That's a pretty good idea, Mark. I've been pondering the solution for this and it might take a bit of lateral thought. The problem to the easy solution is that you wouldn't know in advance where the physical location of the split would be so in terms of GPS position you wouldn't know how close you were to it (unless you had a stab at predicting its location and pre-loading that information). However as you say, there is the chance of adding up the GPS track distance from the last fix (ie split) which would be able to trigger an alarm as it came up to a whole Km or mile. We know the controversy around GPS path measurements but for this application the required accuracy need not be that great. I'll have a look!
Nearly there on the reverse measurement solution - the algorithms have to run in the forward and reverse directions simultaneously and consistently as certain elements (splits etc) are directional as are the tiny elements that ensure positive rounding in the race direction.