"3 minutes away" on a transit app isn't a measurement. It's a forecast, built on a GPS ping from a few seconds ago and an assumption that the bus will keep moving at roughly the speed it's moving now. Most of the time that's close enough. Right before a red light, a stalled intersection, or a stop with six people boarding, it isn't — and the app has no way to know that yet.
This is a smaller version of the same problem every forecasting model has: the prediction is only as good as the assumption that the recent past keeps holding for the next few minutes. Weather forecasting deals with this over days. Transit apps deal with it over minutes, which is exactly why a 3-minute estimate can swing to "arrived" or "missed it" so fast — there's almost no time buffer for the model to be wrong before you find out it was.
None of this means the apps are badly built. It means the next time one is "wrong," it's worth remembering it was never claiming certainty in the first place — it just didn't say "probably" out loud.