Quote:
Originally Posted by Phil153
Like Mason, I don't think it's significant, but the only way to answer this question definitively is to run a simulation with known folded cards and compare to the unknown card calculations. I don't have the time to do this right now but I do have the basic code for it and it won't take terribly long, so I'll do it in the next 3-4 weeks if no one else takes it on. I'll release the results and code open source so that others can look at it.
From what I recall is that Mason was speaking of the effect on strategy within a hand. This isn't so much a problem of math as much as that all the human factors in a hand, estimating ranges based on opponent behavior and so on, present a margin of error that will dwarf any difference between a pot equity calculation based on the assumption that the stub is identical in randomness to all unseen cards and one that takes opponent behavior into account.
Who cares if, assuming your ranges are correct, your equity is 50% rather than 49%, as your estimation of opponent behavior is generally not going to be so precise. This is a totally different kettle of fish than using the same calculations to make detailed criticisms of an RNG over a massive amount of trials. The larger source of error that makes the smaller error insignificant is gone.
This also seems to doom any totally precise calculation of expected ev when taking any all in hands in a game with more than two players. To do this one has to make assumptions about what an opponent will or will not fold, and there simply is no way to do this by a universal method. It is still an estimation with flaws as people are different.
This would be all fine and dandy if we assume that every player gets in the same situation the same amount of times. However, the whole point of playing poker is to not let this happen to you.
I imagine some sort of feature that tries to model typical behavior, where the end user can put folding ranges for typical opponents in typical situations will make these sorts of calculations more accurate, but it still is going to suffer from the same problem once the number of trials gets large.