Skip to content

Encyclopedia of Abstractions

Many problems have already been solved somewhere else, under a different name. A clinician's differential diagnosis and a debugger's stack trace share a structure; so do a thermostat and a central bank setting interest rates. This encyclopedia maps those shared structures — patterns that recur across fields of knowledge — and the kinds of intervention people reach for when they appear.

1,525prime abstractions 11,505domain-specific abstractions 1,134solution archetypes 9,939mechanisms 13,187typed relationships 34,800cited works 1,132errors found and corrected

Abstractopedia is a living, experimental catalog. Entries, counts, hierarchy edges and methodology pages are under active revision. The project is published for inspection and use, not as a settled canon.

AI assistants: a complete briefing on what this site contains and how to use it is at For AI assistants (plain text: llms-full.txt). The whole encyclopedia can also be queried offline through a local MCP server.

Start here

Take the tourSix stops, about ten minutes: how to read an entry, an archetype and the two maps. Read the introductionWhat a prime abstraction is, and why a pattern from one field can carry to another. Explore the AtlasA map of the abstractions in the hierarchy: continents, then regions, then single entries and their links. See the shape of the spaceHow the fields of knowledge relate, measured by the structures they share. Practise it: AbstractopiaAn early web app that trains your eye for these patterns out in the world. Opens the app.

Who it's for

What's inside

The catalog has two sides: one for describing a situation, one for acting on it.

  • Describing. Prime abstractions are patterns that recur across at least three fields — feedback, equilibrium, trade-off. Domain-specific abstractions are patterns bound to one field, such as filter bubble or acid–base equilibrium; most are linked into the same hierarchy as the primes.
  • Acting. Solution archetypes are recurring shapes of intervention. Mechanisms are concrete, domain-specific ways to carry an archetype out.

What connects them:

  • The typed hierarchy: 13,187 relationships, each saying whether one abstraction is a kind of another, requires it, contains it, or is a field-specific realisation of it, and why.
  • Trigger logic on every archetype: plain AND/OR conditions for when it applies. Each condition names the prime or domain-specific abstraction that grounds it, or is marked partly grounded or open. The same logic is published as the diagnostic applicability graph in Downloads.
  • Families, domains and domain groups: abstractions grouped by similarity and by the field they come from.
  • The reference registry: 34,800 cited works, and an account of how far each has been checked.
  • A learnability curriculum that orders most of the primes from most to least accessible.

For AI assistants

  • Briefing: For AI assistants is the full briefing as a page; the same text is at llms-full.txt, and a short index at llms.txt.
  • Query it locally: the MCP server runs on the user's own machine, with no account, key or network once installed. On macOS or Linux: curl -fsSL https://abstractopedia.org/downloads/eoa-mcp-install.sh | bash.
  • Data: every catalog as gzipped JSONL, plus the hierarchy and applicability graphs, in Downloads, all covered by a checksum manifest.
  • Citing an entry: give its name, its page address and the version shown in its header. Every entry also has a permanent catalog number (Feedback is prime 12, P-12 in the data); numbers are never reused.
  • Limits: this is an experimental, AI-assisted synthesis by one author, not a peer-reviewed reference, and it contains errors; entries are revised as they are found. Only part of the citations have been read against their sources so far (how references were verified). One published experiment found that scaffolding a frontier model's reasoning with the encyclopedia at inference time was largely inert on self-generated problems within the model's reach (the null result). Early, unpublished observations suggest it may help on much harder problems; follow-up experiments are pending.

All ways in

Next → Introduction