If I use the combined method it is possible to get 100 points either way.... as an example, if the real figure is 1,000,000 units, to get 100 points when you under-predict, your prediction would have to be 414,214 (though I m only allowing multiples of 5, so to get over 100 points you would have to predict 410k)
To get 100 points when you over-predict your prediction would need to be 2,414.214 (or 2,415k to go over)
Or in other words it is now proportional to whatever the real figure turns out to be.... if you predict double you get the same points as someone who predicts half, which is in fact 75 points (I actually thought that was what method 2 did, but I clearly got that wrong)
So it gives a little extra lee-way to those who over-predict, but starts to penalise those who underpredict more
