The Scriptorium

v192

The Scriptorium

How to Use The Scriptorium

The Scriptorium is a 7-phase AI ghostwriting engine that takes you from a raw idea to a print-ready PDF novel. Each phase feeds into the next — your premise shapes your cast, your cast shapes your world, your world shapes your outline, and your outline drives the Forge. The engine now includes a real-time Manuscript Intelligence Layer that monitors narrative tension, character voice consistency, sensory balance, foreshadowing resolution, and thematic convergence throughout the entire writing process.

Phase 0: Lore Bridge

The Lore Bridge is your series continuity manager. When writing a sequel, it auto-populates with character statuses, world rules, locations, and unresolved hooks from the previous book. You can also seed it manually with your own world-building notes for a standalone novel.

Phase 1: The Spark

Set your target word count (1k–200k+), pick one or more genres, and choose a Story Vibe (e.g., Paranoid & Claustrophobic, Mythic & Ceremonial). Optionally apply a Prose Style Profile to mimic a specific author's fingerprint. Use the Babel Chaos Slider (0–100) to tune how wildly the Babel Engine scrambles genre atoms — Sane (0–30) keeps picks genre-coherent, Mixed (31–70) is the default sweet spot, and Unhinged (71–100) fires full cross-genre atom collisions. The slider is global and persisted — it also controls title generation chaos in the Bindery. Hit 'Ignite the Spark' to generate a polished logline, or 'Randomize' for a wild procedural premise via the Babel Engine.

Phase 2: Character Forge

Cast size scales automatically from your word count. The engine generates Core Cast (protagonist + antagonist + key players) and Supporting Cast (mid-level and minor characters) with full psychological profiles, MBTI types, motivations, flaws, and Visual DNA blueprints. You can pre-seed specific characters, reroll names with a per-project ban list, and generate AI portrait images.

Phase 3: Worldbuilder

The Iceberg model builds your story's thematic depth across 4 layers: Surface (the plot readers see), Shallows (the emotional subtext), Deep (the philosophical argument), and Abyss (the mythic or unconscious resonance). The engine also extracts Forbidden Motifs — recurring thematic symbols that are budgeted and rotated to prevent repetition across the manuscript.

Phase 4: Architect

First, draft a 7-Point Story Skeleton (Hook, Catalyst, Midpoint, Dark Night, Climax, Resolution, Epilogue). Then expand it into a full Master Outline via the Fractal Manager/Worker pipeline — each chapter is fractured into 3 narrative sequences, then each sequence is expanded into 5+ micro-beats. Before advancing to The Forge, the engine asks you to confirm 3 thematic statements derived from your Iceberg. These become the anchor for the Thematic Convergence system, which audits the manuscript at the midpoint and finale to ensure the story delivers on its philosophical promise.

Phase 5: The Forge

The Forge is an autonomous ghostwriter loop. It reads the next beat from your outline, writes a scene, runs it through the AI Prose Auditor (double-pass cleaning), then advances to the next beat. Core mandates: Sensory Rotation (cycles sight/sound/smell/touch/taste with a 1,000-word lockout), Trajectory Enforcement (every 3rd beat must advance plot), Plot Mechanic Seeding, Emotional Opacity Check, and Antagonist Dimensionality Mandate. Use the Live Steering Dials to adjust Tone, Action, Stakes, and Pacing on the fly. The Intelligence Layer runs silently in the background: S1 tags foreshadowing anchors for later verification, S2 monitors character voice drift every N chapters, S3 logs tension self-scores per beat and alerts on plateaus or cliff-drops, S4 checks thematic convergence at the midpoint, S5 tracks sensory channel distribution per chapter. All thresholds are configurable in the Intelligence Layer Controls panel. Alerts appear as non-blocking toasts so they never interrupt the writing loop. For performance, the Beat Decomposer (Option B) automatically detects oversized beats (default threshold: 600 words) and splits them into 3 ordered micro-beats before writing — eliminating the retry cascade that caused 24-hour generation times on 200k novels. The Adaptive Timeout (Option A) calculates a per-call timeout based on the beat's word budget instead of using a fixed value. Both systems are configurable in the Intelligence Layer Controls panel.

