I'm pretty sure it doesn't.
I'll design or steal a new defensive play, run it once and that first time it will have 0% success against all 3 of my opponent's most likely choices, as you might expect as it's never been tried before. Thing is, Regardless of success, the next time I call it, it will usually have 80% chance against all 3 of my opponent's most likely choices, even though we know it's never been tested against at least 2 of them. After that, results are completely up in the air, using no math formula that makes any sense at all.