RTBF cascades

When a customer asks for their data to be deleted, you have to delete it. In a multi-agent system, the original record may have been summarized, paraphrased, or referenced into derivative memories that other agents wrote later. Deleting only the original leaves derivatives behind.

memledger walks the derivation graph downstream and applies a deletion plan: derivatives with a single parent are auto-deprecated; derivatives with multiple parents are flagged for human review.

plan = await ml.plan_cascade(memory_id=incident_id)
# Inspect: which records will be deprecated, flagged, skipped?

result = await ml.execute_cascade(plan)
# {"deprecated": [...], "flagged": [...], "skipped": [...]}

The audit log captures every step: source record, target records, action taken, timestamp, requester ID. Evidence-grade for SOC2, GDPR Art. 17, CCPA reviewers.

The full configuration reference — retention policies, cascade limits, audit-log export formats — ships in a future release.