🌐 Product overview: Page graph on wikantik.com
— a plain-language walkthrough for readers and AI agents.
Wikantik distinguishes two graph subsystems. Confusing them is the single most common source of bugs and miscommunication when working on retrieval, navigation, or admin tooling. Use the right name.
A graph whose edges are real page-to-page wikilinks.
[OtherPage](OtherPage),
[OtherPage], etc.). Period.canonical_id field in
frontmatter (the rename-stable identifier) and the cluster: field
(hub membership)./page-graph view.com.wikantik.pagegraph.* (in wikantik-main),
com.wikantik.api.pagegraph (in wikantik-api)./page-graph (reader); /admin/page-graph/* (operator).A graph whose nodes are LLM-extracted entities and whose edges are co-mention or typed-relation predicates between them.
bin/kg-extract.sh)
reads page text and proposes nodes and edges; admins approve them.com.wikantik.knowledge.* (in wikantik-main),
wikantik-knowledge module, kg_* tables./admin/knowledge-graph/* (operator); /knowledge-mcp MCP
tool surface (agents).| Question | Page Graph | Knowledge Graph |
|---|---|---|
| What is an edge? | A wikilink one author wrote in one page | An LLM-extracted predicate between two entities |
| What is a node? | A page | An entity (concept, person, organisation, etc.) |
| Who curates it? | Authors (by writing links) | Extraction pipeline (with admin review) |
| Where does it live in code? | pagegraph.* | knowledge.*, kg_* |
The frontmatter relations: mechanism — a third concept that let
authors hand-curate typed edges between pages without writing real
wikilinks — was removed. Three of 951 pages used it; nothing
load-bearing depended on it. After removal, the Page Graph is
strictly the wikilinks graph. If curated typed edges between concepts
need to come back later, they belong on the Knowledge Graph as
admin-approved edges, not in page frontmatter.
Code that reads wikilink or cluster data from the Page Graph subsystem:
AgentHintsDeriver (in wikantik-main, package com.wikantik.knowledge.agent) — uses ReferenceManager.findReferrers(slug) to compute intra-cluster inbound link centrality for prefer_pages ranking on the /for-agent projection. See docs/superpowers/specs/2026-05-10-derived-agent-hints-design.md.docs/superpowers/specs/2026-05-02-page-graph-vs-knowledge-graph-design.md
— the spec that drove this separation.