In terms of your problem, do you should probably do a uniform random that returns values between 0 and 100 and use the appropriate cut-off point. This is handy for more resolution (doing 1-5 is too "blocky").
As for the arc, you will need to tell us the behaviour of not only how things change with the difficulty of the action, but more importantly how things change with the nature of what the current score is.
I would recommend something like a modified arc-tangent function:
Inverse trigonometric functions - Wikipedia, the free encyclopedia