracies. Always review changes before publication.
Usage Limitations: Submissions must be content you can lawfully share and comply with our policy. We retain no rights to your manuscript.
No Refunds: Once editing begins, we cannot issue refunds.
✏️ Happy revising!
Let the AI handle the drudge work while you focus on the story.
🧠 Generative AI and Originality Policy
Bookcicle’s services — including Editorial, Translation, Conversion, and Ghostwriting — integrate advanced generative AI systems to enhance writing, editing, and localization. While these tools accelerate creativity, all AI-generated output is provided "as-is" and may include factual inaccuracies or material that resembles existing works.
You are responsible for reviewing, editing, and verifying any AI-assisted deliverable before publication or commercial use. Bookcicle does not guarantee originality or non-infringement of AI-generated content.
For more details, please see the full AI-Generated Content, Ghostwriting & Originality section of our Terms of Service.