Phase 6: The Bindery

The Bindery is your publishing studio. Auto-generate a book title, back-cover blurb and full-bleed AI cover art. Toggle the World Glossary to auto-compile a glossary from your Iceberg and character data. Toggle the Pronunciation Guide to auto-generate phonetic breakdowns (stressed-syllable notation, IPA, and rhyme hints) for every character, place, faction, creature, and lore term in your story — grouped by category across multiple pages. The Pronunciation Guide also feeds the 10-method Slang Engine, which uses phonetic irony, metonymy, clipping, semantic inversion, portmanteau, euphemism drift, shibboleth, onomatopoeia, ironic literalism, and generational drift to generate universe-native slang. Slang comprehension degrades with character age: teens (13-17) understand 100%, adults (31-45) catch about 40%, and anyone 76+ will assume it was an insult. Run the Red Pen Audit for a professional literary critique. Hit Compile PDF to produce a KDP-ready trade paperback with the full back matter: Glossary, Pronunciation Guide, and Slang Appendix with comprehension matrix.

Writing a Sequel

In the Library, find any completed project and click 'Write Sequel'. This creates a new linked project and runs the Lore Engine to scan the finished manuscript — extracting character fates, world-state changes, and unresolved hooks — then drops you into Phase 0 to review and edit the Continuity Bible before starting Book 2. For FFP users, click 'Smart Expand' instead to trigger the full Smart Expand pipeline before writing the sequel.

v18.5 — Gap Guard & Plot Hole Prevention

Implemented the Gap Guard v18.5 to prevent silent generation failures. The Forge loop now audits every beat's word count after deduplication; if a beat results in fewer than 50 words, the engine flags it as a failure and triggers an auto-retry or pause. Added the Dedup Safety Valve, capping sentence stripping at 50% to ensure core content is never lost during seam-smoothing.

v18.0 — Paginated Chapter Mapping

Full rewrite of the chapter mapping architecture. Replaced single-call JSON generation with a paginated batch system (15 chapters per call). This eliminates model truncation on large (200k+) novels and guarantees accurate chapter counts for massive sagas. Replaced hard 4,500w caps with flexible soft bounds (2,000–5,500w) for organic narrative expansion.

v17.0 — The Relationship Web

Introduced a machine-readable social graph engine. Relationships now exist as structured objects with type, label, heat, and status. Canon relationships are extracted from the premise in Phase 1.5, and the Architect and Forge now receive the full social web context in every prompt. Includes the Incest Prevention Guard (v17.1) to block romantic links between family-tagged pairs at the structural, prompt, and safety-validator levels.

v16.0 — Identity Anchor System & Phase 1.5

Introduced the Identity Anchor System — a name-continuity engine that extracts character names from the premise, detects remake/Ship-of-Theseus tropes, and injects a canonical Identity Context Block into every AI call. Phase 1.5 'Who's In Your Story?' is a new confirmation gate between Phase 1 and Phase 2. Zone A: Detected Characters — AI-extracted names with role, gender, and status flags, all editable. Zone B: Remake Detection Banner — auto-detects clone, amnesia, android, resurrection, time loop, and possession tropes from premise text. Zone C: Reconvergence Arc Panel — configures Ship-of-Theseus identity splits with trope type, grief_resonance flag, coexistence toggle, and Soft Reset 1.5x heat timeline. Zone D: Add Character Manually. The Identity Context Block travels with every AI call — character gen, skeleton, outline, chapter expansion, and forge beats — ensuring canonical names are never replaced and Reconvergence instances are tracked correctly in prose.

