Re: Cockup?
"See Kalman filter on Wikipedia"
We're off in the world of conjecture here as I don't think we really know how Tesla mileage metering works. But I should point out that Kalman filters although much loved by mathematicians, don't necessarily work all that well in practice. Basically Kalman filters build a model of whatever is being analyzed using statistical techniques. They then reject "noise" -- which is to say data points that seem improbable based on the model. That can work brilliantly. If it didn't, Kalman filtering wouldn't be used much. But they can have a problem if the data rejection isn't set up properly for the particular problem. In simple terms, the problem is described as "tracking the model". What happens is that the filter works well unless the conditions change. Sometimes, if conditions change too quickly, the filter can fail to adjust it's criteria fast enough. It will then reject the new data as being too unlikely to believe.
Example: An aircraft flies in a straight line for a long time at constant speed. A statistical filter smooths radar data observing the aircraft and, once everything settles down, it comes up with much better estimates of position than the single radar data points would give. However, if the aircraft abruptly changes direction and or speed the model may not "open up" sufficiently quickly to accept the new data. The model will then wander off projecting the aircraft to be somewhere it isn't.
I scanned the Wikipedia article and didn't see that addressed. My guess would be that it is addressed, but not necessarily in terms that are easily understood.
Is it possible that a Muskmobile could have a faulty shiny new fangled digital technological odometer? Seems credible to me