Extract Domain Memory

Use this when the Registry should learn the domain from a source it has not read yet: a requirement, a codebase with no or little Registry, or a change whose domain effect is not yet recorded.

1. Run readiness --repo-root <repo>. When no Registry exists, start one with init-domain-memory, choosing its flags as the workflow describes. Candidates may cite agent-asserted sources; only promotion to reviewed needs a developer to have confirmed them with confirm-sources.
2. Pick the entry that matches the source and follow it in extraction:
   • Requirement: from intended behavior to terms, owner, rules, events and collaborations.
   • Code: from executable behavior to the model it implements, language first and boundaries second.
   • Change: from a requirement change, diff or refactoring to the records it adds, changes or retires.
3. Decide each element by its DDD definition, not by the technique that surfaced it. Event Storming, Domain Storytelling, Example Mapping, code reading and change-coupling are all valid ways to find candidates; none of them is recorded.
4. Write each finding with upsert-candidate and evidence built by cite --show, passing each citation object whole, then run validate. retract-candidate withdraws a candidate that turns out wrong. A record states what the domain is. How it was found, which technique suggested it, and how confident the extraction felt stay out of the Registry.
5. Hand the candidates to the Review capability, domain-memory-review. Extraction never promotes its own findings.

When a reviewed Registry exists for the same code, compare-registry scores an extraction against it.