2026.10.6.2 kept scheduled reports under S in the fork's subspace, a byte scale-out storage (Sb, Sm, Sl, replica markers Sr + index) and the spam-rules updater (the Sr marker, Sf, Sn) already use. Nothing was overwritten (the keys differ in length), but a scan of Sr + id could read the others' keys, and one byte per feature is the rule. They move to G, unused anywhere. ensure_digest moves what 2026.10.6.2 wrote, once per process: what reads as a report from the Sr + 8-byte-id range, and the settings at Ss, deleting only those exact keys. The system test writes a legacy digest, legacy settings and the spam marker, and checks the first two moved and the marker didn't.