Raw AI generation is single-prompt output: characters collapse and plots derail over long chapters. HNT borrows the human publishing workflow — publisher propositions, editor-in-chief plans, writer drafts, proofreader reviews, beta reader blind-evaluates — every step with standard deliverables and quality gates. AI executes; humans direct and approve.
Speed comes from an optimistic parallel pipeline (the next chapter is drafted while the current one is under review); quality comes from uncompromising G0–G5 gates. Every chapter passes proofreader hard-quality review and blind beta reading; failures go back for rework, with a decision log tracing everything.
D0–D8 are nine standardized deliverables: proposition card, editorial plan, writing brief, first draft, dual-track review, revision sign-off, visuals, full read-through, publishing acceptance. G0–G5 are quality gates: G0 proposition freeze, G1 chapter-by-chapter editor review, G2 manuscript sign-off; G3–G5 are publishing-grade gates.
Humans serve as publisher: proposition (theme, audience, length, rating), direction, final approval. All day-to-day execution is done by specialized AI agents. Humans only answer pass/fail questions — never execution ones.
Yes. Hui Nan Tian is HNT's first validation work: 24 chapters across 24 solar terms, from worldbuilding and character dossiers to drafting, review, and revision — all executed under HNT. Five chapters have passed G2 sign-off so far.
Yes. HNT's D0–D8/G0–G5 framework is genre-agnostic; swap in a new D1 editorial plan (world, characters, genre conventions) and it applies to other genres.
Three: 1) Long-form consistency — story bible plus gates keep dozens of chapters coherent; 2) Dual-track review — proofreader catches defects, beta reader judges value, machines checking machines; 3) Reproducibility — standardized process, reusable across subjects, not luck-based.
Humanistic Novel Theory. “Humanistic” states the human-first stance: humans are publishers and value judges; AI is the execution layer. It is first a human theory, then an AI method. HNT is also the abbreviation of Hui Nan Tian — a double meaning.
All AI subagents — but strictly isolated: the writer only writes, the proofreader only reviews, the beta reader only blind-evaluates. Agents are stateless; all handoffs go through standard documents. This isolation is HNT's core anti-crosstalk design.
HNT is currently applied to in-house works. This teaser site is its first public validation; methodology details are on this site's /en/hnt/ page.