Mapping Luhmann's Brain

“What does 45 years of thinking look like when you see it all at once?”

What is this?

Niklas Luhmann (1927-1998) was one of the most productive social scientists in history: more than 50 books and 550 articles. His secret weapon was a system of 90,000 handwritten note cards — the Zettelkasten — connected through tens of thousands of cross-references. It's the most famous example of networked thinking ever built.

Bielefeld University has been digitizing and transcribing this archive since 2015, funded with €5 million, planned through 2030. Their team transcribes and verifies cards by hand — scholarly edition quality. That work is essential and ongoing.

Bielefeld's work focuses on reading every card correctly. But the cross-references — tens of thousands of them — hadn't been analysed as a network. What if you asked a complementary question? Not "how do we read every card?" but "how do we see the network?"

Using the archive's own open API, I downloaded all 73,715 indexed cards, extracted 59,773 references from the full HTML transcriptions of all 73,715 cards — including 33,650 cross-section references and 14,389 neighbourhood links — and built what appears to be the first complete structural map of Luhmann's thinking. What I found: the center of his thinking isn't where you'd expect, 23% of his cards have no cross-references beyond their own branch, and the system helps read itself when you treat it as a network instead of an archive.

Two complementary approaches

	Bielefeld (since 2015)	This project
Goal	Read every card correctly	See the network as a whole
View	One card at a time + local tree	73,715 cards as network graph
References	Outgoing (where does this card point?)	Outgoing + incoming (who points here?)
Cross-ZK	Not visualized	1,976 bridges mapped
Hub analysis	Not available	Top 50 most-connected cards
Metaphor	Reading with a magnifying glass	Seeing the city from above

A note on method: I'm not a scientist. I'm a PKM researcher who wanted to contribute a structural perspective to the extraordinary work Bielefeld has been doing for over a decade. No recognized scientific methodology has been applied to this work. This is an exploration, not a peer-reviewed study. The Bielefeld edition is the scholarly standard — this project aims to complement it.

The numbers

73,715Cards

59,773References

27Drawers

1952-97Span

Zettelkasten I (1951-1962): 22,079 cards on law and administrative science.
Zettelkasten II (1963-1997): 51,636 cards on sociology and systems theory.

Findings

50 Things Nobody Knew About Luhmann's Zettelkasten

The first network analysis reveals surprises that 70 years of scholarship missed.

Findings

What Happens When You Have All 90,000 Cards

We mapped 58%. Here's what becomes possible with the rest.

8 Lessons for Anyone Building a Knowledge System

What the most famous Zettelkasten teaches us about dead notes, cross-domain bridges, and systems that talk back.

Lessons

The Ripple Method — How Context Reads Context

AI couldn't read the handwriting. Then we let the network read itself. CER from 163% to 27%.

Method

Building a Machine That Reads Luhmann's Handwriting Good Enough

Six AI models, 73,715 cards, one day. 15.6% error rate. What happens when you stop aiming for perfection and start aiming for useful.

Pipeline

The Architecture Behind It — How ThetaOS Maps Onto Luhmann

A 10-layer synaptic model, myelination scores, and an 8-dimensional diamond layer. Why this architecture works on the most famous knowledge system in history.

Architecture

How I Got 73,715 Cards in a Couple of Minutes

The hidden API, the download script (80 lines), and why the archive was always open — nobody took it.

Technical

All posts → · RSS Feed

Background reading

I Think I've Built a Mythical Machine

On ThetaOS, the Memex dream, and why the real problem was never storage and retrieval.

Blog

Here Is Everybody

250 years of thinking about information — from Otlet to Bush to Nelson to now.

Blog

The Synaptic Architecture of a Life

Four layers, four neurotransmitters, one convergence. How ThetaOS arrived at the same architecture as the brain.

Blog

Our Intelligence Lives in Synapses, Not Neurons

Why the value is in the connections, not the nodes. The theoretical foundation behind this project.

