A common practice is to use static tests from, say, just one set of captured data -- spins or whatever.
But -- and it's a big but -- any set of fixed data contains its own biases and idiosyncrasies.
And testing with a static, captured database would be fine if you knew you could play the same decisions over and over.
But the real world doesn't offer that opportunity, does it?
The Contrarian View
Would it not be better to do testing only with random numbers generated on-the-spot?
Would that not be the closest thing to a real-life situation where every spin of the wheel or every draw of a card presents a new and not necessarily expected outcome?
A good way is to grab a few spins from here and there and then it becomes your own personal permanence.
It's a bit of a pain but goes some way to achieving what you are suggesting esoito.
You could take spins from...
Spielbank..Dublinbet..Smartlive..Betvoyager..Random.org etc..and the Italian roulette website has a gazillion spins from different casinos. (it's actually 7,527,056) :)
http://www.laroulette.it/risorse/permanenzimetro/folder_listing (http://www.laroulette.it/risorse/permanenzimetro/folder_listing)
cheers