Put it this way, I've never ended a journey and been annoyed I had more miles left in the tank than it promised. But you can bet the opposite would have be stranded and screaming on a highway somewhere instead. Why do you want your customers to retain that angst against your brand? Cars of all things are long term, emotional investments. Screwing one customer and they don't stay quiet.
Either way, we're car beta testers now.
>Just lie to me and tell me the worst case scenario.
The straightforward worst estimate is to assume you'll immediately climb the highest grade road on Earth, against the strongest headwind on Earth, in the coldest conditions and worst precipitation ever recorded on Earth, until you run out of battery. That's probably not the calculation you actually want, however.
Ultimately the question is akin to tuning sensitivity vs. specificity in a medical test. There's a simple answer that can eliminate one class of error (set the tuning threshold to one extreme), but actually that's not the correct answer most of the time.
With the tuning set to the 99th percentile, the computer will underestimate range 99 times for every one instance where it overestimates range. That's not the absolute worst, but is sufficiently worst for you? Or do you want it even more pessimistic?
IMO the answer is cars should simply decline to provide any range estimate at all (just a battery percentage) until the user inputs a destination. None of this (fundamentally intractable) "guess if they're gonna drive up that nearby mountain or not" problem...
Acting like you're entitled to some (vaguely defined) perfect solution isn't a substantive technical discussion. If you don't know what you're even asking for (conjuring nebulous 'clever people' to fill in your blanks), then you don't know if it's even possible or not!
There are lots of "good enough" range estimate solutions (long used on older cars), but there's no perfect solution that will never over-estimate the range ever ever ever (not ever). If you 'solve' this by making the range estimate wildly pessimistic, the driver will simply learn to mistrust-and-adjust, and then you're right back where you started.
Gasoline vechicals do this all the time with their range calculations. They typically exclude the last gallon from the range calculation so that the car won't stop immediately at 0 range.
I feel like if I crank the heat or start driving 80 MPH, it would be surprising if the estimated range didn't budge.
Some might argue: Well, nobody cares how much fuel is in their car or how many amp-hours are left, they care about range. Sure, but these systems are not providing range. They're providing wild-ass guesses based on unreliable inputs and incomplete data. Better to not provide the value at all than to give what's essentially a random guess.
All this to say: I appreciate the mileage estimate in my ICE, but I've had to learn how my PHEV works to be able to interpret it fruitfully.
I've found that the range estimate in my Ford CMAX (PHEV) is highly sensitive to recent driving conditions. For example, I drive one-pedal and am more conscientious about regeneration than my wife. When I drive the car after having been the last driver, the estimate is perhaps 20% greater than if I drive after she drove it.
The result is that the range can drop by a lot even with relatively short trips. In some ways this is bad — it would be better if the vehicle showed accurate range estimates from the time you turn it on. This could be optimized by taking temperature into account more aggressively (I imagine Ford already accounts for it to some extent). But in other ways, there's no way for a vehicle to know if you're going to be heading uphill or downhill on your next trip, or if you're taking the freeway or city streets. It sucks when range estimate swing wildly, but it's possible that this is better than the alternative — that it remains too rosy for too long, and then you're screwed.