Yeah it’s not messaged in the app anywhere. I’ve actually built a mini tutorials view into the app, for mini articles on things like this, to help get the best out of the data and recording. But it’s been stalled on having time to write the articles. Ironically, I instead end up spending more time explaining the same things repeatedly on the forum instead I really should finish up the tutorials/mini articles stuff.
There’s a couple of things going on with very long visits (eg lockdowns or movement restrictions that lead us to end up staying at home for multiple days). The first is within Arc, and the second is within iOS itself.
Within Arc, there’s a system of predicting the probability of you leaving the current place, based on time of day, duration of visit, and some other factors. Longer visits make those predictions less reliable and less useful. The result of that is that Arc becomes less accurate at quickly restarting recording when you do leave the house.
Though that inaccuracy should typically be less than 2 minutes, although it can be worse if Arc goes into “deep sleep”, where it effectively turns off completely, based on the prediction that you’re definitely not going to leave the house for 6+ hours. (During deep sleep state, Arc has several wakeup call systems, to get it restarted before your predicted leaving time. But it’s a gamble, because you could leave before the predicted time, and Arc will then miss maybe 5+ minutes of recording, until less accurate wakeup systems kick in).
But that doesn’t explain the weird, excessive location data drift during these long stays at home. That’s a quirk of iOS itself. I haven’t done extensive testing on it in a while, but the effect is that for some reason iOS / the phone’s location data sometimes gets significantly worse during long stays. I can’t guess what underlying system is causing that - it doesn’t seem a reasonable energy use optimisation - but enough people have seen it that it’s a definite pattern
Aside: iOS also has an underlying location accuracy improvement system, based on wifi hotspot triangulation, that improves the accuracy of location data at new places over the first few days. But that shows itself as typically low quality location data in the first 1-2 days at a new place, which then becomes increasingly accurate and refined after that. So that system doesn’t appear to be related to this problem, as what we’ve seen during lockdowns is effectively the opposite: the location data gets worse over time, rather than improving over time.