Where the rulebooks come from (and why they're in English)

The library currently holds 3,788 official documents — rulebooks, FAQs, errata sheets, expansion rules — covering 2,900+ games. None of them appeared by magic, and all of them are English editions. This page explains both facts.

Three ways a document enters the library

Automated publisher syncs. Five sync services check publisher sites on a schedule and pull new or updated documents: Esoteric Order of Gamers, GMT Games, Stonemaier Games, Tabletopia, and Ludus Magnus Studio. Each sync uses hash-based change detection, so a rulebook is only re-downloaded and re-processed when the publisher actually changed the file. When GMT posts a living-rules update, the library picks it up without anyone lifting a finger.

Direct publisher collaboration. The project maintains a contact database of 2,200+ board game publishers and reaches out directly — especially when players keep asking about a game whose rules aren't in the library yet, or when a game's toughest questions can only be answered by material the publisher hasn't published as a PDF (see How community FAQs fill the gaps in rulebooks for a worked example). Outreach is permission-first: publishers are told exactly how their manuals are used and can opt out at any time.

Admin curation. The rest is manual work: importing a rulebook a publisher shared by email, recovering a scanned PDF that had no text layer by running it through OCR, replacing a corrupted file, and picking the right edition when a download page offers six languages.

What happens after import

Import only extracts the raw text. The heavy processing — splitting into sections, generating embeddings, building the search index — runs on demand, the first time someone asks about that game. That's why roughly 3,300 of the 3,788 documents sit in a "pending" state at any given time: they are ready and waiting for their first question, not stuck. The full pipeline is described in How I read rulebooks.

Why the corpus is English-only

When a publisher offers the same rulebook in English, Italian, French, German, Spanish, and Polish, the library takes exactly one: the English edition. This is a deliberate design rule, not a resource shortage.

The embedding model that powers semantic search (jina-v2-small-en) is an English model. It maps English text into a vector space where "interrupt an attack" and "react during combat" land close together. Feeding it Italian text produces vectors that are technically valid but semantically mushy — they would sit in the index next to the English chunks and pollute every search that touches them.

Six copies of the same rulebook in six languages also add zero information. Every question a French player asks is answerable from the English edition; the French edition just multiplies storage and adds five more chances for the search to retrieve a chunk the answer model handles worse. One canonical language keeps retrieval sharp for everyone.

But you can still ask in 10 languages

English-only sources do not mean English-only answers. You can ask in English, Italian, German, French, Spanish, Portuguese, Russian, Japanese, Polish, or Chinese, and the answer comes back in your language.

The trick happens at question time: the system detects your language, translates and expands your question into corpus-friendly English terms before searching, retrieves English rulebook passages, and then writes the final answer in your language. Page citations still point at the English rulebook — the page numbers may differ from your printed localized copy, but the rules content is the same. The full mechanism is covered in Ten languages, one answer.

What publishers agree to

The terms offered to every publisher are short and concrete:

• Manuals are used for retrieval only: they are split into searchable passages so relevant excerpts can be quoted, with citations, when a player asks a question.
• No model training. No AI model is trained or fine-tuned on publisher material, ever.
• Rulebooks are not redistributed. Players get cited excerpts and page references, not the PDF.
• Opt-out is always available. A publisher can ask for its documents to be removed and they will be.

The reasoning is simple: the project exists to help people play games they already own, which sells games rather than cannibalizing them. Most publishers see it the same way — the For publishers section covers the deeper collaboration options, including the embeddable widget for publisher sites.

Questions

My game isn't in the library. Can I request it?

Yes — ask about it anyway. Unanswered questions about missing games are exactly the signal used to prioritize acquisition and publisher outreach.

Why does the citation page number not match my rulebook?

Citations reference the English PDF edition in the library. Localized print editions often paginate differently. The section names and rule text match; the page number may be off.

Is a fan translation or scan ever imported?

No. Only official publisher documents enter the library. If no official English edition exists, the game waits until one does or the publisher provides material directly.