- Decision
- ADR-090
- Date
- 2026-09-17
- Status
- FindingFinding 13 sub-scenario: inconclusive — the hardware trial was blocked, do not treat as closed. The blocker itself is fully root-caused and is not security-relevant. Dormant findings 3/4: confirmed, that part of ADR-087's item 2 is closed.
- Mentions
- ADR-087
- Mentioned by
- ADR-092
1Decision
Attempt to hardware-verify Finding 13's re-lock-after-earlier-unlock sub-scenario on the owner's real lynx; confirm the two resource-value dormant pre-unlock findings (3, 4) against the owner's real flashed build rather than only the stock reference APK.
2Why and evidence
Real device trial attempted per the 2026-09-17 hw-verify design's Component 2 Step 2, but Communal Hub proved entirely unreachable on this device regardless of the sub-scenario tested: swiping the lock screen did nothing, and Settings$CommunalSettingsActivity's own CommunalDashboardFragment logged "NO dashboard tiles for CommunalFragment" and self-closed within the same second (confirmed via live adb logcat captured around the actual attempt, not assumed).
Root-caused rather than accepted at face value: every Hub-specific capability signal on this build is actually enabled — config_glanceableHubEnabled=true, config_show_glanceable_hub_toggle_setting_mobile=true, config_swipeToOpenGlanceableHub=true (confirmed via aapt2 dump resources against the four real GlanceableHub*Overlay.apk packages pulled off the device, dumped using the build server's own prebuilts/sdk/tools/linux/bin/aapt2 since no aapt2/Android SDK exists on this workstation), and every communal_*/glanceable_hub_* flag in device_config systemui reads true.
No AegisOS patch in aegisos-platform or aegisos-services modifies any Communal/Glanceable Hub code path — the only two textual hits for "communal"/"glanceable" across both repos are benign: an unrelated stock-AOSP context line (int getCommunalProfileId();) inside frameworks-base patch 0019, and a prose reference to GlanceableHubSysuiConfigOverlay.apk in SOURCE.md's own ADR-087 overlay analysis — neither is an AegisOS change to Hub itself, so this is not an AegisOS suppression.
The actual blocker: Settings.Secure.screensaver_components is null (nothing has ever been registered as a selectable dream) and dumpsys dreams showed the live dream as com.android.systemui.lowlightclock.LowLightClockDreamService, never a Hub dream implementation, even after temporarily setting screensaver_enabled to 1 to test (reverted to its original 0 afterward, leaving the device as found).
The one Settings screen meant to let a user select Hub as the active dream exposes zero options — a stock GrapheneOS/AOSP platform gap on this lynx CP2A.260705.006 build, not a security-relevant condition and not caused by anything this project has ever patched. The owner reviewed this root cause and declined further investigation into stock Settings/Dream wiring as out of scope for a duress-unlock security review.
Separately, Findings 3 and 4 — the two dormant findings whose classification rests on a device resource value rather than a source-code-absence fact — were confirmed directly against the real flashed build: config_enableBouncerUserSwitcher=false (pulled /system_ext/priv-app/SystemUI/SystemUI.apk) and config_keyguardUserSwitcher=false (pulled framework-res.apk), both matching the audit doc's stock-reference findings exactly; fw.power_user_switcher (Finding 8, already COVERED regardless) reads empty/unset, matching its documented default.
3Consequences
Finding 13's re-lock-after-earlier-unlock sub-scenario remains formally UNTESTED, not confirmed safe or unsafe — treat it exactly as docs/devices/lynx-duress-decoy-prelock-audit.md's existing Live-state caveat already does. This is a materially weaker claim than "hardware-confirmed no leak" and must never be conflated with one. Findings 3 and 4 are now confirmed on the real flashed build, not only the stock reference.
4Review when
Re-attempt Finding 13's physical trial only if a future AegisOS or GrapheneOS/AOSP release makes Communal Hub genuinely selectable as a dream on this device (screensaver_components becomes non-null with a Hub-family component, or CommunalDashboardFragment starts rendering real tiles) — until then this sub-scenario cannot be dynamically exercised on this hardware and stays open by default, not by neglect.
Unconfirmed source-level hypothesis worth starting from then, never dynamically tested: the widget's own filter (packages/apps/Multiuser patch 0004) keys off isUserSwitcherEnabled() → isDeviceCurrentlyLocked() → live WindowManagerInternal#isKeyguardShowing() state, not one-time-per-boot CE-storage-unlock history — a different mechanism than the one this caveat is actually about, which would make the re-lock sub-scenario plausibly already covered; reasoned from source during this session's plan-writing but never verified.
Read from docs/governance/decision-log.md, entry ADR-090.