We got Parking Sentinel Mode wrong on Android

The assumption we made in the office

When we built Sentinel mode for Android, we based it on how iPhone handles background task management. On iOS, the way the OS manages power while your phone stays mounted is fairly predictable. You tell the system 'I want to record,' and iOS gives you a certain window before it throttles you. We thought Android would work the same way.

It doesn't. Not really.

Android's background recording behavior varies wildly depending on manufacturer. A Samsung handles it one way, a Pixel another, a OnePlus something else entirely. We'd tested on three devices in the office. We shipped with confidence. That confidence lasted exactly 21 days.

The messages started arriving. People complained that Sentinel would wake the screen every 90 seconds, or go silent after two hours, or refuse to start at all if Adaptive Battery was enabled. One user had set up a Hawk on an older Motorola for parking protection while visiting family, and it stopped recording after 43 minutes.

What we thought Sentinel should be

We'd positioned Parking Sentinel as 'continuous recording with minimal power draw.' The idea was sound: if your car is hit while parked, Hawk records it. No external hardware. No subscription required. Just your phone on the dashboard, doing what it's designed to do.

But our implementation treated Android like it was iOS. We assumed the OS would cooperate evenly across devices. We built for a fantasy version of Android that doesn't exist outside of Nexus labs circa 2012.

What we didn't account for: doze mode, manufacturer battery optimization, RAM management differences, and the fact that some phones will kill background tasks if they detect you're recording video while the screen is off. The actual behaviour of the system was far messier than our testing had revealed.

What users actually needed

The real insight came from reading support emails carefully. Users weren't complaining about the idea of Sentinel mode. They were complaining that it behaved unpredictably. On one phone it worked beautifully. On another, it was unreliable. That inconsistency eroded trust faster than a single failure would have.

One Bolt driver told us he'd switched to a dedicated dashcam for parking protection because Hawk was 'too smart.' He meant it as criticism. But he was describing the actual problem: we'd over-engineered the power management, trying to be clever about CPU scaling and frame-rate adjustment, when what people needed was straightforward behaviour they could depend on.

Marcus, the user who started this, eventually sent a follow-up. He'd figured out that disabling Adaptive Battery entirely let Sentinel run for six hours on his Pixel. That wasn't a feature. That was a workaround. He shouldn't have needed to find it.

How we rebuilt it

The fix wasn't elegant. We moved away from background service orchestration and toward simpler, more predictable power states. Instead of trying to optimize record-and-sleep cycles, we let Sentinel run at a steady, lower frame rate. Instead of hoping Android's background task scheduler would behave, we added explicit warnings if the user has power-saving modes enabled, and clear documentation about device compatibility.

We tested again. Not in three devices in the office. This time, we sent test versions to ten users across different manufacturers. We asked them to leave phones in cars for eight hours, parked in direct sun, and report back. The results weren't uniform, but they were predictable. Users knew what they were getting.

We also stopped calling it 'Parking Sentinel mode' on Android. We renamed it 'Parked Recording,' which sounds less magical and more honest about what it actually does.

The thing we're still figuring out

This whole episode taught us something uncomfortable: the gap between what we thought Hawk did and what it actually does was partly our fault for not listening to edge cases early enough.

Android is fragmented. That's not a secret. But we'd treated it as a weakness in the platform rather than a reality we needed to design around. We wanted to ship a universal solution. Android doesn't ship universal solutions. It ships something that works differently on different hardware, and the people using it expect that, and plan for it.

Now, when we ship a feature on Android, we document the constraints. We test on devices people actually own. And we listen to the first ten users who tell us something isn't working the way we described it.

Hawk's continuous recording with SHA-256 integrity hashes on every clip still works beautifully as a dashcam. The evidence locker, the dispute export, the GPS overlay, all of that survives the journey from dashboard mount to insurance claim or police report because those features run in the foreground, where the operating system treats them predictably.

But Parking Sentinel? That taught us to be honest about what the hardware beneath us actually does, rather than what we wish it would do.