Beyond the rulebook: how community knowledge works

Why rulebooks aren't enough

Rulebooks are written before a game ships. They describe the intended system, but they can't anticipate every ambiguous corner case, every interaction between expansion cards, or every question that emerges once thousands of players start probing the rules in real games.

What happens in practice? Players go online. They post questions. Designers respond. Rules clarifications accumulate. Errata gets issued informally in forum posts that never made it into a second printing. Someone discovers a five-card combo that the designer never considered and asks whether it's legal. The answer, buried in a thread from three years ago, is the most authoritative source on that specific interaction.

Board Game Librarian's Tier 2 system is designed to find those answers. When the official rulebook doesn't cover something clearly, the system looks beyond it — into the community knowledge that's been building around the game since it launched.

Where the community knowledge comes from

The community knowledge base is built from forum threads scraped from major board game discussion platforms. The data sits in two database tables: bgg_threads and bgg_posts. As of this writing, that's 591,000 forum threads and 3.9 million posts across 4,533 games with community coverage.

These aren't random internet comments. Board game community forums have a specific culture: rule questions get precise answers, designers participate directly, and the community self-corrects bad rulings through discussion. The signal-to-noise ratio is genuinely better than most online discourse.

The scraping covers official rules forums, general discussion forums, and designer commentary threads. Not every thread is equally useful — a thread about which artwork someone prefers contributes nothing to rules interpretation. The processing pipeline handles this.

How community content is processed

Raw forum threads don't go directly into the answer pipeline. They go through the same processing chain as PDF rulebook content.

Thread content is first chunked into segments that fit within the context window limits of the embedding model. Long threads with hundreds of posts get split; short two-post clarifications might stay together.

Each chunk then gets AI enrichment — a pass where the system analyses the content for rules-relevance, identifies what game mechanic is being discussed, and assigns a "worthiness" score. A thread where the designer explicitly clarifies an edge case scores high. A thread debating house rules scores lower.

Finally, each chunk gets embedded using the same jina-v2-small-en model used for PDF chunks. The embeddings go into the chunk_embeddings table with chunk_type = 'forum_thread' — same table as PDF chunks, distinguished by type.

This means community knowledge and rulebook knowledge live in comparable vector spaces. A query that's semantically close to a rulebook passage will also retrieve nearby community chunks if they discuss the same concept. The retrieval is unified even though the sources are separate.

When community knowledge is used

Community knowledge isn't always retrieved. It's a Tier 2 feature, and Tier 2 activates automatically: when the Tier 1 answer can't be grounded in the rulebook — confidence too low, or the rulebook simply doesn't cover the point — the system escalates on its own. There is no button to press, and it works the same on web chat and Telegram. When the final answer is built from community forums because the official rulebooks don't cover the point, the answer says so directly, so you know it is community knowledge rather than an official ruling.

Edge-case questions (interactions between rules that the rulebook doesn't clearly address) and compound multi-part questions are the most frequent triggers, simply because they are the questions the rulebook most often fails to cover cleanly.

Simple rule lookups — "how many workers do I start with?", "what does the Forest terrain do?" — stay at Tier 1. These have clear rulebook answers. Pulling community threads into a simple lookup would add latency without adding value.

The distinction matters for publishers too: Tier 1 answers in 5–7 seconds. Tier 2 takes 7–35 seconds. Not every question needs the full treatment.

How official and community sources are combined

Tier 2 combines the top-ranked rulebook sections with the top community threads relevant to the question. Both go into the synthesis prompt together.

Authority ordering is explicit in the prompt design. The official rulebook always takes precedence. If the rulebook says one thing and a community thread says another, the answer follows the rulebook and notes the discrepancy. Community content adds context, clarifies ambiguities, and covers gaps — it doesn't override official rules.

Citations are kept separate. Official rules get cited with PDF source and page numbers. Community sources get cited separately with links to the original threads. Users can follow the links to see the full conversation context — not just the extracted chunk, but the thread it came from.

This separation serves a purpose beyond transparency. If someone disagrees with the answer, they need to see where it came from. Conflating official and community sources into a single citation would obscure that.

