LLM FIELD NOTES / LEARNING NOTE

A graph is a hypothesis

How does a document become a representation you can improve? A knowledge graph is not a magical summary of a document. It is a compact, inspectable hypothesis about which concepts matter, how they relate, and what evidence supports each claim. That makes it useful: a person can c

Open in the interactive wiki ↗ Read Markdown

How does a document become a representation you can improve?

The short version

A knowledge graph is not a magical summary of a document. It is a compact, inspectable hypothesis about which concepts matter, how they relate, and what evidence supports each claim. That makes it useful: a person can correct the hypothesis, preserve the correction, and use it to guide the next extraction.

The important loop is:

document → candidate graph → human review → reusable guidance → new graph

The graph is an internal representation. Markdown, Obsidian notes, JSON, and JSON-LD are projections of that representation, not separate sources of truth.

Build it

Start with one document and write down three different layers:

  1. Concepts — the entities or ideas that should remain addressable.
  2. Relations — the claims that connect one concept to another.
  3. Grounding — the source text and location that justify each item.

Then give every candidate a status. inferred means the extractor proposed it. accepted and rejected mean a person reviewed it. Keep those states separate: an extractor should be allowed to make a useful mistake without rewriting what a person already decided.

Export the same graph into a viewer such as Obsidian. A good projection lets a reader inspect evidence, edit a bounded review field, and bring the correction back without creating a second database.

Failure modes

source actually supports it.

one label can describe different ideas in different documents.

context noisy, expensive, and hard to audit.

identity can silently apply a stale correction to a newer graph.

not turn an uncertain hypothesis into an unquestionable fact.

The practical response is to keep provenance, stable identities, bounded review memory, freshness rules, and an explicit queue for unresolved candidates.

Sources

Try it yourself

Take a paragraph you know well and build two graphs: one with only labels, and one with evidence and review status. Ask another person to reject one candidate and accept one missing concept. Re-run the extraction with only those compact decisions as guidance. Compare not just the number of nodes, but which claims became easier to inspect and which unsupported claims disappeared.

← Back to LLM Field Notes