The boring ETS owner (implementation.md §2, §5): holds the instance's routing-cache table. It sits before Fief.Node under the instance's rest_for_one supervisor, so a Fief.Node crash/restart never drops the cache mid-lookup — the restarted node overwrites rows as it re-joins, and readers fall back to the Authority on any miss.

Fief.Node is the table's only writer of routing state. Readers use plain :ets.lookup (the table is :public with read_concurrency: true); nothing ever calls this process or Fief.Node on the read path. Row shapes:

• {{:route, vnode_id}, owner, prev_owner, row_epoch} — one cached table row; prev_owner nil = settled; row_epoch = epoch of the last CAS on this row (the transfer-session epoch, Fief.StateStore M4 decision)
• {:table_epoch, epoch} — the epoch the cached table was read at
• {:node_status, map} — Fief.Node's published upward interface (%{node_id:, owned:, epoch:, lease_live?:, state:}); read it through Fief.Node.status/1
• {:router_config, map} — static instance config the process-less Fief.Router reads on its hot path (store handle for the authority fallback, max_moved_hops, fallback_jitter, the leadership adapter for Fief.leader/1). Written once, here, at init — immutable, so it does not breach the single-writer rule for routing state (M4 phase B decision).

This process handles no messages of its own. If it ever grows handlers, the M2 process discipline applies: bind the sim context at init (done below) and end every handling with Fief.Seam.checkpoint(:handled).