The code is a straight 50/50, with no odd tilting in anyones favor.  With small sample sizes, you get grossly inaccurate numbers.

I purposely made the code simple, and straight forward so that I can share the code with trusted users who are programmers to look at if it ever comes under scrutiny.  The math is stateless, when it figures out what person A rolls and what person B rolls, it is stateless, it does not know or know care how up or down you are, how much you have or anything other than what its rolling currently.  It does not auto correct or get nudged in any direction.

Adding in any code to manipulate the wins/losses to ANYONES favor is an abuseable system and then I can't look anyone in the eye and say 'There is no skewing of numbers what so ever, its is 100% legit 50/50 code'

Looking at a data sample as small as yours, math will never be very accurate.  For example, with a data sample of just 2 that means that if you flip a coin the first time it should be heads, the second time tails.  If it is not tails then something must be broken.  Lol, of course not, the more you flip it the more accurate the random 50/50 is. Where random accuracy comes into play is with very large data samples.

For example, here is the most recent data from House without me having to dig into backups, it's a little over 3 million of the last bets.

House Wins 1,508,060 which is 49.9861% win ratio
House Loses 1,508,898 which is 50.0139% win ratio
House Ties 21,608 (Taken out of the equation so that both wins and losses total to 100%)

Taking into account the last 3,016,958 that ended with a win or a loss, house is actually down by 419.  But in a 3 million data sample, thats pretty damn close to 50/50!