v15.0 — Young Adult Mode

Introduced YA Mode — a top-level narrative mode that reconfigures the entire forge pipeline for Young Adult fiction (ages 12–18). Toggled in Phase 1 alongside Heat Mode. CORE CONSTRAINTS: Heat hard-locked to Sweet (system firewall, not a UI preference). Word count capped at 75,000. 7 genres greyed and unselectable (Adult Romance, Noir, Cyberpunk, Steampunk, Western, Apocalypse, Disaster). Series Architecture toggle: Standalone vs. Series Book 1, with different ending-structure rules injected into every prompt. ROMANCE CEILING: Everything up to the bedroom door. Passionate kissing (including making out), racing pulse, breathlessness, charged physical tension, characters noticing each other in non-explicit terms — all allowed. The line is: sexual acts are not described. What happens is implied, not written. Fade to black. NOT allowed: explicit sexual mechanics, graphic nudity in sexual context, adult erotica language. Note: the initial build incorrectly set this at 'closed-mouth kiss only.' That was overcorrection. Corrected in the same session. VIOLENCE & DARK THEMES: Death on the page with consequence (Rue in Hunger Games, every page of Six of Crows): YES. Fight scenes with real blood: YES. Torture implied or briefly shown: YES. Depression, suicidal ideation, abuse, addiction handled honestly: YES — YA tackles these more directly than adult literary fiction. Mild-to-moderate profanity: YES. Drugs and alcohol with narrative consequence: YES. Gratuitous gore for shock value, torture porn, wall-to-wall profanity as style: NO. Rule: darkness is allowed. Gratuitousness without narrative purpose is not. FORGE PROMPT INJECTIONS: Age-Authentic Voice Mandate (16-year-olds think in fragments, contradictions, hyperbole — not philosophy), Dual Consciousness Narrative (public face vs. inner monologue in every scene), Hope Covenant (protagonist must have done something that mattered by the end — non-negotiable), Villain Mirror Mandate (antagonist = dark reflection of protagonist's potential, the road not taken), Adult Character Mandate (peripheral, flawed, mentor max age 25–35), Zipf Tier 8,000 vocabulary ceiling. YA SLANG COMPREHENSION MATRIX (age-shifted -5 years): Ages 12–17: 100% (they invented it). Ages 18–22: 90% (fluent). Ages 23–28: 65% (gets the gist). Ages 29–38: 35% (partial). Ages 39–50: 15% (lost). Ages 51–65: 5% (completely lost). Ages 66+: 0% (Elvish. Cane-shaking territory). ARCHITECT INJECTIONS: 10 Milestone Moment types auto-seeded (first betrayal, first time an adult is wrong, first impossible choice, etc.). Villain age 17–25 (peer, not distant adult). Mentor arc must include the mentor being wrong about something important. Chapter count 20–28. CHARACTER INJECTIONS: Protagonist 14–18. Cast majority 14–20. Mentor 25–35 max. Emotional/romantic profiles generated (attraction, what holds them back) but no explicit heat-level sexual profiles. UI: Green toggle card in Phase 1. Active banner with 6 constraint badges (Romance: fade-to-black, Word cap: 75k, Vocab: Zipf 8k, Age-authentic voice, Hope covenant, Villain mirror). Series Architecture buttons embedded in banner. Vibe warning if Dark/Edgy or Elegiac vibes are active.

v14.1 — The Word Count Calibration Recalibration

Diagnosed and fixed five compounding bugs causing a 28k explicit-heat novel to complete at 76% of target (~21.4k words). Bug 1 (Key Mismatch, CRITICAL): forge back-pressure and _checkWordDebt() both read state.targetWords, but the state schema only stored targetWordCount. Both systems fell back to 50000, making a 28k run appear to be a 50k run. Fix: targetWords added as canonical alias in getDefaultState(); StateManager.update() now auto-syncs both keys; loadProject() syncs on load. Bug 2 (Dead currentWordCount): _applyBackPressure() reads state.currentWordCount but Phase5UI never wrote it during the forge loop. Back-pressure always saw 0 words written. Fix: Phase5UI now writes currentWordCount and totalBeats to state on every beat completion. Bug 3 (Back-pressure too late/gentle): fired at 70% progress, 10% threshold, 60/40 blend. Fix: fires at 50% progress, 7% threshold, 30/70 blend (stronger correction). Bug 4 (expandChapterChunk no word budgets): the Free/Mid expansion path returned beats without [WORDS:N] tags; Forge defaulted to 600w/beat. Fix: expandChapterChunk now accepts chapterWordTargets map and injects [WORDS:N] into all beats. phase4-ui._expandChunks() builds and passes the map. Bug 5 (Wrong explicit heat prior): HEAT_PRIORS.explicit was 1.18 (assumed AI writes verbose). AI actually writes lean for explicit scenes. Fix: prior changed to 0.95. HEAT_SCENE_AVG_WORDS.explicit reduced from 1000 to 700 (AI averages 600-700w, not 1000). Word debt check fires every 5 beats (was 10) with 10% threshold (was 15%), boosting next 20 beats (was 10).

