[darkmighty]

Well it depends on proper cross-validation, sample size and of course predictability (i.e. 4 * P(Suc.|data) * P(Fail.|data)<<1). But online prediction is of course what will tell the long term reliability of any model.