Notes

The theoretical register of the Commonplace KB: transferable claims, mechanisms, definitions, and synthesis about how agent-operated knowledge bases — LLM wikis — should be built and operated. This is the theory the rest of the repository runs on; the methodology these notes describe is what agents follow to maintain this wiki.

The central thread is deploy-time learning — how deployed AI systems improve through structured knowledge that accumulates alongside the code — developed through the constraining, distillation, and discovery operations.

  • tags index — the top-level hub linking every tag README (foundations, evaluation, learning theory, links, and the rest), plus the workshop layer and open gaps.
  • definitions/ — the project's core vocabulary (register, constraining, distillation, codification, and the rest), one term per file.

How to read these notes

Titles are assertions, not topics, so following a link reads as a chain of reasoning: since [title] and because [title] compose into an argument rather than a table of contents. Each note carries a maturity mark (seedling → current) and links to the sources or notes it rests on, so you can tell how much weight a claim holds before you build on it.

What belongs here

A claim earns a note when it changes how someone would build or operate a knowledge base. Pure pattern-recording without explanation belongs in kb/log.md, not here. For the full authoring contract — register, the reach quality bar, title and link conventions, and types — see COLLECTION.md.

Not the right collection?

  • Current-state docs about the shipped Commonplace system — architecture, type system, ADRs → kb/reference/
  • Procedures, skills, review gates, how-to guidance → kb/instructions/
  • Reviews of external agent-memory and knowledge systems → kb/agent-memory-systems/
  • In-flight drafts, investigations, and migration plans → kb/work/

Complete file listing (generated at build time)