Learning

EmbrAIon learning is evidence-driven and governance-gated.

The system may collect privacy-safe observations from runtime, validation, review, and eval outcomes. Repeated evidence can become a candidate for a reusable rule, skill, workflow, routing adjustment, knowledge entry, or eval.

No candidate becomes canonical automatically.

Promotion follows the Core Learning workflow and ordinary engineering controls: review, validation, appropriate eval coverage, and explicit approval.

Knowledge maintenance compares explicit document/source relationships with a reviewed local baseline. Drift creates a review signal, not a rewritten document or promoted rule. Skill authoring uses focused deterministic checks and appropriate live cases before a candidate is promoted. Checkpoint freshness never restores stale reviewer confirmation.