nvs-datasets

A curated library of public-domain literature converted to Novel Visual Studio (NVS) format — and the community registry the NVS app downloads works from. Every work is a real book (or transcript) you can open in NVS and read with the engine's narrative panels live: character presence, plot threads, character arcs, and continuity all tracked across the whole text.

Sources are public-domain archives — Project Gutenberg, Aozora Bunko, and similar — plus a few publicly-available non-fiction transcripts used as showcases. Conversions run through nvs-parser's agent pipeline (scope → parse → oracle loop), and nothing is listed until it passes a machine quality gate.

What's inside

work	language	scenes	genre
Romance of the Three Kingdoms (三國演義) — Luo Guanzhong	zh	120	classic
Journey to the West (西遊記) — Wu Cheng'en	zh	100	classic
Alice's Adventures in Wonderland — Lewis Carroll	en	50	fantasy
Botchan (坊っちゃん) — Natsume Sōseki	ja	11	classic
Dario Amodei on Scaling (Dwarkesh Podcast)	en	7	interview
Trump v. Barbara (SCOTUS oral argument)	en	12	legal

The full, machine-readable catalog is index.json; the human table is CATALOG.md; quality baselines are in BENCHMARK.md.

Also in this repository (not part of the registry): games/genshin-impact/ is a non-official, fan-compiled dataset of copyrighted game-story text, included only as a technical showcase of the format on long-form branching narrative. It is not public domain, not affiliated with HoYoverse, and not listed in index.json. See NOTICE.

Using a work

In the NVS app (recommended). Open the Store, browse the community registry, and download a work — it arrives as a single .nvsproj bundle and opens with every panel live. No conversion, no re-analysis.

Directly from this repo. Each work is a plain folder of Markdown prose plus a co-located .nvs/ analysis database. Clone the repo and open any work folder in NVS:

git clone https://github.com/<owner>/nvs-datasets
# then in NVS: File → Open Project → nvs-datasets/classics-en/alice-in-wonderland

The prose is readable on its own in any Markdown viewer; the .nvs/ folder is what makes the panels light up.

What "listed" guarantees

Admission is machine-gated, not editorial. A work appears in index.json only if it scores clean ✓ on the nvs_parser.quality oracle — checked on both the source folder and a round-trip re-import of its built bundle. "Clean" means the conversion did its homework:

• Labeled — every spoken line resolves to a named speaker (no UNKNOWN dumps).
• Paged — every speaker has a character page to resolve to.
• De-cluttered — no source markup (［＃…］, **bold**, headings) leaking into the dialogue.

Each entry also carries an oracleScore, a sha256, and a minAppVersion / schemaVersion so a client can refuse an incompatible bundle before downloading.

The NVS format, briefly

{work}/
  content/
    story/chapters/{NNN-code}/{sNNN-slug}.md   ← scenes (YAML frontmatter + dialogue)
    world/characters/{id}.md                   ← character pages
    world/{lore,items,locations,...}/{slug}.md ← optional world pages
  .nvs/
    nvs.db                                     ← the analysis database (prose stays the source of truth)
    trees.json, structure.json                 ← timeline graph + project structure

A scene is frontmatter followed by dialogue in all-caps speaker format:

---
scene_id: alice-s2
title: "Down the Rabbit-Hole"
characters_present: [alice, white-rabbit]
phase: canon
---

WHITE RABBIT
Oh dear! Oh dear! I shall be too late!

ALICE
Curiouser and curiouser!

Character names may be bilingual (劉備 Liu Bei, 坊っちゃん Botchan); aliases in a page's frontmatter fold alternate names onto one entity.

Integrating with NVS

How this repo is produced and consumed — the source → oracle → bundle → registry → app-download pipeline — is documented in INTEGRATION.md. Maintainers regenerate the registry with:

python3 tools/publish.py                        # oracle-gate → bundle → dist/*.nvsproj → index.json
python3 tools/publish.py --release-base <URL>   # point download URLs at a GitHub Release

Contributing a work

1. Convert a public-domain source with nvs-parser until the oracle scores it clean ✓.
2. Drop the work folder under a category (classics-en/, classics-jp/, …).
3. Add a row to CATALOG.md and open a PR. tools/publish.py re-runs the gate; anything that isn't clean is refused automatically.

Please only contribute works that are genuinely public domain (or your own), and note the source archive.

Provenance & license

The listed works are in the public domain (Project Gutenberg / Aozora Bunko / etc.) or publicly available records used as non-fiction showcases (a public podcast transcript; a U.S. Supreme Court oral argument, which is a public record). The structural scaffolding produced by this project — the NVS format files, conversion metadata, and analysis databases — is released under CC0 (see LICENSE); source texts retain their own status.

One exception: games/genshin-impact/ is copyrighted, non-official, fan-compiled content, not covered by the CC0 dedication and not part of the public-domain registry — see NOTICE and LICENSE. Everything in index.json is rights-clean; Genshin is not listed there.