We must first assume that James's #s are taken after a 10minute+ rest period after completion of charging to remove the reported inaccuracies in the process ... if so, then
There's another theory I'd like to propose.
I have heard that on older (possibly much older) firmware the IR would actually increase during long downhill (regen) stretches, but this seems to no longer be the case (at least according to some reports). (The estimated range does, of course, drastically increase on those same downhills.) In other words, the IR gets "stuck" at its last value, then does not decrease for a long time (as you continue moving), but also does not increase when you expect it should.
So ... James's experience could fit that theory. Near the end of the efficient trip the IR should go up, but does not, but at least the battery has a little extra charge as compared to the estimated IR reported. Then when the charging and re-calculation takes place, it ends up a little higher due to the more efficient (but badly IR represented) final trip segment (the real IR is higher than shown at the start of charging and the extra charge adds a fixed amount of IR). On the uphill stretch either a] opposite happens; or b] charging ends with a more realistic IR that is more representative of the state-of-charge.