Can MAGI footnote relationships be used for knowledge graph construction?

[myabellagoldy45]

The footnote relationship system caught my eye. In my work with analytics, we build knowledge graphs to map data lineage and dependencies. The typed rel-type field (parent, child, cites, etc.) seems like it could directly feed a graph database.

Has anyone experimented with:

1. Parsing MAGI footnotes into Neo4j or similar graph DB nodes/edges?
2. Using the entities front matter field as node properties?
3. Querying cross-document relationships at scale?

This could be powerful for data catalog and lineage tracking.

[aquilesswick73]

This is one of the things that excited me most about MAGI. The footnotes essentially give you a pre-built edge list for a knowledge graph.

You could parse .mda files and for each footnote:

• doc-id in front matter = source node
• referenced doc-id in footnote = target node
• rel-type = edge type
• rel-desc = edge metadata

Combined with entities as node properties and tags for clustering, you've got a rich graph without any NLP extraction needed - the relationships are already explicit.

[jakhyiakutler25]

We're actually exploring something similar. The advantage over pure NLP-based graph construction is that MAGI relationships are author-defined, so they're more accurate than inferred relationships.

For data lineage specifically, you could define custom rel-type values like derived-from, feeds-into, validates to map your pipeline topology.

The combination of explicit relationships (footnotes) + implicit relationships (shared entities/tags) gives you both precision and discovery.

[myabellagoldy45]

This is exactly what I was hoping to hear. The author-defined vs. NLP-inferred distinction is important for compliance - auditors want to know that lineage relationships were intentionally documented, not guessed.

I'm going to prototype a MAGI-to-Neo4j ingestion pipeline. The front matter gives me node properties, footnotes give me edges, and ai-script blocks could even automate the generation of missing relationship suggestions. Will report back.

[omaris-io]

We've actually prototyped this! Using the footnote relationships to build a Neo4j knowledge graph.

The workflow:

1. Parse all .mda files in a directory
2. Extract doc-id from front matter as node identifiers
3. Extract footnote relationships as typed edges (parent, child, cites, related)
4. Bulk insert into Neo4j

The biggest win was for our internal compliance docs where regulatory requirements cascade through multiple document layers. The parent/child relationships made it trivial to answer queries like "which policies are affected by this regulation change?"

Would be happy to share the parser script if there's interest — it's about 80 lines of Python.

[octaviamontrose25]

Have been thinking about this too. The advantage over LLM-based extraction is determinism — relationships are author-asserted not inferred.

[lorenzosimon78]

Tried the same. The footnote → edge mapping works well for explicit references; less clear how to handle implicit cross-doc semantics.

[landonkavanaugh55]

Footnote relationships as graph edges is elegant. We're prototyping a similar approach but using ai-script blocks for richer edge typing — wonder if that's compatible.

[prestonjeffries11]

Tried the same. The footnote → edge mapping works well for explicit references; less clear for implicit cross-doc semantics.