I'd bet the farm on 20 samples - and under some circumstances, on as few as 6 samples.
It's a trick played by statisticians to prove all sorts of stuff and to supplement their meager incomes.
You're a braver man than me, then.
If I could make accurate predictions with small sample sizes, I'd stand to save alot of money.
If the universe is small and tesing 20 means testing a significant portion of the whole, that's one thing.
Extrapolating from a small sample to a large population is another.