Sorry, I didn't realize anybody had responded when I deleted my post. > The reason for this is that the range estimator can't predict things the driver might d

Sorry, I didn't realize anybody had responded when I deleted my post. > The reason for this is that the range estimator can't predict things the driver might do to spoil the forecast Notice that all of the lines are extremely linear, meaning that Broder's behavior (or a miscalibration) caused a consistent bias in the estimate -- for 500 miles! It's very possible to design an estimator that can take this into consideration and make more accurate predictions. > All estimations in this class must make assumptions about future behavior. Changes in behavior change the estimate. Obviously. So what assumptions is it making? "My estimate has been consistently off for the last 250 miles, but it's going to be correct in the next 250 miles.... I can just feel it!" > Changes in behavior change the estimate. They should, but apparently they don't in this case.