Summary: Most companies sit at Waypoint 1 — undifferentiated piles of tables, files, and JSON blobs — and call it data architecture. Benny Cheung lays out the five waypoints that separate raw data from a reasoning-grade knowledge graph, and pairs each one with the corresponding step in the I Ching's 5,000-year-old ontology of situations. The W3C rediscovered the same structure in 1999 and called it OWL. Staying at Waypoint 1 isn't pragmatism — it's a ceiling.

The five waypoints

  1. Data Piles — The undifferentiated. Tables, files, exports nobody owns. Every relationship implicit in column names or buried in app code. "Where most 'RAG over the database' projects quietly stall." Most teams live here and call it "data architecture."
  2. Taxonomy — First division. Yin and yang. Films break into Sci-Fi, Drama, Action. A vertical hierarchy from a single binary cut. Teams that reach Waypoint 2 typically call it "data governance."
  3. Thesaurus — Three lines stacked make eight trigrams. Heaven, Earth, Thunder, Water, Mountain, Wind, Fire, Lake. A vocabulary of primal types. Teams with a glossary here typically call it "metadata."
  4. Ontology — Two trigrams stacked make sixty-four hexagrams. A finite vocabulary from which the world's complexity can be re-derived. The W3C rediscovered this in 1999 and called it OWL. Almost no enterprise gets here — but Waypoint 4 is where AI-grade reasoning starts.
  5. Knowledge Graph — Populate the ontology with instances. Consulting the I Ching is mapping a situation to a hexagram. The recommendation that needs three SQL JOINs becomes a single graph traversal.

Why most teams ceiling at Waypoint 1

"𝗧𝗵𝗲 𝗳𝗼𝗿𝗺𝗹𝗲𝘀𝘀 𝗰𝗮𝗻 𝗯𝗲 𝗺𝗮𝗱𝗲 𝗹𝗲𝗴𝗶𝗯𝗹𝗲 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝗯𝗲𝗶𝗻𝗴 𝗺𝗮𝗱𝗲 𝘀𝗺𝗮𝗹𝗹𝗲𝗿."

Cheung's central wager: structuring data does not lose detail. Sixty-four hexagrams hold every possible situation; an enterprise ontology runs the same wager. Your data doesn't lose detail by being structured — it gains the road that lets you walk between the pieces.

The W3C stack has shipped this for over twenty years. The I Ching has run it for five thousand. What's missing isn't capability — staying at Waypoint 1 is a ceiling, not pragmatism.

Why this matters for LLM systems

LLMs generate probabilistic outputs. Ontologies provide structure and shared meaning. When you work with knowledge graphs or a semantic layer, you're doing applied philosophy in practice:

  • defining entities,
  • structuring relationships,
  • deciding what "exists" in your business.

That's why every RAG project that tries to skip the ontology layer (and read straight off the data pile) stalls. The retrieval surface has no shape to retrieve along.

This is also the empirical-memory layer underneath Context Crafting in Kapil Viren Ahuja's Four Crafts — the knowledge base that turns architectural decisions from guesswork into repeatable engineering. Without an ontology, the system can't reason about which architecture, which patterns, which tools; it can only re-invent them.

Three verbs run the stack

Cheung's coming-up-next preview: knowing what each waypoint contains doesn't tell you how to build one. Three verbs run the entire stack:

  • MODEL — design the ontology.
  • POPULATE — load instances.
  • QUERY — traverse the graph.

Different domains, different formats, same pipeline. "The waypoints don't care what the data is about."

  • Ashleigh Faith — How to Model Taxonomy, to Thesaurus, to Ontology, to Knowledge Graph (video; same progression from a knowledge-engineering practitioner's lens). YouTube: FJ3ZqtnJ_go.
  • Hogan et al. — Knowledge Graphs (Morgan & Claypool, 2022). Open-access at kgbook.org. Canonical academic reference covering data models, query languages, schemas, identity, ontologies, reasoning, and embeddings.