Agent-facing knowledge: declarative Kinds, not generated prose

DNA represents agent-facing knowledge as declarative, curated Kinds with provenance — Research (a synthesis of N cited References, each finding evidence-rated), LessonLearned, and the SDLC timeline itself — rather than as LLM-generated prose (the OpenWiki / DeepWiki approach of auto-writing a repo wiki for coding agents).

The difference is epistemic. A generated wiki page is trusted by the next agent that reads it, so a confidently-wrong page is worse than no page. A DNA Research doc is deterministic and verifiable: its findings carry an evidence_rating, cite specific sources, and live under the same review and lifecycle (draft → published → superseded) as any other tracked artifact. The knowledge is curated once and stays auditable, instead of being regenerated and re-trusted on every run.

This is a positioning choice, not a rejection of the tools: a generated wiki can still be run as an external tool against the repo. But on DNA's critical path, agent knowledge is data with a citation graph — proof over prose.