Foreground VPN polling and callback lifecycle validation

2026-09-17, refactor/ui-situation, based on 54bd755e plus the working-tree polling/callback changes. This is device acceptance evidence, not a CI result.

Build and device

• Pixel 8 Pro, Android SDK 37, built-in Native backend and LSPosed active.
• Installed minified debug APK: 1.2.5-253-g54bd755e-dirty.
• APK SHA-256: ff6232626dd562c197cbb3378b97b9637164cc543787e14eee752e365bbc19d2. Installed base.apk matched the local build hash.
• Rebooted after install; the hook report showed the matching dirty version and current boot ID, with both callback overloads attached and zero attach errors.
• 800 JVM tests passed, together with warnings-as-errors compilation, detekt, CPD, Android lint, ktlint, shellcheck and debug APK assembly.

App refresh

Kept the Dashboard foreground and controlled Floppa VPN through its ADB receiver. Agent getState(refresh=false) and a continuous logcat capture were used; no Retry or forced agent refresh was used to cause the transitions.

• VPN off -> on: protection changed from VPN_OFF to ROUTED automatically.
• VPN on -> off: protection returned to VPN_OFF automatically.
• Stable polls did not repeatedly start full gate/root refreshes.
• HOME stopped polling after the current worker finished. No new poll appeared between 02:09:52 and foreground return at 02:10:33. VPN was enabled while the app was backgrounded; foreground return refreshed it to ROUTED without Retry.
• Final screenshot showed "VPN hidden", Native/Java OK, three modules active, and zero issues.
• 39 observed lightweight samples: 106–627 ms, median 344 ms. These are durations on this device, not energy measurements or guarantees for other ROMs. The four-second delay is between completed polling iterations, not fixed-rate work. (The interval was subsequently set to one second at the maintainer's request; that changes how often a sample runs, not the per-sample cost measured above.)

Callback captures

Used scripts/network-view-probe.py under VPN Hide's target UID. App-process diagnostics separately passed the pushed-capabilities/link-properties check.

• Unchanged Wi-Fi, VPN on and off within a 20-second capture: default registration received only its initial Available/Capabilities/LinkProperties/Blocked bundle. No repeated lifecycle or property events; VPN-specific listener stayed silent.
• Two concurrent probe processes under the same target UID each retained an independent default registration. Both recorded Wi-Fi 106 -> cellular 101 -> Wi-Fi 107, with initial properties for every replacement and no VPN handle.
• Airplane mode plus Wi-Fi disabled produced one Lost(107); restoring airplane mode off and Wi-Fi on produced Available(108) and its initial properties.
• Dynamic captures explicitly reported topology changes during collection. Their synchronous before/after snapshots are not treated as stable invariant proofs.
• The final stable VPN-on capture had no errors or violated invariants: active handle/list/properties agreed, no phantom networks or VPN transports, legacy VPN type inactive, VPN listener silent. PendingIntent was not captured.
• No callback-adapter failure or fatal exception appeared in the captured logs.

Temporary offline does not necessarily mean the VPN tunnel is down: Floppa kept its VPN active during the airplane test. The offline assertion concerned loss and recovery of the app-visible cover network, not a forced VPN_OFF gate.

Boundaries and restoration

This run covers one Android 17 ROM and the built-in backend. It does not establish behavior on Android 9–12/non-Bundle dispatchers, OEM callback variants, other root managers, or PendingIntent registrations. Two same-UID probe processes are not a test of every possible within-process registration pattern.

Airplane mode was restored to off, Wi-Fi and mobile data to on, VPN left connected. The logcat recording was stopped. Local evidence files are in /tmp with prefix vpnhide- (state JSON, callback captures, dashboard screenshot and device log); they are temporary artifacts and are not committed user data.