Versioned, validated knowledge base for a software cooperative, designed to be read by people first and AI agents second
- Foundation, not adopted (private)
- 2026
- 4 commits last push Sept 2026
- Python Markdown Make
- Private repository
If agents are going to read a company's documents, the documents need the discipline of code: one owner per truth, provenance per claim, and checks that fail the build.
Private repository. Built as a reference; not adopted by the cooperative.
01
Principles
- The language model is not company memory. Knowledge lives in canonical documents in Git; agents read it.
- One truth, one owner. A canonical registry names the document that owns each kind of fact.
- Every material claim has a class and a basis: fact, owner decision, inference, contradiction, open, proposed, or historical.
- Unknowns are recorded, not guessed. Access is not authority.
02
Checks
A stdlib-only validator and contract tests enforce the structure: required headings, responsibility statements, resolvable links, defined identifiers, decision and ADR structure, provenance fields, agent capability grants, and absence of template placeholders or contact data.