Wikis

Why Wiki Page Types Exist

5 min read

Why Wiki Page Types Exist

Different kinds of knowledge need different structures. Page types (entity, concept, overview, how-to, comparison) are not bureaucracy — they are the mechanism that keeps a wiki navigable as it grows and makes AI maintenance cheap.

The Core Problem

A flat wiki — everything in one folder, no enforced structure — degrades predictably:

  1. Pages accumulate. You can’t tell what’s in them without reading them.
  2. Related facts scatter across multiple files. No canonical place for a named thing.
  3. Procedures mix with facts. How-to steps buried in entity descriptions are hard to follow.
  4. Maintenance cost grows faster than value. Humans abandon the wiki.

The LLM Wiki pattern (Karpathy, 2025) is explicit about this: “The tedious part of a knowledge base is bookkeeping — updating cross-references, keeping summaries current, catching contradictions. Humans abandon wikis because maintenance grows faster than value.”

Page types solve this by giving every piece of information a natural home.

Why Each Type Exists

Entity

A named, concrete thing — a product, service, tool, API, or system. Entities are the nouns of the wiki.

Why: The same named thing gets referenced across many pages and many ingests. Without a canonical entity page, facts about Elasticsearch end up scattered across every source that mentions it. An entity page is the accumulation point — each new ingest adds to it rather than creating a duplicate.

Concept

An idea or mechanism — not a named product but an abstract principle.

Why: Concepts don’t have a home in entity pages (they’re not named things) and don’t belong in how-tos (they’re not procedures). Without a concepts folder, ideas like “context distillation” or “watcher actions” either bloat an entity page or get lost. Separating them lets the wiki explain how things work independently of what the things are.

Overview

The big picture of an entire topic — how all pieces fit together.

Why: After ingesting several entities and concepts, no single page tells the whole story. The overview is written last (or after a synthesis pass) and links everything together. A reader entering a topic for the first time needs a map before diving into details. Without overviews, the wiki has facts but no narrative.

How-To

A step-by-step procedure for doing something specific.

Why: Procedural content has a fundamentally different structure from factual content. Steps need to be followed in order; facts need to be looked up and cross-referenced. Mixing them makes both worse. A how-to page answers “how do I do X?” without requiring the reader to extract steps from a fact-dense entity page.

Comparison

Two or more things placed side by side, with explicit tradeoffs.

Why: Without a comparison page, the wiki might have excellent entity pages for Elastic Agent and Logstash but no answer to “which should I use?” A comparison forces the wiki to make tradeoffs explicit rather than leaving the synthesis to the reader. It is the natural output of a SYNTHESIZE pass when multiple sources have accumulated enough to tell a coherent story.

Why This Matters for AI Maintenance

When an AI opens the wiki and encounters a new source, page types answer two questions immediately:

  1. Where does this information go? A fact about a named product → entity page. An abstract mechanism → concept page. A step-by-step guide → how-to page.
  2. What already exists here? The AI can scan entities/ to find canonical pages rather than reading everything.

Without types, the AI has to guess. Guesses compound: a second AI session, or a different AI tool, makes different guesses. The wiki fragments.

With types, any AI — Claude, Codex, Gemini — navigates and maintains the wiki the same way. This is the point of the schema: “Without a schema, the LLM is a generic chatbot. With one, it’s a consistent collaborator that maintains your wiki the same way across sessions and across different AI tools.”

The Deeper Reference: Memex and Associative Trails

The LLM Wiki pattern draws on Vannevar Bush’s Memex (1945) — a vision for a personal, curated knowledge store where ideas are linked by associative trails rather than linear indexes. Bush’s insight was that the mind works by association, not hierarchy, and a knowledge store should reflect that.

Page types implement this: entities link to concepts, concepts link to overviews, overviews link to how-tos. The cross-reference graph is the associative trail. The wiki compounds in value because links already exist when you need them.

Bush couldn’t solve who maintains the trails. The LLM handles that.

See Also

Graph View

Backlinks