Execution epic for the daemon AI-enrichment-of-KBs initiative. Design landed in PR #682 (ADR-101 structured CRDT edges, ADR-102 KB engine, ADR-103 autonomous local-LLM link enrichment) + docs/KB_ENRICHMENT_EPIC.md.
Why: the enrichment sweep computes embeddings only (ADR-061 §575); the link-derivation half — a local model proposing typed links with calibrated confidence, kept healthy and reviewable — is unbuilt. Two prerequisites block it: machine links have no durable CRDT home, and the hosted DB engine is bottlenecked (~8 concurrent sessions, ADR-054) vs a ~50-user target.
Dependency order: ADR-101 (hard prereq of ADR-103) · ADR-102 (deployment prereq, parallelizable) → ADR-103.
Each milestone ships behind an adversarial test that fails first (principle #14):
Design docs: docs/adr/101-103, docs/research/103-kb-link-enrichment-prior-art.md, docs/KB_ENRICHMENT_EPIC.md (all in PR #682). Numbering note: ADR-097–100 belong to PR #669 (Browser MAE) — no collision.
Execution epic for the daemon AI-enrichment-of-KBs initiative. Design landed in PR #682 (ADR-101 structured CRDT edges, ADR-102 KB engine, ADR-103 autonomous local-LLM link enrichment) +
docs/KB_ENRICHMENT_EPIC.md.Why: the enrichment sweep computes embeddings only (ADR-061 §575); the link-derivation half — a local model proposing typed links with calibrated confidence, kept healthy and reviewable — is unbuilt. Two prerequisites block it: machine links have no durable CRDT home, and the hosted DB engine is bottlenecked (~8 concurrent sessions, ADR-054) vs a ~50-user target.
Dependency order: ADR-101 (hard prereq of ADR-103) · ADR-102 (deployment prereq, parallelizable) → ADR-103.
Each milestone ships behind an adversarial test that fails first (principle #14):
Design docs:
docs/adr/101-103,docs/research/103-kb-link-enrichment-prior-art.md,docs/KB_ENRICHMENT_EPIC.md(all in PR #682). Numbering note: ADR-097–100 belong to PR #669 (Browser MAE) — no collision.