diff --git a/.beans/ATFS-t46t--emmc-is-ext4-now-pin-the-choice-sync-the-docs.md b/.beans/ATFS-t46t--emmc-is-ext4-now-pin-the-choice-sync-the-docs.md
new file mode 100644
index 0000000..5d3fc9b
--- /dev/null
+++ b/.beans/ATFS-t46t--emmc-is-ext4-now-pin-the-choice-sync-the-docs.md
@@ -0,0 +1,16 @@
+---
+# ATFS-t46t
+title: 'eMMC is ext4 now: pin the choice, sync the docs'
+status: todo
+type: task
+priority: normal
+created_at: 2026-08-08T23:14:47Z
+updated_at: 2026-08-08T23:14:47Z
+---
+
+The 2026-08-08 bench reflash surfaced that the gosd module bump changed the eMMC default filesystem from FAT32 to ext4 (journaled, crash-safe) without atfs choosing it: the labelled-ATFSDATA-but-wrong-format guard refused the mount until ATFS_FORMAT_DRIVES_IF_NOT_ATFS authorized a wipe. The bench NanoPi migrated (new peer ID 12D3KooWBKCynScP2zz5UEH8wWdSC8RPogNaSAcsBzBbhxF728hT, record recreated). Follow-ups:
+
+- [ ] Pin the filesystem choice explicitly in cmd/atfs-gosd (pass gosd's Options.Filesystem rather than inheriting the default) so the next gosd default change is a conscious atfs decision, with a comment recording why ext4
+- [ ] Sync docs: CLAUDE.md sharp edges still describe the data volume as FAT ("Never put rotating logs on the FAT data volume", emmcLabel "FAT volume label" comment in datadir.go); reword for ext4 while keeping the crash-safety lesson
+- [ ] Revisit ATFS-c2ny in the ext4 world: the journal removes most of the corruption class, and e2fsck exists — its rebuild-flag design may shrink or simplify
+- [ ] The old dev.atfs.server record at the pre-migration rkey (12D3KooWGsFsFUCx9oVR7aA8AmhwyxcaBSkbJA6VQccj77ANykdx) is stale — delete or document