Actually wait, now I've confused myself.
Someone good with maths check this logic:
We know(?) based on the Bernoulli process (
) that we eventually converge around a 63% chance of any random event occurring given an equal number of chances, the more chances and lower probability the closer you get to 63%, eg;
Probability of NOT guessing a random number between:
1-10 with 10 guesses: 34.868%
(0.9^10)
1-100 with 100 guesses: 36.603%
(0.99^100)
1-1mil with 1mil guesses: 36.787%
(etc...)
So that gives us a rough idea that the probability of a random fair event occurring given the equal number of chances required is around 63%
Given that the slot provider has calculated the average payout of the feature as 100x bet and we are working on 100% trtp we can assume average cost to achieve feature is also 100x.
Now it may have been easier to do this with 96% RTP because you can then ascertain that you will have on average 248 chances with a 100x bankroll (see dunovers second paragraph) and so the chance would be 247/248(as a decimal
0.99596774193, not sure why it doesn't work as a fraction)^248 = 36.7% chance of NOT getting the feature and so ~63.3% chance of getting it.. which would seem inline with the initial proposition of 64% on a 100% RTP machine.
I do not believe basegame/freespin RTP split comes into the equation at all.
And it's definitely not 50% because "you either get it or you don't" because then 1 in 2 people would be winning the lotto every night.