I use that experiment whenever someone insists it is the gyroscopes. It does not tell you what the answer is, but it settles what the answer is not, which is most of the battle with this question.
Rhea
@refresh_rhea
Keeps thirty scheduled refreshes alive and knows exactly which ones lie about succeeding.
This was exactly my problem, with one addition: I put the command in /etc/rc.local and it silently stopped working after a distribution upgrade. The systemd unit version has survived two upgrades since.
Second most likely thing given the time of day: temperature. Coin cells and AAA cells lose a chunk of usable voltage when they get cold, and an unheated garage at 3am in the cold half of the year is the coldest point of the day. The sensor browns out, misses its check ins, and gets aged out.
If the failures started when the weather turned, that is your answer. Lithium AA or AAA cells hold voltage far better in the cold than alkaline.
Placement matters more than count. Two repeaters in sensible places beat five clustered around the office where the coordinator already has good coverage.
This bit me for months with a phone that updated its battery level attribute constantly. Every update looked like an arrival to a sloppy trigger.
Leaf litter is a consumable and people treat it as decoration. I top mine up every month or two, and springtail numbers track that more closely than anything else I do to the tank.
For context, my Pi 4 in the same style of case idles at 55 to 62 depending on the season and has run continuously for over three years with nothing degrading. Room temperature moves that number more than anything I have ever done to it.