Disks

Storage favors safety over raw capacity. The owned hosts write to datacenter-grade NVMe SSDs arranged as mirrors — every block lands on two drives, so a single disk failure costs nothing but the drive. ZFS sits underneath, checksumming data to catch silent corruption and snapshotting state on a schedule.

Bulk and overflow capacity comes from attached network storage, where raw speed matters less. The models, capacities, and replication intervals are in the Hardware Manifest; the constant across all of them is mirrored, checksummed, recoverable.