v14.0 — Hybrid Word Count Calibration

Replaced the flat 750w/beat projection with a Bayesian self-calibrating system. Ships with hardcoded style/vibe/heat/genre priors (e.g. Donna Tartt lush = 1.10×). Every completed novel deposits a calibration receipt. The system blends priors with observed data — 0 receipts = 100% prior, 12+ receipts = ~96% self-calibrated. Heat scene word budget is now subtracted from the beat pool before chapter mapping, so heat counts toward the external word target. The projection display shows only the external target (never the internal adjusted number). Back-pressure rebalancing kicks in at 70% completion if running >10% off target.

Word Count Calibration Engine

v13.4.4 — Character Name Consistency Fix

Fixed AI hallucinating character names in skeletons and outlines (e.g., inventing 'Riley Thorne' instead of using canonical 'Dante'). Root cause: character registry was buried mid-prompt with no enforcement mechanism. Two-pronged fix: (1) Every Architect prompt now opens with a === CHARACTER LOCK LIST === mandating ONLY canonical characters may appear, with full role and description context to anchor the AI. Non-compliance triggers a regeneration threat. (2) Post-generation _normalizeNamesInSkeleton() uses Levenshtein distance matching to detect and replace hallucinated names with canonical counterparts automatically after every generation step (skeleton, fracture, sequence expansion, chapter chunk).

v13.4 — The Global Babel Chaos Slider

Introduced a global, persisted Babel Chaos Slider (range 0–100, default 65) that gives users direct control over how wildly the Babel Engine scrambles genre atoms during premise randomization and title generation. The slider appears in two places: Phase 1 (The Spark), directly below the Randomize Idea button, and Phase 6 (The Bindery), below the Book Title input. Both sliders are wired to the same babelChaosLevel state key — moving either one syncs all others instantly. The value is persisted across sessions via miniappsAI.storage. Mechanically, 0–30 activates biased picks only (genre-filtered atom selection, coherent combinations), 31–70 is the mixed sweet spot (some filtering, some wildness — the previous hardcoded 65% behavior), and 71–100 is pure chaos (no filtering, full cross-genre atom collisions, maximum weirdness). The previous hardcoded BABEL_CHAOS constant of 65% has been replaced by a live read from state on every Randomize and Auto-Generate Title call. The slider UI uses a cyan→purple→pink gradient track and a glowing white thumb that shifts from purple to pink on hover, matching the app's aesthetic.

v13.3 — The Smart Expand & Surgical Rewrite Update

Introduced the Smart Expand pipeline — a multi-agent system that converts a Red Pen Audit into a fully revised manuscript before the next Forge run. The pipeline has five stages. Stage 1: if no Red Pen data exists, the audit runs automatically (~60 seconds) before anything else begins. Stage 2: Grok 4.1 Fast Reasoning translates all audit scores and critique notes into 5-8 concrete Forge directives, which are pre-loaded into the Author Correction Cache so every future beat is written with the fixes already active. Stage 3: Grok 4.1 Fast scores every chapter 1-10 across all Red Pen metrics and flags any chapter scoring below 5 for surgical rewrite. Stage 4: DeepSeek V3.2 Speciale re-runs the Architect with the new word target (Option C hybrid — settings preserved, no Iceberg wipe, no cast regeneration). Stage 5 (FFP only): the Surgical Rewrite Queue processes each flagged chapter through a three-pass loop. Pass 1 — Kimi K2.5 reads the full chapter plus all Red Pen directives and produces a JSON rewrite plan identifying which beats need work and why. Kimi decides the scope itself — it will not nuke a chapter that only has one weak paragraph. Pass 2 — beat-by-beat rewrite loop: the chapter is split into paragraph-level beats, flagged beats are rewritten one at a time by Kimi K2.5 with a 150-word rolling Context Diet cache, unflagged beats are kept verbatim at zero cost. BeatDecomposer applies — heavy beats (>600 words) are split into micro-beats before rewriting so Kimi never writes more than one beat per call. AdaptiveTimeout scales per beat. Pass 3 — Gemma 4 31B smooths the joins between original and rewritten beats. All rewrites are stored as a separate Revised Draft layer — the original manuscript is never touched. The Bindery gains a Revised Draft Banner, a side-by-side diff viewer, and a PDF source toggle to choose Original or Revised at compile time.

