architecture.md

Architecture

MetaGate helps teams know when AI is allowed to act.

flowchart TD
  A["DataHub"] --> B["Metadata change events"]
  B --> C["MetaGate scanner"]
  C --> D["Affected graph resolver"]
  D --> E["Multi-hop evidence collector"]
  E --> F["Policy engine"]
  F --> G["Deterministic readiness evaluation"]
  G --> H["Confidence engine"]
  G --> I["Capability certification"]
  G --> J["Gap and impact analysis"]
  G --> K["MetaGate Certificate"]
  H --> K
  I --> K
  J --> K
  K --> L["Context Contract"]
  K --> M["Readiness Diff and history"]
  K --> N["Write back to DataHub"]
  N -. "new metadata event" .-> B

Loading

Components

• ReadinessEngine: deterministic scoring, confidence, gap classification, capability certification, recommendations, and context contracts.
• PolicyProfile: YAML policy-as-code controlling evidence weights, stale thresholds, capability gates, and graph propagation.
• DataHubEvidenceExtractor: maps DataHub metadata into SDK evidence.
• BackgroundScanner: handles event-driven rescans for changed entity URNs.
• DataHubWriteback: publishes certificates and remediation tasks through a client adapter.
• ReadinessHistory: stores issued certificates for audit and diffing.

Live DataHub wiring

Implement DataHubClient with:

• get_entity: fetch ownership, glossary terms, lineage, assertions, incidents, freshness, usage statistics, and policy tags/aspects.
• get_neighbors: fetch upstream and downstream nodes for graph propagation.
• write_certificate: write an AI readiness custom aspect or equivalent assertion.
• create_remediation_task: create tasks or incidents for missing controls.