Reference
Reference pages are quick lookups, not tutorials — the thing you open mid-design-review to check Raft’s exact leader-election timeout behavior, not the thing you read start to finish. Each one is short, dense, and link-rich back to the Learn module or Architecture page that goes deeper.
Every lookup has the same five sections, enforced in CI: At a Glance, Key Concepts, Numbers That Matter, Common Gotchas, and Where to Go Deeper. The shape is deliberately narrow — those are the four questions a lookup actually gets opened for, plus the route back into the deeper material. Anything that needs more than that belongs in a Learn module.
Published
Section titled “Published”| Topic | Category |
|---|---|
| Python | Language |
| asyncio | Concurrency |
| Kafka | Messaging |
| Redis | Data store |
| CAP theorem | Distributed systems theory |
| Raft | Consensus |
| Docker | Runtime |
| Kubernetes | Orchestration |
| AWS | Cloud platform |
| MCP | Agent protocol |
| LangGraph | Agent orchestration |
| RAG | Retrieval architecture |
| Vector DB | Data store |
The fast-moving four
Section titled “The fast-moving four”The last four cover subject matter that moves fast enough for editorial age to be the wrong instrument — a page on MCP can be two months old and already wrong, while the Raft page can be two years old and still correct. Each therefore declares what it was checked against, and CI fails once one passes its 90-day review window rather than letting it quietly go stale.
| Topic | Verified against | Note |
|---|---|---|
| MCP | MCP Specification 2026-07-28 |
A protocol revision, so the spec date is the version |
| LangGraph | langgraph==1.2.11 |
A library, so the release is the version |
| RAG | LangChain 1.3.15 |
RAG is a pattern with no version; LangChain is the implementation baseline |
| Vector DB | Pinecone API 2026-04 |
A category, not a product; Pinecone is the versioned anchor |
The last two rows are the interesting ones. Neither RAG nor “vector database” has a version, and inventing one — “RAG 2.0” — would be a claim about currency that nothing backs. Naming a real implementation baseline instead keeps the conceptual material framework-neutral while still giving the review window something concrete to check against.
See the Roadmap for what ships next.