Connections, backlinks and the knowledge graph

Every article ends with a connections block. It answers one question: where should I go next if this page was not enough? The block is built from the database, not from a hand-maintained index, and it is readable without JavaScript.

🔗 The three groups

Order matters: explicit links first, then backlinks, then neighbours. The page shows at most eight connections — a graph with forty edges is decoration, not navigation.

🧭 Why a list and a map

The block contains two things at once:

1. A plain list with real links. It works with JavaScript disabled, on a screen reader, and when the map fails to draw.
2. A mini-map drawn by the page script from the same data. It gives the shape of the neighbourhood at a glance: the current article in the centre, its connections around it.

The map is progressive enhancement. It never contains information the list does not have, and it is hidden from assistive technology because it duplicates the list.

For the full picture there is the knowledge graph in the main product (/kb): all published articles, not one neighbourhood.

📚 What a connection is not

• It is not an automatic embedding similarity. Nothing links two articles just because a model thinks they look alike.
• It is not a promise of freshness. A linked page can be older than the link.
• It is not a ranking signal. The wiki does not chase link counts; it tries to answer the next question.

✍️ How an editor links articles

1. Open the article in the Django admin.
2. Find Related articles and add the target.
3. Save. The change is recorded as a revision like any other content edit.

There is one rule: link to a page that exists and is published. A link to a draft is invisible on the public page, and the editor will not notice the mistake until someone reports a dead end.

🚧 When this does NOT work

• You want a cluster of twenty related pages. Eight is the cap; nobody reads the ninth.
• You expect cross-language links. Each language has its own page; the connection block follows the current language and falls back to English titles for pages that are not translated.
• You need an external graph database. The wiki reads the two fields it already has. A separate graph store would be a second source of truth for the same edges.

➡️ Next

• How this wiki is edited and published — the rest of the editorial workflow.
• Sources and verification — what a linked page must prove.
• Open the article index: /wiki/articles.json.