[jaydreyer]

Summary

• persist honest seeded origin metadata on the 12 canonical policy system, document, and training dependency rows
• expose relationship source, trust basis, policy scope, and explicit absence of AI involvement in the existing catalog detail view
• preserve typed relationship identities and deterministic assessment behavior while deferring imported, curated, and AI-backed relationship lineage

Reviewer outcome

A reviewer can now answer, from the Manager Travel Approval Guide detail:

• why the document is connected to MANAGER_APPROVAL_REQUIRED
• that the relationship came from seeded demonstration data
• that it is trusted deterministic enterprise input, not an AI recommendation or human approval

Scope and integrity

• no relationship generation, inference, editing, or write API
• no impact-analysis, Confluence, approval, execution, audit-timeline, or event-sourcing changes
• migration backfills only the 12 canonical seed-owned IDs and fails rather than labeling an unexpected existing row as seeded
• only seeded_demonstration is representable until other origin types have real persisted lineage

Verification checklist

• Migration upgrades from PR B, downgrades, and re-upgrades with populated dependency rows
• Provenance category, authority, reference, and timestamp constraints cover all three dependency tables
• Seed reruns and guarded demo reset preserve provenance
• Backend: 274 passed
• Backend lint/format: Ruff clean
• Frontend: TypeScript, ESLint, 48 passed, production build
• Browser: flagship detail shows two distinct policy scopes, seeded source, trusted level, and no-AI language
• Browser console: 0 errors, 0 warnings
• Existing deterministic assessment tests and golden counts remain unchanged
• Architecture, ADR, milestone, scenario, inventory, and README documentation updated
• Independent implementation review: approved with no remaining findings