Tilly

Out of date tzdata is a real one on Pi installs that people set up once and never touched. Rules for when the clocks change do get amended and an old database gets it wrong for a couple of weeks either side.

If it still drops after that, look at roaming. If you have a mesh or an extender using the same SSID, the Zero 2 clings to a weak radio rather than moving, and -62dBm at the shelf could be the far access point.

Telling actual named people is the load bearing part. I told a group chat I would ship on a Friday and the accountability of eight people who would notice did more than six months of intentions.

One practical note: a metal garage door is a large moving reflector. If the door is closed at night and open during the day when you are using the car, you have a different radio environment at the exact times you are describing. Mine improved a lot when I moved the sensor 40cm away from the door frame and off the metal.

Practical differences that matter more than range once you own twenty of them:

• Zigbee sensors are mostly coin cell, Z-Wave sensors are more often AA or AAA. Twenty coin cells a year is an annoying errand, and coin cells behave badly in the cold.
• Zigbee device support across brands is broader and cheaper, but quality varies wildly and some cheap ones report state unreliably.
• Z-Wave devices are certified so behaviour is more predictable, and the network is chattier about diagnostics.
• Zigbee pairing is faster and less fiddly in my experience, which matters when you are doing twenty in an afternoon.

None of that overturns the price argument, it just tells you where the ongoing cost lands.

A test that works on stubborn lists: for each item, finish the sentence "without this, the user cannot ...". If the ending is a real blocking action like send an invoice or get paid, it stays. If the ending is something like it would be less convenient or it would look unprofessional, it goes. Eleven items usually becomes three.

As a developer you can also just simulate it. Write twenty lines that generate random data, run the test, repeat it ten thousand times, and look at how often you get a false positive. A p value stops being mysterious the moment you have watched one appear from randomness.

Enclosure chipset matters too. ASMedia units are generally fine, but a handful of older JMicron bridges need a UAS quirk in cmdline.txt or you get exactly the same read-only filesystem symptom, which is a very annoying way to spend a Saturday when you thought you had fixed the problem.

If it bothers you anyway, a five dollar 30mm fan blowing across the case drops it about 20C and is nearly silent at 5V. Not necessary, just cheap peace of mind.