1Decision
Fix the audit's three Critical items and four High items: the legacy power-menu lock-screen name leak, Settings' Users management screen, the storage move/migrate credential dialogs, and the four getUserCount()-based Wi-Fi/uninstall/factory-reset sibling call sites.
2Why and evidence
LegacyGlobalActions.addUsersToMenu() (frameworks/base) called raw getUsers() and rendered every real user's name with a checkmark directly on the lock screen; fixed with a one-line substitution to getUsersForUiDisplay(true) (patch 0027). UserSettings.updateUserList() (packages/apps/Settings) called raw getAliveUsers() — confirmed via source to be exactly getUsers(excludeDying=true), i.e.
getUsersForUiDisplay(true)'s own non-decoy-branch semantics — fixed by the same substitution (patch 0004).
The storage move/migrate dialogs (StorageWizardMigrateConfirm, StorageWizardMoveConfirm) needed more care than a substitution: both genuinely must iterate every real user's raw getUsers() to unlock their CE storage before migration can proceed, so that enumeration was left untouched; only the credential dialog's templated description text (which named the locked user directly) was made decoy-aware, via a new generic string (storage_wizard_move_unlock_generic) shown instead whenever the current session is the decoy (patch 0005).
The four sibling getUserCount() call sites — three uninstall-dialog single/multi-user text sites (frameworks/base/packages/PackageInstaller, patch 0028) and three Wi-Fi/factory-reset sites (WifiEntryPreference's "shared" badge, WifiDetailPreferenceController2's "forget shared network" dialog, MainClear's "other users present" warning, all packages/apps/Settings, patch 0006) — were all fixed with getUsersForUiDisplay(true).size(), confirmed against getUserCount()'s own real implementation (getUsers().size()) first.
Every fix was authored under the established local-first worktree discipline (each touched file's sha256sum confirmed byte-identical to the live remote before editing, landed via git am only after confirming no build was in flight, patches regenerated fresh from real post-git am branch history) and independently build-verified (m services, m PackageInstaller, m Settings ×3, all clean, no diagnostics on any edited file).
3Consequences
Patches exported to aegisos-platform/credential/duress-unlock-decoy/patches/{frameworks-base,packages-apps-settings}/, SOURCE.md/README.md updated.
All seven fixes are compile-time verified only. An emulator dynamic-validation attempt for the earlier getUsersForUiDisplay fix (ADR-066) hit a real tooling limitation, not a fix-correctness problem: synthetic adb taps did not register reliably against this build's Compose-based QS footer UI (confirmed via uiautomator dumps showing correct element bounds/ids but no resulting dialog, and a control tap on an unrelated button — the settings gear — also failing to navigate), so the specific dialog the original leak was found in could not be re-opened for a clean before/after comparison on this attempt.
One partial, inconclusive positive signal was obtained (the QS avatar's own accessibility label, content-desc="Signed in as <decoy name>", correctly showed only the decoy) but this label is understood to be sourced from a different, simpler "who is currently selected" code path than the leaking enumeration list itself, so it does not by itself confirm the fix.
4Review when
Following this project's own established precedent (ADR-060 through ADR-063: batch fixes, then one consolidated hardware-validation pass) rather than fighting the emulator's UI-automation limitation fix-by-fix, dynamic re-validation of this whole batch (plus ADR-066) is deferred to a single consolidated round — ideally on physical hardware, where interactive adb taps have already been proven reliable earlier in this same session — once the remaining audit items (8, 12, 13, and the three low-confidence items) have also landed or been explicitly resolved as not needing a fix.
Read from docs/governance/decision-log.md, entry ADR-068.