v13.2 — The Expand to Novel Fix Pack

Diagnosed and fixed five compounding bugs that degraded novel quality on Expand to Novel runs. Fix 1 (Word Count Shortfall): the GUTENBERG_SAFETY_FACTOR was being applied twice — once in the beat budget calculation and once in the word target — causing sparse-style novels (McCarthy, Tartt) to generate at ~60% of target. The double-application is now removed. Fix 2 (Settings Wipe): expandToNovel() was calling a full state reset before re-running the Architect, wiping genre, vibe, style profile, and cast data. The function now performs a targeted word-target update only, preserving all creative settings. Fix 3 (Iceberg Regeneration): the expand flow was re-triggering the Worldbuilder, generating a new Iceberg that could contradict the existing manuscript. The flow now jumps directly to the Architect. Fix 4 (Resume State): the Forge loop was not correctly reading the last completed beat index after an expand, causing it to restart from beat 0 instead of resuming. The beat cursor is now correctly restored from state. Fix 5 (Word Count Display): the Library card was showing the pre-expand word count until the next manual save. The display now updates immediately after expand completes.

v13.1 — The Performance Restoration Update

Diagnosed and eliminated the retry cascade that caused 200k novel generation to regress from 2-3 hours to 24+ hours. Root cause: a single fixed timeout for all beats regardless of complexity, combined with growing beat sizes from the Fractal Manager pipeline. Three systems deployed simultaneously. Option D (Context Diet): rolling prose context truncated from 300 to 150 words per beat call, reducing payload size and latency on every single forge call. Option A (Adaptive Timeout): timeout is now calculated dynamically as baseline + (estimatedWordTarget × 140ms), capped at 180 seconds. A 100-word beat gets 39 seconds; a 500-word beat gets 95 seconds. No more premature cuts on complex scenes, no more 4-minute waits on simple ones. Option B (Beat Decomposer): before any beat reaches the forge, it is classified by estimated word target. Beats exceeding the threshold (default 600 words) are sent to a fast flash-class model (Gemini 3.0 Flash Preview in FFP) which splits them into 3 ordered micro-beats with their own word budgets. Each micro-beat is written independently and stitched together. Optional seam-smoothing pass available. All three thresholds are configurable in the Intelligence Layer Controls panel. Decomposer activity is logged and displayed as live stats.

v13.0 — The Manuscript Intelligence Layer

Added a six-system quality monitoring architecture that runs silently throughout the entire writing loop. S1 (Foreshadowing Tracker): the ghostwriter tags depth events with [DEPTH_EVENT] markers during generation; a verification pass at manuscript completion checks that all planted anchors were resolved and surfaces orphaned ones. S2 (Voice Drift Monitor): extracts a lexical and syntactic fingerprint for each named character on first appearance; audits every N chapters (configurable) and flags drift with severity ratings. S3 (Tension Gradient): the ghostwriter self-scores each beat 1-10 with a [TENSION:N] tag; the system checks for plateaus (5+ beats within ±1 point), cliff drops (>4 point fall in 3 beats), ceiling lock (5+ beats at 9-10), and floor lock (5+ beats at 1-2). S4 (Thematic Convergence): author confirms 3 thematic statements at the end of Phase 4; the system audits at the midpoint and manuscript completion to verify the story is delivering on its philosophical promise. S5 (Sensory Budget): zero-cost keyword classification per beat into 6 channels (visual, auditory, olfactory, gustatory, tactile, proprioceptive); chapter-boundary alerts when any channel dominates above threshold. S6 (Health Report): aggregates all five systems into a composite score out of 100, tension gradient visualization, sensory distribution chart, orphaned anchor list, voice drift log, dormant theme list, learning directives, and Stylistic Bible aphorisms. All thresholds configurable. All alerts are non-blocking toasts that never interrupt the writing loop. Tags are stripped from the final manuscript before display.

