One thing to remember when thinking about % is that most people equate numbers like 20% to equal 1 out of 5.
In computers, that is not true. It would be nice if it was, but, typically developers program RNGs (Random Number Generators) based on a principal of a non-repeatable non-linear numeric result.
This means that there is no one set of circumstances where you can produce a predictable numeric result from the number.
Where in Mathematics, probability is done through Matrices, you don't typically want to, or need to do that much work in a video game RNG. If a probability Matrix was used, then 20% would be 1 out of 5 with a certain margin of error.
Instead, the RNG *should* generate a number between 1-100 randomly 1% of the time. With each number being likely or possible to come up.
However, anyone who has ever worked on an RNG should know, there are a series of traps that occur, the most common when learning tend to be:
1) Numbers form a notable incrementing amount.
2) A single number will never actually repeat
3) Numbers tend to spike from high to low values, avoiding mid ranges
4) Numbers tend to cluster around a value for large sampling
As a result, many computer based (and SE has shown this very much through the history of their programs) to have a VERY random generator that does not produce numbers that fall within any viable probability chart.
With a standard thought on probability, you would have a probability chart like this for something that has a 20% chance of success
Attempt Chance of Success based on Success Based on Fails
1 20% 59% 80%
2 20% 18% 97%
3 20% 2.7% 99.8%
4 20% 0.1% 100%
Where with SE's algorithms, what we actually see is more like:
Attempt Chance of Success based on Success Based on Fails
1 20% 20% 20%
2 20% 20% 20%
3 20% 20% 20%
4 20% 20% 20%
*all numbers a crude, and while, *should be* functionally accurate, may not stand up to a math test (for any math geeks who are looking to prove me wrong) and are based off of a quick run through the Binomial Probability formula (P(x) = nCx * p^x * q^(n-x))...