HNT: Humanistic Novel Theory

HNT (Humanistic Novel Theory) is an AI novel-writing methodology modeled on human literary creation and publishing workflows. It maps traditional publishing roles β€” publisher, editor-in-chief, writer, proofreader β€” onto a human-AI collaboration architecture, using standardized D0–D8 deliverables and G0–G5 quality gates to keep AI writing character-consistent and plot-coherent at 100,000 words per day.

δΈ­ζ–‡η‰ˆ

Core Principles

Humans as publishers

Proposition, direction, content rating, and final approval belong to humans. AI never decides what to write β€” only how to write it well.

AI as the execution layer

Writing, reviewing, beta reading, and illustration are done by specialized AI agents β€” writer, proofreader, illustrator β€” each with a defined role, handing off via standard documents.

Publishing workflow as backbone

No invented process: HNT reuses a century of publishing practice β€” proposition β†’ editorial plan β†’ writing β†’ review β†’ final check β†’ publishing β€” every step with deliverables and gates.

Three-Tier Architecture

Publisher (human)
proposition Β· approval
β†’Editor-in-chief (AI)
planning Β· final review
β†’Writer (AI)
drafting
β†’Proofreader (AI)
hard-quality review
β†’Beta reader (AI)
blind evaluation
β†’Illustrator (AI)
visuals

Agents are stateless: all handoffs go through standard documents, never verbal agreements

D0–D8 Deliverables

PhaseNameDeliverable
D0Proposition cardTheme, audience, length, content rating (frozen once locked)
D1Editorial planCreative intent, character iron laws, hard settings, causal skeleton, chapter outline, story bible
D2Writing briefChapter goals, forbidden list, anti-defect checklist
D3First draftChapter manuscript from the writer
D4Dual-track reviewProofreader hard-quality report + blind beta-reading report
D5Revision sign-offRevision list, decision log
D6VisualsCharacter art, chapter illustrations, cover (face-locked consistency)
D7Full read-throughEditor-in-chief book-level review
D8Publishing acceptancePublisher pass/fail, release record

G0–G5 Quality Gates

G0 Proposition freeze

Once the publisher approves, theme and rating are permanently frozen β€” no mid-course pivots.

G1 Editor review

Chapter-by-chapter read: scene defects, character consistency, spatial continuity. Failures go back for rework.

G2 Manuscript sign-off

Issued after dual-track review passes; recorded in the decision log, fully traceable.

G3–G5 are publishing-grade gates (visual final check, publisher acceptance, release confirmation), enabled as the work progresses

Throughput

100,000 words / day

Optimistic parallel pipeline: while one chapter is under review, the writer drafts the next; illustrators start as soon as a chapter passes. Gates never compromise β€” speed comes from parallel architecture.

HNT vs Alternatives

HNTRaw AI generationTraditional writing
Daily output100,000 wordsHigh but uncontrolledA few thousand
Long-form consistencyD1 bible + gatesCollapses over chaptersAuthor's memory
Quality reviewDual-track (proofreader + blind read)NoneHuman editor
Human rolePublisher: proposition + approvalPrompt operatorDoes everything
ReproducibilityStandardized, repeatableLuck-basedTalent-based

Live Case: Hui Nan Tian

The novel Hui Nan Tian (24 chapters across 24 solar terms, one full year) is HNT's first validation work, with 5 chapters through G2 sign-off so far.