For blindfold solving a lucky case is defined by:
...
3) more than 3 corners are correctly positioned, or
4) more than 4 edges are correctly positioned.
@Stachu, the WCA regs have NOTHING about whether a case is "lucky" or not.
EDIT: And also Stachu, it can change. There are more scenarios of "lucky" for 3OP (orientation luckiness and permutation luckiness) where methods like M2/Old Pochmann only have permutation luckiness.
I don't get credit for that - lots of people suggested 20% before me. I just jumped on the bandwagon. I guess I made it a lot more visible because I talk about it a lot more than other people.
But yeah, I think 20% is a good rule, and it even works pretty well for even-sized cubes (where many centers can be solved by properly orienting).