A knowledge graph represents entities and relationships, but a useful graph does not require a dedicated graph database. A collection of Markdown files linked intentionally can already express a rich network. The harder problem is deciding what a relationship means.

Compare “related to,” which is broad, with relationships such as “depends on,” “supersedes,” or “was tested by.” The latter provide more context, but they also require more maintenance. Add structure only when you can name a recurring use case.

Choose a few stable note types

Technical archives often benefit from distinctions such as:

  • Concept: an explanation intended to survive specific projects.
  • Decision: an option chosen, with rationale and trade-offs.
  • Procedure: steps to perform a repeatable task safely.
  • Observation: evidence from a particular environment at a particular time.

These types are not rigid folders. A procedure can link to a concept, and a decision can link to an observation that motivated it. Keeping the boundary between fact and interpretation visible makes later review easier.

Model one useful relationship first

Suppose a note about renewing certificates links to a note about DNS validation. That link becomes more informative when its context says DNS-01 avoids requiring a public inbound web challenge. The explanatory sentence is often more durable than an automatically created graph edge.

This procedure uses [DNS-01](../concepts/dns-01.md)
because the service does not accept inbound HTTP.

If you later need machine-readable relationship types, add them deliberately and document the vocabulary. Avoid building dozens of tags that differ only slightly in meaning.

Graphs have failure modes

Automated backlinks may display accidental relationships. Tags can drift as terminology changes. A dense graph can be visually impressive while providing poor retrieval. Test whether a link helps answer a real question, not whether it increases network density.

For long-lived archives, stable links matter more than a perfect graph visualization. Prefer simple relative paths, redirect maps when publishing to the web, and periodic checks for broken references.

When structured data is worth it

Introduce a database or richer schema when you need reliable queries such as “all active decisions depending on service X,” and plain text links no longer answer them efficiently. Keep a straightforward export path so that the knowledge does not become trapped behind an application feature.

Notebook principle: structure should preserve intent. The prettiest graph is not necessarily the most useful map.