Procne wrote:No, once you have (5,5,5,85) seeds then the only way to increase its purity is to increase last component's purity, simply because it alone has so high impact on it. I'm pretty sure that if we took those values and ran simulations with the system you described we would get lowered purity in majority of cases.
initial values:
[5,5,5,85]
final values:
[5+a,5+b,5+c,85+d] where a,b,c,d are generated from independent uniform distributions between -5 and 5.
condition to increase the lead component:
100*(85+d)/(100+a+b+c+d) > 85
rearrange
0.15*d-0.85*(a+b+c) > 0
since d, a, b, c are as likely to be positive as negative, this happens on average half the time (a symmetry argument, if you want to prove this more rigorously i can as well)
I think if we simulate the system I have said it will confirm this, which means that if you see something in game that is different (and I'm pretty sure you are a more experienced farmer than me, so I'll trust you on that), this probably means that me thinking it works this way (+/-5 then renormalize) is wrong.
You are right that to increase purity at high purities you essentially have to increase the high component, but that is only true after normalization, and not for the rolls themselves - for example with the values we are using a roll of [-5,-5,-5,-5] will result in 100% pure output.
Has anyone got data to determine what the distribution of the rolls actually is? I admit I haven't done anything but look at what loftar wrote and think that this is what happened, so I'd love some hard data.