The Information Problem Was Never the Real Problem

The academic paper: confirmation-driven knowledge graphs and four design principles.

Paper

PKM Summit 2026

The European conference on Personal Knowledge Management, Utrecht.

Event

About

Martijn Aslander

Independent thinker and builder. Creator of ThetaOS, a Life Lens System (LLS) — what emerges when you move beyond PKM by adding more elements: relations, context, time, confirmation. Powered by LLM, with 339 tables, 91,000+ records and 2.5 million words of structured content. This Luhmann project is a proof of concept for AI-augmented knowledge cartography.

The network map was built with Claude (Anthropic) as an AI collaborator — from discovering the API to parsing the XML to rendering the visualization.

Blog · LinkedIn · GitHub

Aria Khodaverdi and Martijn Aslander at PKM Summit Europe 2026

Aria Khodaverdi and Martijn Aslander at PKM Summit Europe, Utrecht, March 2026

What's happening now

The reference count has been updated from 18,289 to 59,773 after fetching the complete HTML for all 73,715 cards via the individual card API. The original count was based on truncated preview text. The full transcriptions — available for 43,262 of 73,715 cards — reveal a network 3.3× denser than first reported. References are now separated into three types: 33,650 cross-section references (Fernverweise), 14,389 neighbourhood links, and 11,182 unclassified. The remaining 30,453 cards await transcription and will add more references once readable.

In parallel, we're running an AI transcription pipeline that reads the handwritten scans of those 30,453 untranscribed cards, achieving 15.6% character error rate using Gemini's vision models. The full run is expected to complete by March 30, 2026. This will produce a searchable, confidence-scored version of every card — not replacing Bielefeld's scholarly edition, but supplementing it with broad coverage.

Updates

March 29, 2026

Reference count updated from 18,289 to 59,773 after fetching complete HTML for all 73,715 cards via the individual card API. The original count was based on truncated preview text.
References now separated into three types: 33,650 Fernverweise (cross-section), 14,389 neighbourhood links, 11,182 unclassified. Thanks to Ton Zijlstra for identifying the truncation issue and the three-layer link structure.
Explorer rebuilt with color-coded edges: blue = Fernverweise, orange = cross-ZK, green = neighbourhood.
New blog post: Building a Machine That Reads Luhmann's Handwriting Good Enough — AI transcription pipeline achieving 15.6% CER across 200 test cards.
AI transcription of all 30,453 untranscribed cards is running — expected to complete March 30.
Corrected explanation of the missing ~16,000 cards (bibliographic sections, not blank cards).
"Isolated cards" clarified: 23% have no cross-references beyond their branch, but are structurally positioned in the Folgezettel tree.
Tone revised across the site based on feedback from Aria Khodaverdi: less boasting, more humility.

March 28, 2026

Initial release: 73,715 cards downloaded, network explorer launched, six blog posts published.

Technical details

The data comes from the Niklas Luhmann-Archiv at Bielefeld University, licensed under CC BY-SA 4.0. The archive provides a search API that returns card metadata and transcription previews as JSON, and individual cards as TEI-XML.

The download script makes 9 API calls with paginated requests of 10,000 results each. The network is extracted by parsing Luhmann's cross-reference notation from the transcription data. The visualization uses canvas-based rendering with force-directed layout.

Everything is open source. View the repository on GitHub.

Limitations

This project maps the structure of Luhmann's thinking, not the content. Important caveats:

The 59,773 references are extracted from the full HTML transcriptions of 43,262 cards. The remaining 30,453 cards have no transcription yet — references from those will be added as transcriptions become available.
The visualization shows the top 2,000 most-connected cards, not all 73,715.
No semantic analysis has been performed — we see which cards link to each other, not why.
This is not a substitute for scholarly engagement with the archive. Read the cards, not just the map.

We're actively working on several of these limitations — including a context-driven transcription method for the remaining 41% and expanding the visualization beyond 2,000 nodes. This is an ongoing project.