1Decision
Confirm the combined 0038+0039+0040 storage fix on physical lynx hardware against the exact original trigger; close the incident ADR-083 explicitly left open.
2Why and evidence
The exact original trigger scenario was repeated on the flashed combined candidate: fresh boot, then the decoy's PIN entered directly with the owner's own credential never entered this boot.
Device-side evidence, checked directly rather than taken from a visual report: adb shell dumpsys user showed user 0 genuinely RUNNING_LOCKED (proving the owner's credential really was never entered) and the decoy (user 10, org.aegisos.usertype.full.DECOY) RUNNING_UNLOCKED. adb shell sm list-volumes showed both private and emulated;10 as mounted.
adb logcat showed the full volume lifecycle for emulated;10 complete cleanly — UNMOUNTED → CHECKING → MOUNTED, StorageSessionController creating and starting its session, the mount broadcast delivered to UserHandle{10} — with zero occurrences of ExternalStorageServiceException anywhere in the buffer. adb shell pidof com.android.settings confirmed the Settings process alive and not killed.
adb shell dumpsys dropbox --print showed no new crash entry since the boot the trigger was run against (only benign system_server_strictmode/system_app_strictmode noise, and the SYSTEM_BOOT entry itself).
One warning-level stack trace appeared in logcat at the exact point storage initializes (UserManager: The System (uid 1000) is fetching a copy of UserProperties on behalf of callingUid 0..., from StorageSessionController.getConnectionUserIdForVolume via UserManager.isMediaSharedWithParent) and was investigated rather than assumed benign: it is AOSP's own pre-existing self-instrumentation warning about a clearCallingIdentity best practice, logged unconditionally at every call regardless of outcome, not a caught or swallowed exception — unrelated to the three-patch fix and present on ordinary (non-decoy) boots too.
3Consequences
The storage-unmount incident (docs/devices/lynx-duress-decoy-storage-unmount-incident.md) is now fully closed, not just build-verified. This is the fourth confirmed instance in this project of the same underlying pattern (a system service assuming user 0 always unlocks first) — keep watching for a fifth. Per ADR-083's own explicit instruction, this closure rests on the combination of volume state, logcat, dropbox, and process-liveness checks together, not on a visual "it didn't crash" report alone.
4Review when
If a future AOSP/GrapheneOS source sync touches StorageManagerService.getVolumeList() or StorageSessionController.java between onUnlockUser() and initExternalStorageServiceComponent(), re-run this exact trigger and re-check all four signals before trusting a merge kept the fix intact.
Read from docs/governance/decision-log.md, entry ADR-084.