v12.4 — The Archival & Navigation Refactor

Completely rebuilt the Library Ledger system to automatically detect and repair index corruption using a self-healing storage fallback. Introduced non-destructive Project Forging (bypassing the old state wipe constraint) and persistent Red Pen back-navigation via state-preserving overlays. Implemented 'The Basement' archival tier to cleanly manage legacy manuscripts without polluting the Active Workspace.

v12.3 — Publication Ready (KDP Trade Paperback)

Full bleed cover art (zero-margin first page), industry-standard trade paperback margins (0.75 inside / 0.5 outside), automatic Chapter TOC with placeholder page numbering, smart sanitized PDF naming (Title_by_Author), and optional AI-generated world glossary.

v12.2 — The Deep Scan Integration

Addressed prose-to-plot ratio imbalance and late-introduced plot mechanics identified by Deep Scan analysis of a 250k adversarial stress-test. Added Trajectory Enforcement Mandate (every 3rd beat must advance plot), Plot Mechanic Seeding System (auto-seeds climax mechanics into early chapters), Hybrid Vocabulary Rotation (budget-tracked thematic word exhaustion with automatic substitution), Emotional Opacity Check (forces precise emotional naming per chapter), and Antagonist Dimensionality Mandate (ensures sympathetic internal logic for antagonists).

v12.1 — The Free-for-Pro Fractal Hybrid Update

Launched the multi-tier Fractal Manager/Worker pipeline. For FFP users, the engine now fractures chapters into narrative sequences before expansion, ensuring logic-dense outlines with 15+ beats per chapter. This hierarchy prevents 'summary-happiness' and maintains high-resolution pacing from the first page to the last.

v12.0 — The Atomic & Fractal Expansion Update

Refactored the Architect Engine to handle extreme word counts (200k+) without timeouts. Implemented the Atomic Chapter Loop to eliminate contextual drag. Introduced Vibe-Based Density Throttling, automatically adjusting processing resolution based on narrative complexity (Paranoid, Mythic, and Elegiac modes).

v11.1 - The 'Unconventionally Complimentary' Babel Update

Refined the Babel Engine to eliminate 'The Paradox Trope.' Replaced literal disabilities and logical dead-ends (like hemophobic surgeons) with specialized, functional methodologies (like forensic echo-analysts). This ensures character hooks feel unconventionally skilled rather than jokingly incapable.

v11.0 - The 'Library of Babel' Procedural Scrambler Update

Implemented the Babel Engine: a procedural scrambler that forces AI models to synthesize high-entropy 'Conceptual Atoms' (Discordant Archetypes, Bizarre Locations, and Conceptual Curses). Added a hard-coded 'Anti-Trope Mandate' to purge the engine of Archivists, Forgers, and dusty libraries. Introduced Tension-Based Synthesis, requiring the engine to justify the relationship between conflicting plot atoms before generating the final premise.

v10.1 - The Anatomical Fixation & Name Consistency Update

Implemented the Character Physical Motif Lockout system to prevent obsessive repetition of character physical descriptions. Each character's physical details are now tracked in a 3-tier system: Signature (once per 10 chapters focus, once per 3 chapters background), Secondary (once per chapter background only), and Functional (exempt only during active plot use). Added the Entity Lock Protocol for name consistency: canonical character names are now injected into all Bindery API calls, and a post-generation Levenshtein distance validator catches name corruption before text reaches the manuscript. Fixed Library button non-responsiveness caused by escaped HTML quotes.

