Start with the actual dependency map
Reports of permanent customer-data loss after strikes on Amazon data centers are a reminder that a crypto asset can be onchain while its user experience remains dependent on cloud backups, indexers, notifications, and identity providers. Treat every one of those services as a potential recovery point.
Make recovery concrete
Verify that seed phrases or hardware backup procedures can restore access without a particular app account. Export transaction records and addresses in a durable format. For organizations, document who can rotate keys and where a signing policy is stored if the normal dashboard disappears.
The uncomfortable caveat
Self-custody changes the failure mode; it does not eliminate it. Lost secrets, phishing, and rushed recovery remain material risks. A useful drill is a controlled restoration to an empty device, performed before an outage rather than during one.