DEC-0003 — Knowledge Model v0

Context

We need an atomic concept taxonomy for institutional memory that matches the client's mental model ("processes, workflows, roles, everything") and stays OKF-compliant, graph-connectable, and extensible.

Options considered

1. Free-form markdown, no taxonomy.
2. A heavy ontology (rigid schema, many required fields).
3. A minimal core taxonomy + typed edges + extensibility (only type mandatory).

Decision

Adopt the v0 model in Dossier — The Knowledge Model (v0):

• Core types: role, process, workflow, decision.
• Supporting types: system, policy, artifact, term, concept (fallback).
• Typed edges (owner, uses, governed_by, steps, produces, relates_to, supersedes…) in frontmatter + [[wiki-links]] in body.
• Provenance fields on every atom: source, confidence, timestamp.

Rationale

• Atomic & single-source-of-truth — one concept per file, link don't copy.
• The decision type captures judgment — the tacit IP, not just facts.
• Extensible — verticals add types freely (OKF only mandates type).
• Provenance makes memory auditable → safe to feed agents.

Consequences

• The model governs all concept authoring and the extraction pipeline's output target.
• v0 will evolve; ids are immutable, changes tracked via status/supersedes and git.

Review

Revisit after the first real client KB and the first vertical template stress-test the taxonomy.