hi,
thanks for enlightening me.
Yes the simulator would be very good if the algorithm
behind the model is based on a large sample of past data
and using a hold out sample for testing.
Otherwise simulating using 'random by rank' as an example
would be akin to being blindfolded and sticking a pin in the form guide
and selecting Notts County playing away to beat Man U.
I may be on the wrong track and please put me right if I am.
I thrive on constructive criticism.
R