v10.0 - The Forbidden Motif & Thematic Integrity Update

Implemented the Forbidden Motif Rule to prevent thematic tics. Motifs are now detected in Phase 3 and strictly budgeted: 1 primary focus per 5 chapters and 2 background appearances per chapter. Added 'Thematic Substitution' logic to force stylistic variety when budgets are hit. Fully integrated the Free-for-Pro high-reasoning stack (Grok 4.1, RolePlay v5, MiMo-V2-Omni, Gemma 4) for logic-dense architectural planning and prose execution.

v9.4 - The Global Mosaic & Grit Mandate

Eroded demographic homogeneity by implementing the Global Mosaic protocol. Severed the ethnicity-link between protagonist and secondary cast. Mandated 'Ungramorous Anchors' (physical flaws and regional grit) for every character to prevent generic perfection. Upgraded the Magic Wand (⚡) with a Contrast Ratio check to ensure new descriptions diversify the existing cast.

v9.3 - The Visual DNA & Diversity Update

Implemented the Visual DNA Blueprint system for permanent physical identity. Integrated Vision AI for photo-to-text character profiling. Added a global Diversity Mandate to character generation prompts to ensure varied ethnicities, ages, and body types. Synced Visual DNA with the Forge writing loop to maintain descriptive continuity across the manuscript.

v9.2 - The Sensory Engineering & Rotation Update

Transitioned from mathematical sensory budgeting to a dynamic Channel Rotation system (Sight, Sound, Smell, Touch, Taste). Implemented a 1,000-word lockout for repetitive sensations (curing the 'copper taste' infection) and established the 'Ungramorous Anchor' as a foundational mandatory detail for all new scenes. Recalibrated Red Pen heuristics to protect sensory immersion while aggressively targeting purple adjective-stacking.

v9.0 - The Chronos & Cosmos Update

Introduced the Lore Engine for deep manuscript scanning and fact extraction. Added the Phase 0 Lore Bridge for series continuity management. Implemented Just-In-Time (JIT) lore injection in the Forge writing loop to ensure Book 2 never forgets the rules established in Book 1.

v8.0 - The Evolutionary Engine & Red Pen Audit

Introduced the multi-stage Evolutionary Loop. Stage 1 (Prose Governor) implements a 0.75 Gutenberg safety factor and sensory budgets to eliminate wordiness. Stage 2 (Adaptive Pulse) triggers background audits every 10k words to self-correct stylistic decay. Stage 3 (Red Pen Post-Mortem) adds a professional literary audit tool in the Bindery. Stage 4 (Stylistic Bible) allows the engine to learn from its mistakes across different projects using persistent storage.

v7.2 - The High-Density Masterpiece Overhaul

Total recalibration for professional-grade 200k+ novels. Implemented the GUTENBERG_SAFETY_FACTOR (0.9) to manage AI wordiness. Replaced linear chapter scaling with a density curve, targeting ~3,600 words per chapter. High-density beat allocation (7-10 beats for climaxes) ensures depth over volume. Increased rolling context memory to 300 words for superior scene continuity.

v7.1 - The Hybrid Engine Overhaul

Introduced the Director/Worker Bee Hybrid model. Bulk expansion tasks are now routed to high-speed Flash-class models (Gemini 3.0 Flash) with large 6-chapter chunking, while core logic remains on Reasoning-class models. This reduced expansion time for 200k novels from 120+ minutes to under 10 minutes while maintaining rolling memory consistency.

v6.3 - The KDP Print Typesetter

Professional trade paperback CSS, automated front matter, manuscript typography, drop caps, and scene break ornaments.

v6.0 - The Pro Unlimited & Vibe Overhaul

Pro Unlimited tier, Global Vibe steering, per-project name ban list, Red Pen auditor, and Stylistic Bible persistence.

The Lore Bridge

Sequel — Unlock Names from Previous Projects

Characters + Add

Locations + Add

World Rules

Unresolved Hooks

The Spark

Who's In Your Story?

Detected Characters

Relationship Web

Identity Context Block Preview

The Character Forge

Core Cast

Supporting Cast (Mid-Level & Minor)

Universe Slang Glossary

The Worldbuilder

Thematic Anchors (Forbidden Motifs)

The Architect

The Forge

Smart Expand Pipeline

Professional Audit: The Red Pen Report Card

The Bindery