What community sources actually contain

To be concrete about what's actually in these threads:

Designer clarifications are the highest-value content. When a designer responds directly to a rules question in a forum, that response has authority that clarifies what the rule is intended to mean. These are often the only official source for questions the rulebook left ambiguous.

Rules Q&A threads where experienced players work through edge cases are not authoritative in the same way as designer comments, but often accurate and detailed. The AI enrichment process scores these on apparent accuracy and community consensus.

Errata and corrections issued informally also show up here. Publishers sometimes correct rules mistakes in forum posts rather than releasing updated rulebook PDFs. Community threads capture this; official PDFs often don't.

Expansion and promo interaction clarifications make up another major category. "Does the expansion mechanic interact with the base game rule this way?" questions are extremely common. The rulebook for the base game obviously doesn't cover expansions released three years later. Community threads do.

Then there are documented common misplays — cases where players regularly get a rule wrong, with the correct interpretation explained. These are useful when the user might be describing a misunderstanding rather than a genuine gap in the rules.

What the system deprioritises: opinion threads, strategy discussions, debates about game balance, artwork commentary. The enrichment scoring pushes these down, but they're still in the corpus. Vector search retrieves by semantic similarity to the question, which naturally filters toward rules-relevant content.

The coverage problem

Coverage is uneven, and there's no way around it.

Popular games have extensive community knowledge. A highly discussed worker placement game might have 10,000 threads spanning eight years of community play. An obscure abstract game from a small publisher might have 40. That difference matters for Tier 2 quality.

For popular games, Tier 2 can answer edge cases that the rulebook genuinely doesn't address, because thousands of players have encountered and debated those exact edge cases. The community has effectively produced an unofficial FAQ over years of play. For niche games, Tier 2 has less to work with. The answer might still be good — even a few well-written threads can cover the common edge cases — but the comprehensiveness isn't there.

The stats reflect this: 4,533 games have community coverage in the database. That's not every game in the catalogue. Games added recently, games with small player bases, or games from regions with less active online communities may have limited or no forum data available.

Coverage reflects actual community discussion. The system can't manufacture community knowledge that doesn't exist.

What community knowledge cannot do

Community knowledge doesn't replace official rulings. A forum thread, even one with a designer response, isn't a rulebook update. The system treats it as clarifying evidence, not authoritative rule text.

It doesn't handle contradictions automatically. If five forum threads give five different answers to the same edge case question, the system will synthesise the available evidence and note the disagreement rather than confidently picking one. "The community disagrees on this point" is a valid answer when the community genuinely disagrees.

Games with no forum presence have no community knowledge layer. This affects some classic games, some regional games, and some games that predate widespread online board game discussion.

The forum data has a scrape date. Designer clarifications posted after the last scrape aren't in the system — an inherent limitation of any scraped dataset.

And if the rulebook is clear, community speculation doesn't override it. If the rulebook says "draw two cards" and a community thread speculates it should be three, the answer is two.

Questions

How do I trigger Tier 2?

You don't have to. Tier 2 triggers automatically when your question isn't grounded in the rulebook or Tier 1 confidence is low. Escalation is silent — the answer just takes a few extra seconds and arrives with community citations.

Will I always get community sources cited?

Only if relevant community threads were found for your specific question. If vector search doesn't find sufficiently similar community content, the answer stays rulebook-only. No community sources doesn't mean the feature is broken — it means the question was well-covered by official rules, or no relevant threads exist.

Can I tell which parts of the answer came from community sources?

Yes. Official citations (with page numbers) and community citations (with thread links) are listed separately. You can follow community links to read the full original threads.

What if a community source is wrong?

The synthesis process prioritises official rules over community content. If a thread contradicts the rulebook, the official ruling takes precedence and the discrepancy may be noted.

Why does Tier 2 take longer?

Tier 2 retrieves and synthesises more content: top PDF chunks plus community threads, then a longer synthesis pass with more context. More inputs means more processing time. The 7–35 second range reflects both retrieval and synthesis variability.

Does community knowledge cover expansions?

Yes, where threads exist for expansion interactions. This is actually one of the stronger use cases — expansion interactions are rarely covered in base game rulebooks, but they're extensively discussed in community forums.